Scalable hierarchical data-driven navigation system and method for information retrieval

Abstract

A data-driven, hierarchical information navigation system and method enable search of sets of documents or other materials by certain common attributes that characterize the materials. The invention includes several aspects of a data-driven, hierarchical navigation system that employs this navigation mode. The navigation system of the present invention includes features of an interface, a knowledge base and a taxonomy definition process and a classification process for generating the knowledge base, a graph-based navigable data structure and method for generating the data structure, World Wide Web-based applications of the system, and methods of implementing the system. Users are able to search or browse a particular collection of documents by selecting desired values for the attributes. A data-driven, hierarchical information navigation system and method enable this navigation mode by associating terms with the materials, defining a set of hierarchical relationships among the terms, and providing a guided search mechanism based on the relationship between the terms. In another aspect of the invention, implementations of the invention may be scalable through parallel or distributed computation. Implementations of the invention may employ master and slave servers in a hierarchical configuration.

Description

FIELD OF THE INVENTION

[0002] The present invention generally relates to information navigation systems and search engines.

BACKGROUND OF THE INVENTION

[0003] Information retrieval from a database of information is an increasingly challenging problem, particularly on the World Wide Web (WWW), as increased computing power and networking infrastructure allow the aggregation of large amounts of information and widespread access to that information. A goal of the information retrieval process is to allow the identification of materials of interest to users.

[0004] As the number of materials that users may search increases, identifying materials relevant to the search becomes increasingly important, but also increasingly difficult. Challenges posed by the information retrieval process include providing an intuitive, flexible user interface and completely and accurately identifying materials relevant to the user's needs within a reasonable amount of time. Another challenge is to provide an implementation of this user interface that is highly scalable, so that it can readily be applied to the increasing amounts of information and demands to access that information. The information retrieval process comprehends two interrelated technical aspects, namely, information organization and access.

[0005] Current information navigation systems usually follow one of three paradigms. One type of information navigation system employs a database query system. In a typical database query system, a user formulates a structured query by specifying values for fixed data fields, and the system enumerates the documents whose data fields contain those values. PriceSCAN.com uses such an interface, for example. Generally, a database query system presents users with a form-based interface, converts the form input into a query in a formal database language, such as SQL, and then executes the query on a relational database management system. Disadvantages of typical query-based systems include that they allow users to make queries that return no documents and that they offer query modification options that lead only to further restriction of the result set (the documents that correspond to the user's search specifications), rather than to expansion or extension of the result set. In addition, database query systems typically exhibit poor performance for large data sets or heavy access loads; they are often optimized for processing transactions rather than queries.

[0006] A second type of information navigation system is a free-text search engine. In a typical free-text search engine, the user enters an arbitrary text string, often in the form of a Boolean expression, and the system responds by enumerating the documents that contain matching text. Google.com, for example, includes a free-text search engine. Generally a free-text search engine presents users with a search form, often a single line, and processes queries using a precomputed index. Generally this index associates each document with a large portion of the words contained in that document, without substantive consideration of the document's content. Accordingly, the result set is often a voluminous, disorganized list that mixes relevant and irrelevant documents. Although variations have been developed that attempt to determine the objective of the user's query and to provide relevance rankings to the result set or to otherwise narrow or organize the result set, these systems are limited and unreliable in achieving these objectives.

[0007] A third type of information navigation system is a tree-based directory. In a tree based directory, the user generally starts at the root node of the tree and specifies a query by successively selecting refining branches that lead to other nodes in the tree. Shopping.yahoo.com uses a tree-based directory, for example. In a typical implementation, the hard-coded tree is stored in a data structure, and the same or another data structure maps documents to the node or nodes of the tree where they are located. A particular document is typically accessible from only one or, at most, a few, paths through the tree. The collection of navigation states is relatively static—while documents are commonly added to nodes in the directory, the structure of the directory typically remains the same. In a pure tree-based directory, the directory nodes are arranged such that there is a single root node from which all users start, and every other directory node can only be reached via a unique sequence of branches that the user selects from the root node. Such a directory imposes the limitation that the branches of the tree must be navigationally disjoint—even though the way that documents are assigned to the disjoint branches may not be intuitive to users. It is possible to address this rigidity by adding additional links to convert the tree to a directed acyclic graph. Updating the directory structure remains a difficult task, and leaf nodes are especially prone to end up with large numbers of corresponding documents.

[0008] In all of these types of navigation systems, it may be difficult for a user to revise a query effectively after viewing its result set. In a database query system, users can add or remove terms from the query, but it is generally difficult for users to avoid underspecified queries (i.e. too many results) or overspecified queries (i.e. no results). The same problem arises in free-text search engines. In tree-based directories, the only means for users to revise a query is either to narrow it by selecting a branch or to generalize it by backing up to a previous branch.

[0009] Having an effective means of revising queries is useful in part because users often do not know exactly what they are looking for. Even users who do know what they are looking for may not be able to express their search criteria precisely. And the state of the art in search technology cannot guarantee that even a precisely stated query will be interpreted as intended by the user. Indeed, it is unlikely that a perfect means for formation of a query even exists in theory. As a result, it is helpful that the information retrieval process be a dialogue with interactive responses between the user and the information retrieval system. This dialogue model may be more effectively implemented with an effective query revision process.

[0010] Various other systems for information retrieval are also available. For example. U.S. Pat. Nos. 5,715,444 and 5,983,219 to Danish et al., both entitled “Method and System for Executing a Guided Parametric Search,” disclose an interface for identifying a single item from a family of items. The interface provides users with a set of lists of features present in the family of items and identifies items that satisfy selected features. Other navigation systems include i411's Discovery Engine, Cybrant's Information Engine, Mercado's IntuiFind, and Requisite Technology's BugsEye.

SUMMARY OF THE INVENTION

[0011] The present invention, a highly scalable, hierarchical, data-driven information navigation system and method, enables the navigation of a collection of documents or other materials using certain common attributes associated with those materials. The navigation system interface allows the user to select values for the attributes associated with the materials in the current navigation state and returns the materials that correspond to the user's selections. In some embodiments, the user's selections may be constrained using Boolean operators. The present invention enables this navigation mode by associating terms (attribute-value pairs) with the documents, defining a set of hierarchical refinement relationships (i.e., a partial order) among the terms, and providing a guided navigation mechanism based on the association of terms with documents and the relationships among the terms.

[0012] The present invention includes several components and features relating to a hierarchical data-driven navigation system. Among these are a user interface, a knowledge base, a process for generating and maintaining the knowledge base, a navigable data structure and method for generating the data structure, WWW-based applications of the system, and methods of implementing the system. Although the invention is described herein primarily with reference to a WWW-based system for navigating a product database, it should be understood that a similar navigation system could be employed in any database context where materials may be associated with terms and users can identify materials of interest by way of those terms.

[0013] The present invention uses a knowledge base of information regarding the collection of materials to formulate and to adapt the interface to guide the user through the collection of navigation states by providing relevant navigation options. The knowledge base includes an enumeration of attributes relevant to the materials, a range of values for each attribute, and a representation of the partial order that relates terms (the attribute-value pairs). Attribute-value pairs for materials relating to entertainment, for example, may be Products: Movies and Director: Spike Lee. (Attribute-value pairs are represented throughout this specification in this Attribute: Value format; navigation states are represented as bracketed expressions of attribute-value pairs.) The knowledge base also includes a classification mapping that associates each item in the collection of materials with a set of terms that characterize that item.

[0014] The knowledge base is typically organized by domains, which are sets of materials that conform to natural groupings. Preferably, a domain is chosen such that a manageable number of attributes suffice to effectively distinguish and to navigate among the materials in that domain. The knowledge base preferably includes a characterization of each domain, which might include rules or default expectations concerning the classification of documents in that domain. A particular item may be in more than one domain.

[0015] The present invention includes a user interface for navigation. The user interface preferably presents the user's navigation state as an expression of terms organized by attribute. For a given expression of terms, the user interface presents materials that are associated with those terms in accordance with that expression and presents relevant navigation options for narrowing or for generalizing the navigation state. In one aspect of the present invention, users navigate through the collection of materials by selecting and deselecting terms. In one aspect of the present invention, the user interface responds immediately to the selection or the deselection of terms, rather than waiting for the user to construct and to submit a comprehensive query composed of multiple terms. Once a query has been executed, the user may narrow the navigation state by conjunctively selecting additional terms, or by refining existing terms. Alternatively, the user may broaden the navigation state by deselecting terms that have already been conjunctively selected or by generalizing the terms. In preferred embodiments, the user may broaden the navigation state by deselecting terms in an order different from that in which they were conjunctively selected. For example, a user could start at {Products: Movies}, narrow by conjunctively selecting an additional term to {Products: Movies AND Genre: Drama }, narrow again to {Products: Movies AND Genre: Drama AND Director: Spike Lee }, and then broaden by deselecting a term to {Products: Movies AND Director: Spike Lee}.

[0016] In another aspect of the present invention, the user may broaden the navigation state by disjunctively selecting additional terms. For example, a user could start at {Products: DVDs}, and then broaden by disjunctively selecting a term to {Products: DVDs OR Products: Videos}, and then narrow by conjunctively selecting a term to {(Products: DVDs OR Products: Videos) AND Director: Spike Lee}.

[0017] In another aspect of the present invention, the user may narrow the navigation state by negationally selecting additional terms. For example, a user could start at {Products: DVDs}, narrow by conjunctively selecting a term to {Products: DVDs AND Genre: Comedy}, and then narrow by negationally selecting a term to {Products: DVDs AND Genre: Comedy AND (NOT Director: Woody Allen)}.

[0018] In another aspect of the present invention, the user interface allows users to use free-text search to find terms of interest. In another aspect of the present invention, the user interface also allows users to use free-text search on descriptive information associated with the materials.

[0019] In another aspect of the present invention, the user interface presents users with context-dependent navigation options for modifying the navigation state. The user interface does not present the user with options whose selection would correspond to no documents in the resulting navigation state. Also, the user interface presents new navigation options as they become relevant. The knowledge base may contain rules that determine when particular attributes or terms are made available to users for navigation.

[0020] In another aspect of the invention—for example, when the materials correspond to products available for purchase from various sources—the knowledge base includes a catalog of canonical representations that have been aggregated from the materials.

[0021] In another aspect of the invention, the knowledge base may include definitions of stores, sets of materials that are grouped to be searchable at one time. A store may include documents from one or more domains. An item may be assigned to more than one store. The knowledge base may also include rules to customize navigation for particular stores.

[0022] In another aspect of the invention, the knowledge base is developed through a multi-stage, iterative process. Workflow management allocates resources to maximize the efficiency of generating and of maintaining the knowledge base. The knowledge base is used to generate data structures that support navigation through a collection of materials. In one aspect of the invention, the navigation system consists of a hierarchy (i.e., a partial order) of navigation states that map expressions of terms to the sets of materials with which those terms are associated. In another aspect of the invention, the navigation states are related by transitions corresponding to terms used to narrow or broaden from one navigation state to another. The navigation states may be fully or partially precomputed, or may be entirely computed at run-time. In another aspect of the invention, implementations of the invention may be scalable through parallel or distributed computation. In addition, implementations of the invention may employ master and slave servers arranged in a hierarchical configuration.

BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The invention, including these and other features thereof, may be more fully understood from the following description and accompanying drawings, in which:

[0024] FIG. 1 is a view of a user interface to a navigation system in accordance with an embodiment of the present invention.

[0025] FIG. 2 is a view of the user interface of FIG. 1, showing a drop-down pick list of navigable terms.

[0026] FIG. 3 is a view of the user interface of FIG. 1, showing a navigation state.

[0027] FIG. 4 is a view of the user interface of FIG. 1, showing a navigation state.

[0028] FIG. 5 is a view of the user interface of FIG. 1, showing a navigation state.

[0029] FIG. 6 is a view of the user interface of FIG. 1, showing a navigation state.

[0030] FIG. 7 is a view of the user interface of FIG. 1, showing a navigation state.

[0031] FIG. 8 is a view of the user interface of FIG. 1, showing a navigation state.

[0032] FIG. 9 is a view of the user interface of FIG. 1, showing the result of a free-text search for terms.

[0033] FIG. 10 is a view of the user interface of FIG. 1, showing information about a particular document.

[0034] FIGS. 11A-C are representative examples of how the range of values for an attribute could be partially ordered in accordance with an embodiment of the present invention.

[0035] FIG. 12 is a block diagram of a process for collecting and classifying documents in accordance with an embodiment of the present invention.

[0036] FIG. 13 is a table illustrating how a set of documents may be classified in accordance with an embodiment of the present invention.

[0037] FIG. 14 is a representative partial order of navigation states in accordance with an embodiment of the present invention.

[0038] FIG. 15 is a block diagram of a process for precomputing a navigation state in accordance with an embodiment of the present invention.

[0039] FIG. 16 is a view of a user interface to a navigation system in accordance with an embodiment of the invention, showing disjunctive selection.

[0040] FIG. 17 is a view of a user interface to a navigation system in accordance with an embodiment of the invention, showing disjunctive selection.

[0041] FIG. 18 is a view of a user interface to a navigation system in accordance with an embodiment of the invention, showing negational selection.

[0042] FIG. 19 is a view of a user interface to a navigation system in accordance with an embodiment of the invention, showing negational selection.

[0043] FIG. 20 is a block diagram of a system and a method for processing a request across multiple servers in accordance with an embodiment of the present invention.

[0044] FIG. 21 is a flow diagram of steps for combining refinement options from slave servers in accordance with an embodiment of the present invention.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0045] User Interface

[0046] In accordance with one embodiment of the present invention, FIG. 1 shows a user interface 10 to a hierarchical, data-driven navigation system. The navigation system operates on a collection of documents defined in a knowledge base. As is shown, the user is preferably presented with at least two alternative methods of using the navigation system: (1) by selecting terms to navigate through the collection of documents, or (2) by entering a desired keyword in a search box.

[0047] The navigation system preferably organizes documents by domain. In accordance with one embodiment of the present invention, the user interface 10 shown in FIGS. 1-10 is operating on a set of documents that are part of a wine domain. Preferably, a domain defines a portion of the collection of documents that reflects a natural grouping. Generally, the set of attributes used to classify documents in a domain will be a manageable subset of the attributes used to classify the entire collection of documents. A domain definition may be a type of product, e.g., wines or consumer electronics. A domain may be divided into subdomains to further organize the collection of documents. For example, there can be a consumer electronics domain that is divided into the subdomains of televisions, stereo equipment, etc. Documents may correspond to goods or services.

[0048] The user interface may allow users to navigate in one domain at a time. Alternatively, the user interface may allow the simultaneous navigation of multiple domains, particularly when certain attributes are common to multiple domains.

[0049] The user interface allows the user to navigate through a collection of navigation states. Each state is composed of an expression of terms and of the set of documents associated with those terms in accordance with that expression. In the embodiment shown in FIGS. 1-10, users navigate through the collection of navigation states by conjunctively selecting and deselecting terms to obtain the navigation state corresponding to each expression of conjunctively selected terms. Preferably, as in FIG. 4, the user interface 10 presents a navigation state by displaying both the list 50 of terms 52 and a list 41 of some or all of the documents 42 that correspond to that state. Preferably, the user interface presents the terms 52 of the navigation state organized by attribute. Preferably, the initial navigation state is a root state that corresponds to no term selections and, therefore, to all of the documents in the collection.

[0050] As shown in FIG. 2, the user interface 10 allows users to narrow the navigation state by choosing a value 28 for an attribute 22, or by replacing the currently selected value with a more specific one (if appropriate). Preferably, the user interface 10 presents users with the options available to narrow the present navigation state, preferably with relevant terms organized by attribute. In some embodiments of the present invention, as shown in FIG. 2, users can select values 28 from drop-down lists 26 denoted by indicators 24, that are organized by attributes 22 in the current navigation state. The user interface may present these navigation options in a variety of formats. For example, values can be presented as pictures or as symbols rather than as text. The interface may allow for any method of selecting terms, e.g., mouse clicks, keyboard strokes, or voice commands. The interface may be provided through various media and devices, such as television or WWW, and telephonic or wireless devices. Although discussed herein primarily as a visual interface, the interface may also include an audio component or be primarily audio-based.

[0051] Preferably, in the present navigation state, the user interface only presents options for narrowing the navigation state that lead to a navigation state with at least one document. This preferred criteria for providing navigation options ensures that there are no “dead ends,” or navigation states that correspond to an empty result set.

[0052] Preferably, the user interface only presents options for narrowing the navigation state if they lead to a navigation state with strictly fewer documents than the present one. Doing so ensures that the user interface does not present the user with choices that are already implied by terms in the current navigation state.

[0053] Preferably, the user interface presents a new navigation state as soon as the user has chosen a term 28 to narrow the current navigation state, without any further triggering action by the user. Because the system responds to each user with immediate feedback, the user need not formulate a comprehensive query and then submit the query.

[0054] In accordance with one embodiment of the present invention, as shown in FIGS. 3 and 4, the user interface 10 may enable broadening of the current navigation state by allowing the user to remove terms 52 from the list 50 of terms conjunctively selected. For example, the interface 10 may provide a list 50 with checkboxes 54 for removing selections and a button 56 to trigger the new search. In the illustrated embodiment, the user can remove conjunctively selected terms 52 in any order and can remove more than one selection 52 at a time.

[0055] Preferably, the navigation options presented to the user are context-dependent. For example, terms that refine previously selected terms may become navigation options in the resulting navigation state. For example, referring to FIG. 5, after the term Flavors: Wood and Nut Flavors 52 is conjunctively selected (the user has selected the value Wood and Nut Flavors 23 for the attribute Flavors), Wood and Nut Flavors 23 then appears in the interface for the new navigation state in the list 20 of attributes and allows conjunctive selection of values 28 that relate to that specific attribute for further refinement of the query. The user interface may also present certain attributes that were not presented initially, as they become newly relevant. For example, comparing FIG. 3 to FIG. 2, the attribute French Vineyards 25 appears in the list 20 of attributes only after the user has already conjunctively selected the term Regions: French Regions in a previous navigation state. Attributes may be embedded in this way to as many levels as are desired. Presenting attributes as navigation options when those attributes become relevant avoids overwhelming the user with navigation options before those options are meaningful.

[0056] Additionally, for some attributes 22, multiple incomparable (non-refining) conjunctive selections of values 28 may be applicable. For example, for the attribute Flavor, the values Fruity and Nutty, neither of which refines the other, may both be conjunctively selected so that the terms Flavors: Fruity and Flavors: Nutty narrow the navigation state. Thus, users may sometimes be able to refine a query by conjunctively selecting multiple values under a single attribute.

[0057] Preferably, certain attributes will be eliminated as navigation options if they are no longer valid or helpful choices. For example, if all of the documents in the result set share a common term (in addition to the term(s) selected to reach the navigation state), then conjunctive selection of that term will not further refine the result set; thus, the attribute associated with that term is eliminated as a navigation option. For example, comparing FIG. 6 with FIG. 4, the attribute Wine Types 27 has been eliminated as a navigation option because all of the documents 42 in the result set share the same term, Wine Types: Appellational Wines. In preferred embodiments, an additional feature of the interface 10 is that this information is presented to the user as a common characteristic of the documents 42 in the result set. For example, referring to FIG. 6, the interface 10 includes a display 60 that indicates the common characteristics of the documents 42 in the result set. Removing a term as a navigation option when all of the documents in the result set share that term prevents the user from wasting time by conjunctively selecting terms that do not refine the result set.

[0058] Preferably, the user interface also eliminates values as navigation options if their selection would result in no documents in the result set. For example, comparing FIG. 8 to FIG. 7, after the user selects the term Wine Spectator Range: 95-100, the user interface eliminates as navigation options all the values 28, 29 in the list 26 of values for the attribute Appellations 22 except for the values Alexander Valley 29 and Napa Valley 29. Alexander Valley 29 and Napa Valley 29 are the only two values in the list 26 of values for the attribute Appellations that return at least one document in the result set; all other values 28 return the empty set. Removing values as navigation options that would result in an empty result set saves the user time by preventing the user from reaching dead-ends.

[0059] Preferably, the user interface allows users to search for desired words using free text search. In accordance with one embodiment of the present invention, illustrated in FIG. 9, a search box 30 preferably allows users to perform a free-text search for terms of interest, rather than performing a full-text search of the documents themselves. Preferably, the user interface responds to such a search by presenting a list 32 of terms 34 including terms organized by attribute 36, and allowing the user to select from among them. Preferably, the user interface responds to the user's selection by presenting the user with the navigation state corresponding to the selection of that term. The user may then either navigate from that state (i.e., by narrowing or broadening it) or perform additional free-text searches for terms.

[0060] Preferably, the user interface 10 presents a full or partial list 41 of the documents that correspond to the current navigation state. Preferably, if a user is interested in a particular document 42, the user may select it and obtain a record 70 containing further information about it, including the list 72 of terms 74 that are associated with that document, as shown in FIG. 10. Preferably, the user interface 10 allows the user to conjunctively select any subset of those terms 74 and thereby navigate to the navigation state that corresponds to the selected term expression.

[0061] Preferably, the user interface 10 also offers navigation options that directly link to an associated navigation state that is relevant to, but not necessarily a generalization or refinement of, the present navigation state. These links preferably infer the user's interests from the present navigation state and enable the user to cross-over to a related topic. For example, if the user is visiting a particular navigation state in a food domain, links may direct the user to navigation states of wines that would complement those foods in the wine domain.

[0062] In accordance with another embodiment of the present invention, the user is preferably presented with additional methods of using the navigation system such as: (1) by conjunctively selecting terms, (2) by disjunctively selecting terms, (3) by negationally selecting terms, or (4) by entering a desired keyword in a search box.

[0063] In another aspect of the present invention, the user may broaden the navigation state by disjunctively selecting additional terms. For example, a user could start at {Products: DVDs}, and then broaden by disjunctively selecting a term to {Products: DVDs OR Products: Videos}, and then narrow by conjunctively selecting a term to {(Products: DVDs OR Products: Videos) AND Director: Spike Lee}. FIG. 16 shows a user interface 300 to a hierarchical, data-driven navigation system. The user interface 300 is operating on a collection of records relating to mutual funds. The interface 300 presents navigation options, including a list of attributes 310 relating to mutual funds and a list of terms 314 for a particular attribute 312, such as Fund Family, under consideration by a user. A selected term 316 is highlighted. As shown, the attribute-value pair {Fund Family: Fidelity Investments} has previously been selected. The illustrated navigation system allows the user to select attribute-value pairs disjunctively. As shown in FIG. 17, after the user subsequently selects {Fund Family: Vanguard Group}in addition, the interface 300 presents a new navigation state {Fund Family: Fidelity Investments OR Fund Family: Vanguard Group}, including mutual funds 320 that match either selected attribute-value pair. Accordingly, both selected attribute-value pairs 316 are highlighted. In some embodiments, for example, to reduce computational requirements, disjunctive combination of attribute-value pairs may be limited to mutually incomparable attribute-value pairs that correspond to the same attribute.

[0064] In another aspect of the present invention, the user may narrow the navigation state by negationally selecting additional terms. For example, a user could start at {Products: DVDs}, narrow by conjunctively selecting a term to {Products: DVDs AND Genre: Comedy}, and then narrow by negationally selecting a term to {Products: DVDs AND Genre: Comedy AND (NOT Director: Woody Allen)}. FIG. 18 shows another interface 400 to a hierarchical, data-driven navigation system. The user interface 400 is operating on a collection of records relating to entertainment products. The user interface 400 includes a header 410 and a navigation area 412. The header 410 indicates the present navigation state {Products: DVDs AND Genre:Drama}, and implies the refinement options currently under consideration by the user. The leader “Not Directed By” 414 indicates a negational operation with respect to the Director attribute. The interface lists the attribute-value pairs 416 that can be combined with the expression for the present navigation state under this operation. As shown in FIG. 19, after the user selects the term Director: Martin Scorsese, the interface 400 presents a new navigation state {Products: DVDs AND Genre:Drama AND (NOT Director: Martin Scorsese}.

[0065] Although the interface to the navigation system has been described herein as a user interface, the interface could provide other forms of access to the navigation system. In alternative embodiments, the interface may be an applications program interface to allow access to the navigation system for or through other applications. The interface may also enhance the functionality of an independent data-oriented application. The interface may also be used in the context of a WWW-based application or an XML-based application. The navigation system may also support multiple interface modes simultaneously. The navigation system may be made available in a variety of ways, for example via wireless communications or on handheld devices.

[0066] Knowledge Base

[0067] Preferably, the navigation system stores all information relevant to navigation in a knowledge base. The knowledge base is the repository of information from two processes: taxonomy definition and classification. Taxonomy definition is the process of identifying the relevant attributes to characterize documents, determining the acceptable values for those attributes (such as a list or range of values), and defining a partial order of refinement relationships among terms (attribute-value pairs). Classification is the process of associating terms with documents. The knowledge base may also be used to maintain any information assets that support these two processes, such as domains, classification rules and default expectations. Additionally, the knowledge base may be used to maintain supplementary information and materials that affect users' navigation experience.

[0068] The taxonomy definition process identifies a set of attributes that appropriately characterize documents. A typical way to organize the taxonomy definition process is to arrange the collections of documents into domains, which are sets of documents that conform to a natural grouping and for which a manageable number of attributes suffice to effectively distinguish and navigate among the documents in that domain. The knowledge base preferably includes a characterization of each domain, which might include rules or default expectations concerning the classification of documents in that domain.

[0069] The taxonomy definition process also identifies a full set of values, at varying levels of specificity when appropriate, for each attribute. The values preferably identify the specific properties of the documents in the collection. The values may be enumerated explicitly or defined implicitly. For example, for a “color” attribute, a full set of valid color values may be specified, but for a “price” or “date” attribute, a range within which the values may fall or a general data type, without defining a range, may be specified. The process of identifying these values may include researching the domain or analyzing the collection of documents.

[0070] The taxonomy definition process also defines a partial order of refinement relationships among terms (attribute-value pairs). For example, the term Origin: France could refine the term Origin: Europe. The refinement relationship is transitive and antisymmetric but not necessarily total. Transitivity means that, if term A refines term B and term B refines term C, then term A refines term C. For example, if Origin: Paris refines Origin: France and Origin: France refines Origin: Europe, then Origin: Paris refines Origin: Europe. Antisymmetry means that, if two terms are distinct, then both terms cannot refine each other. For example, if Origin: Paris refines Origin: France, then Origin: France does not refine Origin: Paris.

[0071] Further, the partial order of refinement relationships among terms is not necessarily a total one. For example, there could be two terms, Origin: France and Origin: Spain, such that neither term refines the other. Two terms with this property are said to be incomparable. Generally, a set of two or more terms is mutually incomparable if, for every pair of distinct terms chosen from that set, the two terms are incomparable. Typically, but not necessarily, two terms with distinct attributes will be incomparable.

[0072] Given a set of terms, a term is a maximal term in that set if it does not refine any other terms in the set, and it is a minimal term in that set if no other term in the set refines it. For example, in the set {Origin: France, Origin: Paris, Origin: Spain, Origin: Madrid}, Origin: France and Origin: Spain are maximal, while Origin: Paris and Origin: Madrid are minimal. In the knowledge base, a term is a root term if it does not refine any other terms and a term is a leaf term if no other term refines it.

[0073] FIGS. 11A, 11B, and 11C illustrate attributes 112 and values 114, arranged in accordance with the partial order relationships, that could be used for classifying wines. The attributes 112 are Type/Varietal, Origin, and Vintage. Each attribute 112 corresponds to a maximal term for that attribute. An attribute 112 can have a flat set of mutually incomparable values (e.g., Vintage), a tree of values (e.g., Origin), or a general partial order that allows a value to refine a set of two or more mutually incomparable values (e.g., Type/Varietal). The arrows 113 indicate the refinement relationships among values 114.

[0074] Attributes and values may be identified and developed in several ways, including manual or automatic processing and the analysis of documents. Moreover, this kind of analysis may be top-down or bottom-up; that is, starting from root terms and working towards leaf terms, or starting from leaf terms and working towards root terms. Retailers, or others who have an interest in using the present invention to disseminate information, may also define attributes and terms.

[0075] The classification process locates documents in the collection of navigation states by associating each document with a set of terms. Each document is associated with a set of mutually incomparable terms, e.g., {Type/Varietal: Chianti, Origin: Italy, Vintage: 1996}, as well as any other desired descriptive information. If a document is associated with a given term, then the document is also associated with all of the terms that the given term refines.

[0076] The classification process may proceed according to a variety of workflows. Documents may be classified in series or in parallel, and the automatic and manual classification steps may be performed one or more times and in any order. To improve accuracy and throughput, human experts may be assigned as specialists to oversee the classification task for particular subsets of the documents, or even particular attributes for particular subsets of the documents. In addition, the classification and taxonomy processes may be interleaved, especially as knowledge gained from one process allows improvements in the other.

[0077] FIG. 12 illustrates the stages in a possible flow for the classification process 250. The data acquisition step 252, that is, the collection of documents for the database, may occur in several different ways. For example, a retailer with a product catalog over which the navigation system will operate might provide a set of documents describing its products as a pre-defined set. Alternatively, documents may be collected from one source, e.g., one Web site, or from a number of sources, e.g., multiple Web sites, and then aggregated. If the desired documents are Web pages, the documents may be collected by appropriately crawling the Web, selecting documents, and discarding documents that do not fit in the domain. In the data translation step 254, the collected documents are formatted and parsed to facilitate further processing. In the automatic classification step 256, the formatted and parsed documents are processed in order to automatically associate documents with terms. In the manual classification step 258, human reviewers may verify and amend the automatic classifications, thereby ensuring quality control. Preferably, any rules or expectations violated in either the automatic classification step 256 or the manual classification step 258 would be flagged and presented to human reviewers as part of the manual classification step 258. If the collection of documents is divided into domains, then there will typically be rules that specify a certain minimal or preferred set of attributes used to classify documents from each domain, as well as other domain-specific classification rules. When the classification process is complete, each document will have a set of terms associated with it, which locate the document in the collection of navigation states.

[0078] In FIG. 13, table 180 shows a possible representation of a collection of classified wine bottles. Preferably, each entry is associated with a document number 182, which could be a universal identifier, a name 184, and the associated terms 186. The name is preferably descriptive information that could allow the collection to be accessed via a free-text search engine as well as via the term-based navigation system.

[0079] In another aspect of the invention, the knowledge base also includes a catalog of canonical representations of documents. Each catalog entry represents a conceptually distinct item that may be associated with one or more documents. The catalog allows aggregation of profile information from multiple documents that relate to the item, possibly from multiple sources. For example, if the same wine is sold by two vendors, and if one vendor provides vintage and geographic location information and another provides taste information, that information from the two vendors can be combined in the catalog entry for that type of wine. The catalog may also improve the efficiency of the classification process by eliminating duplicative profiling. In FIG. 12, the catalog creation step 260 associates classified documents with catalog entries, creating new catalog entries when appropriate. For ease of reference, an item may be uniquely identified in the catalog by a universal identifier.

[0080] The knowledge base may also define stores, where a store is a subcollection of documents that are grouped to be searchable at one time. For example, a particular online wine merchant may not wish to display documents corresponding to products sold by that merchant's competitors, even though the knowledge base may contain such documents. In this case, the knowledge base can define a store of documents that does not include wines sold by the merchant's competitors. In FIG. 12, the store creation step 262 may define stores based on attributes, terms, or any other properties of documents. A document may be identified with more than one store. The knowledge base may also contain attributes or terms that have been customized for particular stores.

[0081] In FIG. 12, the export process step 264 exports information from the knowledge base to another stage in the system that performs further processing necessary to generate a navigable data structure.

[0082] Navigation States

[0083] The navigation system represents, explicitly or implicitly, a collection of navigation states. A navigation state can be represented either by an expression of terms, or by the subset of the collection of documents that correspond to the term expression.

[0084] By way of example, types of navigation states include conjunctive navigation states, disjunctive navigation states and negational navigation states. Conjunctive navigation states are a special case of navigation states in which the term expression is conjunctive—that is, the expression combines terms using only the AND operator. Conjunctive navigation states are related by a partial order of refinement that is derived from the partial order that relates the terms.

[0085] In one aspect of the present invention, a conjunctive navigation state has two representations. First, a conjunctive navigation state corresponds to a subset of the collection of documents. Second, a conjunctive navigation state corresponds to a conjunctive expression of mutually incomparable terms. FIG. 14 illustrates some navigation states for the documents and terms based on the wine example discussed above. For example, one navigation state 224 is {Origin: South America} (documents #1, #4, #5); a second navigation state 224 is {Type/Varietal: White AND Origin: United States} (documents #2, #9). The subset of documents corresponding to a conjunctive navigation state includes the documents that are commonly associated with all of the terms in the corresponding expression of mutually incomparable terms. At the same time, the expression of mutually incomparable terms corresponding to a conjunctive navigation state includes all of the minimal terms from the terms that are common to the subset of documents, i.e., the terms that are commonly associated with every document in the subset. A conjunctive navigation state is preferably unique and fully specified; for a particular conjunctive expression of terms, or for a given set of documents, there is no more than one corresponding conjunctive navigation state.

[0086] One way preferred to define the collection of conjunctive navigation states is to uniquely identify each conjunctive navigation state by a canonical conjunctive expression of mutually incomparable terms. A two-step mapping process that maps an arbitrary conjunctive expression of terms to a canonical conjunctive expression of mutually incomparable terms creates states that satisfy this property. In the first step of the process, an arbitrary conjunctive expression of terms is mapped to the subset of documents that are associated with all of those terms. Recalling that if a document is associated with a given term, then the document is also associated with all of the terms that the given term refines, in the second step of the process, this subset of documents is mapped to the conjunctive expression of minimal terms among the terms that are common to all of the documents in that document set. The result of this second step is a conjunctive expression of mutually incomparable terms that uniquely identifies the corresponding subset of documents, and, hence, is a canonical representation for a conjunctive navigation state. By way of illustration, referring to the wine example in FIG. 14, the term expression {Origin: France} maps to the subset of documents {documents #8, #11 }, which in turn maps to the canonical term expression { Type/Varietal: Red AND Origin: France}.

[0087] The conjunctive navigation states 222, 224, 226 are related by a partial order of refinement relationships 220 derived from the partial order that relates terms. This partial order can be expressed in terms of either the subsets of documents or the term expressions that define a conjunctive navigation state. Expressed in terms of subsets of documents, a navigation state A refines a navigation state B if the set of documents that corresponds to state A is a subset of the set of documents that corresponds to state B. Expressed in terms of term expressions, a conjunctive navigation state A refines a conjunctive navigation state B if all of the terms in state B either are in state A or are refined by terms in state A. Referring to FIG. 14, the navigation state 226 corresponding to the term expression {Type/Varietal: Red AND Origin: Chile} (document #4) refines the navigation state 224 corresponding to {Origin: Chile} (documents #4, #5). Since the refinement relationships among navigation states give rise to a partial order, they are transitive and antisymmetric. In the example, {Type/Varietal: Red AND Origin: Chile} (document #4) refines {Origin: Chile} (documents #4, #5) and {Origin: Chile} (documents #4, #5) refines {Origin: South America} (documents #1, #4, #5); therefore, {Type/Varietal: Red AND Origin: Chile} (document #4) refines {Origin: South America} (documents #1, #4, #5). The root navigation state 222 is defined to be the navigation state corresponding to the entire collection of documents. The leaf navigation states 226 are defined to be those that cannot be further refined, and often (though not necessarily) correspond to individual documents. There can be arbitrarily many intermediate navigation states 224 between the root 222 and the leaves 226. Given a pair of navigation states A and B where B refines A, there can be multiple paths of intermediate navigation states 224 connecting A to B in the partial order. For convenience of definition in reference to the implementation described herein, a navigation state is considered to refine itself.

[0088] A user browses the collection of documents by visiting a sequence of one or more navigation states typically starting at the root navigation state 222. In one embodiment of the present invention, there are three basic modes of navigation among these states. The first mode is refinement, or moving from the current navigation state to a navigation state that refines it. The user can perform refinement either by adding a term through conjunctive selection to the current navigation state or by refining a term in the current navigation state; i.e., replacing a term with a refinement of that term. After the user adds or refines a term, the new term expression can be mapped to a canonical term expression according to the two-step mapping described above. The second mode is generalization, or moving from the current navigation state to a more general navigation state that the current state refines. The user can perform generalization either by removing a term from the current navigation state or by generalizing a term in the current navigation state; i.e., replacing a current term with a term that the current term refines. After the user removes or generalizes a term, the new term expression can be mapped to a canonical term expression. The third mode is simply creating a query in the form of a desired term expression, which again can be mapped to a canonical term expression to obtain a navigation state.

[0089] In other embodiments of the present invention, there are additional modes of navigation. In systems that support the corresponding types of navigation states, these modes may include generalization of the navigation state through disjunctive selection, as shown in FIG. 16, as well as refinement of the navigation state through negational selection, as shown in FIG. 17. In general, terms can be combined using Boolean logic. Although term expressions that are not conjunctive do not necessarily have canonical forms, some implementations may be based on a system that uses a collection of conjunctive navigation states. One implementation is based on logical equivalence rules as described below.

[0090] Implementation

[0091] The knowledge base is transferred to a navigable data structure in order to implement the present invention. The navigation states may be fully precomputed, computed dynamically at run-time, or partially precomputed. A cache may be used to avoid redundant computation of navigation states.

[0092] In preferred embodiments, the collection of conjunctive navigation states may be represented as a graph—preferably, a directed acyclic multigraph with labeled edges. A graph is a combinatorial structure consisting of nodes and edges, where each edge links a pair of nodes. The two nodes linked by an edge are called its endpoints. With respect to the present invention, the nodes correspond to conjunctive navigation states, and the edges represent transitions that refine from one conjunctive navigation state to another. Since refinement is directional, each edge is directed from the more general node to the node that refines it. Because there is a partial order on the navigation states, there can be no directed cycles in the graph, i.e., the graph is acyclic. Preferably, the graph is a multigraph, since it allows the possibility of multiple edges connecting a given pair of nodes. Each edge is labeled with a term. Each edge has the property that starting with the term set of the more general end point, adding the edge term, and using the two-step map to put this term set into canonical form leads to a refinement which results in the navigation state that is the other endpoint. That is, each edge represents a refinement transition between nodes based on the addition of a single term.

[0093] The following definitions are useful for understanding the structure of the graph: descendant, ancestor, least common ancestor (LCA), proper ancestor, proper descendant, and greatest lower bound (GLB). These definitions apply to the refinement partial order among terms and among nodes. If A and B are terms and B refines A, then B is said to be a descendant of A and A is said to be an ancestor of B. If, furthermore, A and B are distinct terms, then B is said to be a proper descendant of A and A is said to be a proper ancestor of B. The same definitions apply if A and B are both nodes.

[0094] If C is an ancestor of A and C is also an ancestor of B, then C is said to be a common ancestor of A and B, where A, B, and C are either all terms or all nodes. The minimal elements of the set of common ancestors of A and B are called the least common ancestors (LCAs) of A and B. If no term has a pair of incomparable ancestors, then the LCA of two terms-or of two nodes-is unique. For example, the LCA of Origin: Argentina and Origin: Chile is Origin: South America in the partial order of terms 110 of FIG. 11B. In general, however, there may be a set of LCAs for a given pair of terms or nodes.

[0095] In an implementation that fully precomputes the collection of nodes, computation of the nodes in the graphs is preferably performed bottom-up.

[0096] The leaf nodes in the graph—that is, the nodes corresponding to leaf navigation states—may be computed directly from the classified documents. Typically, but not necessarily, a leaf node will correspond to a set containing a single document. The remaining, non-leaf nodes are obtained by computing the LCA-closure of the leaf nodes—that is, all of the nodes that are the LCAs of subsets of the leaf nodes.

[0097] The edges of the graph are determined according to a refinement function, called the R function for notational convenience. The R function takes as arguments two nodes A and B, where A is a proper ancestor of B, and returns the set of maximal terms such that, if term C is in R (A, B), then refining node A with term C results in a node that is a proper descendant of A and an ancestor (not necessarily proper) of B. For example, in FIG. 14, R ({Type/Varietal: Red}, {Type /Varietal: Merlot AND Origin: Argentina AND Vintage: 1998})={Type/Varietal: Merlot AND Origin: South America AND Vintage: 1998}. If B1 is an ancestor of B2, then R (A, B1) is a subset of R (A, B2)—assuming that A is a proper ancestor of both B1 and B2. For example, R ({Type/Varietal: Red}, {Type/Varietal: Red AND Origin: South America})={Origin: South America}

[0098] In the graph, the edges between nodes A and B will correspond to a subset of the terms in R (A, B). Also, no two edges from a single ancestor node A use the same term for refinement. If node A has a collection of descendant nodes {B1, B2, . . .} such that term C is in all of the R (A, Bi), then the only edge from node A with term C goes to LCA (B1, B2, . . . ), which is guaranteed to be the unique maximal node among the Bi. In FIG. 14, for example, the edge from node {Type/Varietal: Red} with term Origin: South America goes to node {Type/Varietal: Red AND Origin: South America} rather than to that node's proper descendants {Type/Varietal: Merlot AND Origin: South America AND Vintage: 1998} and {Type/Varietal: Red AND Origin: Chile}. The LCA-closure property of the graph ensures the existence of a unique maximal node among the Bi. Thus, each edge maps a node-term pair uniquely to a proper descendant of that node.

[0099] The LCA-closure of the graph results in the useful property that, for a given term set S, the set of nodes whose term sets refine S has a unique maximal node. This node is called the greatest lower bound (GLB) of S.

[0100] The graph may be computed explicitly and stored in a combinatorial data structure; it may be represented implicitly in a structure that does not necessarily contain explicit representations of the nodes and edges; or it may be represented using a method that combines these strategies. Because the navigation system will typically operate on a large collection of documents, it is preferred that the graph be represented by a method that is scalable.

[0101] The graph could be obtained by computing the LCAs of every possible subset of leaf nodes. Such an approach, however, grows exponentially in the number of leaf nodes, and is inherently not scalable. An alternative strategy for obtaining the LCA closure is to repeatedly consider all pairs of nodes in the graph, check if each pair's LCA is in the graph, and add that LCA to the graph as needed. This strategy, though a significant improvement on the previous one, is still relatively not scalable.

[0102] A more efficient way to precompute the nodes is to process the document set sequentially, compute the node for each document, and add that node to the graph along with any other nodes necessary to maintain LCA-closure. The system stores the nodes and edges as a directed acyclic multigraph. The graph is initialized to contain a single node corresponding to the empty term set, the root node. Referring to FIG. 15, in process 230 for inserting a new node into the graph, in step 232, for each new document to be inserted into the graph that does not correspond to an existing node, the system creates a new node. In step 234, before inserting the new node into the graph, the system recursively generates and inserts any missing LCA nodes between the root node (or ancestor node) and the new node. To ensure LCA-closure after every node insertion, the system inserts the document node last, in steps 236 and 238, after inserting all the other nodes that are proper ancestors of it.

[0103] Inserting a new node requires the addition of the appropriate edges from ancestors to the node, in step 236, and to descendants out of the new node, in step 238. The edges into the node are preferably determined by identifying the ancestors that have refinement terms that lead into the new node and do not already have those refinement terms used on edges leading to intermediate ancestors of the new node. The edges out of the node are preferably determined by computing the GLB of the new node and appropriately adding edges from the new node to the GLB and to nodes to which the GLB has edges.

[0104] The entire graph of conjunctive navigation states may be precomputed by following the above procedures for each document in the collection. Computation of other types of navigation states is discussed below. Precomputing of the graph may be preferred where the size of the graph is manageable, or if users are likely to visit every navigation state with equal probability. In practice, however, users typically visit some navigation states more frequently than others. Indeed, as the graph gets larger, some navigation states may never be visited at all. Unfortunately, reliable predictions of the frequency with which navigation states will be visited are difficult. In addition, it is generally not practical to precompute the collection of navigation states that are not conjunctive, as this collection is usually much larger than the collection of conjunctive navigation states.

[0105] An alternative strategy to precomputing the navigation states is to create indexes that allow the navigation states to be computed dynamically. Specifically, each document can be indexed by all of the terms that are associated with that document or that have refinements associated with that document. The resulting index is generally much smaller in size than a data structure that stores the graph of navigation states. This dynamic approach may save space and precomputation time, but it may do so at the cost of higher response times or greater computational requirements for operation. A dynamic implementation may use a one-argument version of the R function that returns all refinement terms from a given navigation state, as well a procedure for computing the GLB of a term set.

[0106] It is also possible to precompute a subset of the navigation states. It is preferable to precompute the states that will cost the most to compute dynamically. For example, if a state corresponds to a large subset of the documents, it may be preferable to compute it in advance. In one possible partial precomputation approach, all navigation states, particularly conjunctive ones, corresponding to a subset of documents above a threshold size may be precomputed. Precomputing a state is also preferable if the state will be visited frequently. In some instances it may be possible to predict the frequency with which a navigation state will be visited. Even if the frequency with which a navigation state will be visited cannot be predicted in advance, the need to continually recompute can be reduced by caching the results of dynamic computation. Most recently or most frequently visited states may be cached.

[0107] As described above with respect to the interface, the system supports at least three kinds of query operations—namely refinement, generalization, and query by specifying an expression of terms. These operations may be further described in terms of the graph. For query refinement, the system enumerates the terms that are on edges from the node corresponding to the current navigation state. When the user selects a term for refinement, the system responds by presenting the node to which that edge leads. Similarly, for query generalization options, the system enumerates and selects edges that lead to (rather than from) the node corresponding to the current navigation state. Alternatively, query generalization may be implemented as a special case of query by specifying a set of terms. For query by specifying a set of keywords, the system creates a virtual node corresponding to the given term set and determines the GLB of the virtual node in the graph. If no GLB is found, then there are no documents that satisfy the query. Otherwise, the GLB node will be the most general node in the graph that corresponds to a navigation state where all documents satisfy the query.

[0108] The above discussion focuses on how the system represents and computes conjunctive navigation states. In some embodiments of the present invention, the user interface only allows users to navigate among the collection of conjunctive navigation states. In other embodiments, however, users can navigate to navigation states that are not conjunctive. In particular, when the system supports navigation states that are not conjunctive, the user interface may allow users to select terms disjunctively or negationally.

[0109] If the system includes navigation states that are both conjunctive and disjunctive (e.g., {(Products: DVDs OR Products: Videos) AND Director: Spike Lee}), then in some embodiments, the system only precomputes a subset of the states, particularly if the total number of navigation states is likely to be too large to maintain in memory or even secondary (e.g., disk) storage. By using rules for equivalence of Boolean expressions, it is possible to express any navigation state that mixes conjunction and disjunction in terms of a union of conjunctive navigation states. The above example can be rewritten as {(Products: DVDs AND Director: Spike Lee) OR (Products: Videos AND Director: Spike Lee)}. This approach leads to an implementation combining conjunctive and disjunctive navigation states based on the above discussion, regardless of whether all, some, or none of the graph of conjunctive navigation states is precomputed.

[0110] In preferred embodiments, disjunctive selections may be made within, but not between, attributes. When determining the set of disjunctive generalizations, the system does not consider other terms from the attribute of the given disjunction to be in the navigation state. For example, if the navigation state is {Type/Varietal: Red AND Origin: Chile} and the system is allowing the disjunctive selection of other countries of origin, then the GLB and R function will be applied to the set {Type/Varietal: Red} rather than to {Type/Varietal: Red AND Origin: Chile}. Accordingly, the other terms for the attribute of country of origin that are incomparable to “Chile” become generalization options for the navigation state.

[0111] If the system includes navigation states that use negation (e.g., {Products: DVDs AND Genre: Comedy AND (NOT Director: Woody Allen)}), then the negationally selected terms can be applied to navigation states as a post-process filtering operation. The above example can be implemented by taking the conjunctive navigation state {Products: DVDs AND Genre: Comedy} and applying a filter to it that excludes all movies associated with the term Director: Woody Allen. This approach leads to an implementation including negational navigation states based on the above discussion, regardless of whether all, some, or none of the graph of conjunctive navigation states is precomputed.

[0112] As with disjunction, when determining the set of negational generalizations, the system does not consider other terms from the attribute of the given negation to be in the navigation state. For example, if the navigation state is {Medium: Compact Disc AND Artist: Prince} and the system is allowing the negational selection of other artists (e.g., {Artist: Prince AND NOT (Artist: The Revolution)}), then the GLB and R function will be applied to the set {Medium: Compact Disc} rather than to {Medium: Compact Disc AND Artist: Prince}.

[0113] Another aspect of the present invention is its scalability through parallel or distributed computation. One way to define scalability in a navigation system is in terms of four problem dimensions: the number of materials in the collection, the number of terms associated with each material in the collection, the rate at which the system processes queries (throughput), and the time necessary to process a query (latency). In this definition, a system as scalable if it can be scaled along any of these four dimensions at a subquadratic cost. In other words:

[0114] 1. If the number of materials in the collection is denoted by the variable n1 and the other three problem dimensions are held constant, then the resource requirements are subquadratic in n1.

[0115] 2. If the number of terms associated with each material in the collection is denoted by the variable n2 and the other three problem dimensions are held constant, then the resource requirements are subquadratic in n2.

[0116] 3. If the number of queries that the system processes per second (i.e., the throughput) is denoted by the variable n3 and the other three problem dimensions are held constant, then the resource requirements are subquadratic in n3.

[0117] 4. If the time necessary to process a query (i.e., the latency) is denoted by the variable n4 and the other three problem dimensions are held constant, then the resource requirements are subquadratic in 1/n4.

[0118] Preferably, these resource requirements would be not only subquadratic, but linear. Also included within the concept of scalability, there is an allowance for overhead in creating a network of distributed resources. Typically, this overhead will be logarithmic, since the resources may be arranged in a hierarchical configuration of bounded fan-out.

[0119] In some embodiments, the present invention surmounts the limitations of a single computational server's limited resources by allowing for distributing the task of computing the information associated with a navigation state onto a hierarchy of multiple computational servers that act in parallel.

[0120] One insight that drives this aspect of the present invention is that it is possible to partition the collection of materials among multiple “slave” servers, all of which implement the single-server algorithm for multidimensional navigation, and then to have a “master” server compute navigation states by passing requests onto the set of slave machines and combining the responses. From the outside, the collection of servers appears to act like a single server, but with far greater computational resources than would be possible on a single computational device. Indeed, the distinction between master and slave servers is arbitrary; a slave server can itself have slaves, thus creating a nested hierarchy of servers. Such nesting is useful when the number of slaves exceeds the fan-out capability of a single master server. An exemplary embodiment of such a system is illustrated in FIG. 20. In the hierarchical arrangement 500, a master server 520 works with slave servers 530, 540. In the hierarchical arrangement shown, slave servers 530 are in turn master servers with respects to slave servers 540. The search results are made available to a user on a terminal 510 through a user interface in accordance with the present invention.

[0121] The collection of materials may be partitioned by splitting (arbitrarily or otherwise) the materials into disjoint subsets, where each subset is assigned to its own server. The subsets may be roughly equal in size, or they might vary in size to reflect the differing computational resources available to each server.

[0122] The algorithm for distributing the task of computing the information associated with a navigation state includes three steps. The steps of the algorithm are indicated in FIG. 20. In the first step, the query, which is a request for a valid navigation state, is submitted to the master server 520, which forwards the query to each of the slave servers 530. If the servers are nested, the requests are forwarded through the hierarchy of servers 500 until they reach the leaf servers 540 in the hierarchy. In the second step, each slave server 530, 540 processes the query independently, based on the subset of the collection of materials that is in its partition. In the third step, the master server 520 combines the responses from the slave servers to produce a response for the original query. The master server 520 returns the response to the terminal 510.

[0123] The master server receives the original request and farms it out to the slave servers. Thus, in preferred embodiments, the only computation performed by the master server is to combine the results from the slave servers. Each slave server that receives a request computes the navigation state based on the subset of the collection assigned to it. The computation may involve any combination of conjunction, disjunction, and negation.

[0124] The master server, in contrast, only performs a combination step. The combination step involves producing a valid navigation state, including documents and corresponding refinement options, from the responses from the slave servers. Since the collection of materials has been partitioned into disjoint subsets, the documents identified by each of the slave servers can be combined efficiently as a disjoint union. Combining the various refinement options returned by each of the slave servers may require additional processing, as described below.

[0125] The slave servers all process the same query, but on different partitions of the collection of materials. They will generally return different sets of refinement options because a set of refinement options that is valid for one partition may be invalid for another. If the different sets are disjoint, and if the refinement options involve terms that do not themselves have refinement relationships, then the combination is a disjoint union.

[0126] Typically, there will be some overlap among the different sets of refinement options returned by each slave server. If the sets are not disjoint, duplicates can be eliminated in this combination step.

[0127] When there are refinement relationships among the terms that are refinement options returned by the slave servers, the combination algorithm computes, for every set of related terms, the least common ancestor or ancestors (LCA) of the terms, as defined by the partial order among the terms. One algorithm for combining the refinement options is outlined in FIG. 21. In step 552, the master server receives and takes the union of all of the terms, X1, X2, . . . Xn, returned as refinement options for the navigation state from the slave servers. In step 554, the master server computes the set of ancestors A1, A2, . . . An, for each of the terms, X1, X2, . . . xn, respectively. In step 556, the master server computes the intersection A of all of the sets of ancestors, A1, A2, . . . An. In step 558, the master server computes the set M of minimal terms in A. The set M, formed of the least common ancestors of the terms X1, X2, . . . Xn, returned by the slave servers, is the set of refinement options corresponding to the result navigation state. This combination procedure is applied whether the refinement options are conjunctive, disjunctive, or negational.

[0128] In summary, the master server receives a request for a navigation state, forwards this request to each of the slave servers, combines their results with a union operation, and then computes, for every set of terms, the least common ancestor or ancestors of the set.

[0129] There are at least two ways to compute the LCA of the terms. One approach is to store all non-leaf terms on the master server. This strategy is reasonably memory efficient, since, in practice, most of the terms are leaves (minimal elements) in the partial order. A second approach is to include the ancestors when returning the terms that are refinements. This approach saves memory at the expense of increasing the size of the data being transferred. The latter overhead is reasonable, since, in practice, a term typically has very few ancestors.

[0130] The navigation system of the present invention allows information providers to overlay a navigation system over any collection of documents. The knowledge base and navigation aspects of the invention can be performed independently by different providers, and information providers may outsource these functions to separate entities. Similarly, a generated knowledge base may be imported by a navigation specialist. Information providers may also outsource this navigation requirement to a navigation system provider. A navigation system provider could charge customers a license fee for the system independent of the amount of its usage. Alternatively, a navigation system provider could charge customers on a per-click basis, a per-purchase basis if products are available via the system, or per-transaction generated from a click through the navigation system. A navigation system provider could also function as an aggregator—compiling records from a number of sources, combining them into a global data set, and generating a navigation system to search the data set. The navigation system can be implemented as software provided on a disk, on a CD, in memory, etc., or provided electronically (such as over the Internet).

[0131] A navigation system in accordance with the present invention may also enhance user profiling capability and merchandising capability. The navigation system may maintain a profile of users based on the users' selections, including the particular paths selected to explore the collection of navigation states. Using the knowledge base, the system may also infer additional information regarding the users' preferences and interests by supplementing the selection information with information regarding related documents, attributes and terms in the knowledge base. That information may be used to market goods and services related to the documents of interest to the user.

[0132] The foregoing description has been directed to specific embodiments of the invention. The invention may be embodied in other specific forms without departing from the spirit and scope of the invention. The embodiments, figures, terms and examples used herein are intended by way of reference and illustration only and not by way of limitation. The scope of the invention is indicated by the appended claims and all changes that come within the meaning and scope of equivalency of the claims are intended to be embraced therein.

Claims

1. A navigation system for a set of materials, comprising: a plurality of attributes characterizing the materials; a plurality of values describing the materials, wherein each of the values has an association with at least one of the attributes and each association defines an attribute-value pair, and wherein some of the attribute-value pairs refine other of the attribute-value pairs; a plurality of navigation states, wherein each navigation state corresponds to a particular expression of attribute-value pairs and to a particular subset of the materials; and an interface, the interface providing a plurality of transitions, each transition providing a direct path between two of the navigation states, wherein each transition represents a change from the expression of attribute-value pairs corresponding to an originating navigation state to the expression of attribute-value pairs corresponding to a destination navigation state, wherein a series of one or more transitions provides a path between any two navigation states, there being more than one path between at least a first of the navigation states and a second of the navigation states.

2. The navigation system of claim 1, wherein at least one navigation state corresponds to an expression of attribute-value pairs that relates at least one attribute value pair conjunctively.

3. The navigation system of claim 1, wherein at least one navigation state corresponds to an expression of attribute-value pairs that relates at least one attribute value pair disjunctively.

4. The navigation system of claim 1, wherein at least one navigation state corresponds to an expression of attribute-value pairs that relates at least one attribute value pair negationally.

5. The navigation system of claim 1, wherein at least one of the transitions from an originating navigation state to a destination navigation state represents a refinement of the value of one of the attribute-value pairs in the corresponding expression of attribute-value pairs for the originating navigation state.

6. The navigation system of claim 1, wherein at least one of the transitions from an originating navigation state to a destination navigation state represents a generalization of the value of one of the attribute-value pairs in the corresponding expression of attribute-value pairs for the originating navigation state.

7. The navigation system of claim 1, wherein at least one of the transitions from an originating navigation state to a destination navigation state represents a deselection of the attribute of one of the attribute-value pairs in the corresponding expression of attribute-value pairs for the originating navigation state.

8. The navigation system of claim 1, wherein a first of the transitions from an originating navigation state to a first destination navigation state represents a selection of an attribute-value pair corresponding to a first attribute and a second of the transitions from the originating navigation state to a second destination navigation state represents a selection of an attribute-value pair corresponding to a second attribute.

9. The navigation system of claim 1, wherein at least one of the transitions from an originating navigation state to a destination navigation state represents a selection of an attribute-value pair corresponding to an attribute for which there is no corresponding attribute-value pair in the expression of attribute-value pairs corresponding to the originating navigation state.

10. The navigation system of claim 1, wherein at least one of the transitions from an originating navigation state to a destination navigation state represents disjunctive relation of an attribute-value pair to the expression of attribute-value pairs for the originating navigation state.

11. The navigation system of claim 1, wherein at least one of the transitions from an originating navigation state to a destination navigation state represents negational relation of an attribute-value pair to the expression of attribute-value pairs for the originating navigation state.

12. The navigation system of claim 1, wherein an originating navigation state corresponds to one of the materials in the set of materials, at least one of the transitions from the originating navigation state to a destination navigation state representing a selection of a particular attribute-value pair associated with the originating navigation state.

13. The navigation system of claim 1, wherein the values associated with at least one of the plurality of attributes are defined explicitly.

14. The navigation system of claim 1, wherein the values associated with at least one of the plurality of attributes are defined implicitly.

15. The navigation system of claim 1, wherein, for attribute-value pairs sharing a common attribute, no attribute-value pair refines a plurality of mutually incomparable attribute-value pairs.

16. The navigation system of claim 1, wherein, for attribute-value pairs sharing a common attribute, at least one attribute-value pair refines a plurality of mutually incomparable attribute-value pairs.

17. The navigation system of claim 1, wherein no attribute-value pair refines a plurality of mutually incomparable attribute-value pairs.

18. The navigation system of claim 1, wherein at least one attribute-value pair refines a plurality of mutually incomparable attribute-value pairs.

19. The navigation system of claim 1, wherein, for any two attribute-value pairs corresponding to different attributes, the two attribute-value pairs are incomparable.

20. The navigation system of claim 1, wherein the set of materials includes materials related to a single subject area.

21. The navigation system of claim 1, wherein the set of materials includes materials related to a plurality of subject areas.

22. The navigation system of claim 1, wherein the set of materials includes a subset of the materials, the subset being integrally navigable, a portion of the materials in the set of materials being assigned to the subset.

23. The navigation system of claim 22, wherein the interface is adapted to provide a plurality of transitions related to the subset of materials.

24. The navigation system of claim 22, wherein the set of materials includes a plurality of subsets, each of the plurality of subsets being independently integrally navigable, a portion of the materials in the set of materials being assigned to each subset, at least one of the materials being assigned to more than one subset.

25. The navigation system of claim 1, further including a profile for each of the materials in the set of materials, the profile including a set of attribute-value pairs.

26. The navigation system of claim 25, the profile further including descriptive information.

27. The navigation system of claim 1, the interface including a human user interface.

28. The navigation system of claim 1, the interface including an applications program interface.

29. The navigation system of claim 1, wherein the interface is operable in a World Wide Web-based environment.

30. The navigation system of claim 1, wherein the interface is operable in an XML-based environment.

31. The navigation system of claim 1, wherein the interface supplements the functionality of an independent data-oriented program.

32. The navigation system of claim 1, the interface including a guided search tool for enabling navigation from a current navigation state based on the plurality of transitions among the plurality of navigation states.

33. The navigation system of claim 32, the interface including a free-text search tool for searching the attributes.

34. The navigation system of claim 32, the interface including a free-text search tool for searching the values.

35. The navigation system of claim 32, further including a profile for each of the materials in the set of materials, the profile including descriptive information, the interface including a free-text search tool for searching the descriptive information in the profiles.

36. The navigation system of claim 32, the interface including access to the materials in the set of materials.

37. The navigation system of claim 32, the interface including a presentation of attribute-value pairs corresponding to the current navigation state.

38. The navigation state of claim 37, the presentation of attribute-value pairs corresponding to the current navigation state including user-selected attribute-value pairs and inferred attribute-value pairs, the interface including an indication of user-selected attribute-value pairs and of inferred attribute-value pairs.

39. The navigation system of claim 37, the presentation of attribute-value pairs corresponding to the current navigation state including only mutually incomparable attribute-value pairs.

40. The navigation system of claim 37, wherein the presentation organizes the attribute-value pairs corresponding to the current navigation state by attribute.

41. The navigation system of claim 37, wherein the presentation organizes the attribute-value pairs corresponding to the current navigation state by more general attribute-value pairs.

42. The navigation system of claim 32, the guided search tool including a presentation of navigation options for selection from the current navigation state, the options corresponding to transitions from the current navigation state.

43. The navigation system of claim 42, the navigation options including attribute-value pairs that are refinements of the attribute-value pairs corresponding to the current navigation state.

44. The navigation system of claim 43, wherein the options include a presentation of a set of lists of attribute-value pairs, each list corresponding to one of the attributes, some lists including attribute-value pairs that refine the attribute-value pairs corresponding to the current navigation state and some lists including attribute-value pairs that are incomparable to the attribute-value pairs corresponding to the current navigation state.

45. The navigation system of claim 42, the navigation options including disjunctive selection of attribute-value pairs.

46. The navigation system of claim 45, wherein the disjunctive selection options include a second attribute-value pair that corresponds to an attribute for which a first corresponding attribute-value has previously been selected, the second attribute-value pair being mutually incomparable to the first.

47. The navigation system of claim 42, the navigation options including negational selection of attribute-value pairs.

48. The navigation system of claim 42, wherein the presentation organizes the navigation options by attribute.

49. The navigation system of claim 42, wherein the presentation organizes the navigation options by more general attribute-value pairs.

50. The navigation system of claim 42, the navigation options including attribute-value pairs that are incomparable to the attribute-value pairs corresponding to the current navigation state.

51. The navigation system of claim 42, the navigation options including attribute-value pairs that are generalizations of the attribute-value pairs corresponding to the current navigation state.

52. The navigation system of claim 42, the navigation options including deselection of attribute-value pairs from the expression of attribute-value pairs corresponding to the current navigation state.

53. The navigation system of claim 42, the navigation options further including a link to an associated navigation state.

54. The navigation system of claim 53, wherein the associated navigation state is a generalization of the present navigation state.

56. The navigation system of claim 53, wherein the associated navigation state is a refinement of the present navigation state.

57. The navigation system of claim 53, wherein the link corresponds to a path of two or more transitions.

58. A method for enabling a user to browse a set of materials, a plurality of attributes characterizing the materials, a plurality of values describing the materials, each of the values having an association with at least one of the attributes, each association defining an attribute-value pair, some of the attribute-value pairs refining other of the attribute-value pairs, comprising the steps of: defining a plurality of navigation states, each navigation state corresponding to a particular expression of attribute-value pairs and to a particular subset of the materials; providing an interface, the interface providing a plurality of transitions, each transition providing a direct path between two of the navigation states, wherein each transition represents a change from the expression of attribute-value pairs corresponding to an originating navigation state to the expression of attribute-value pairs corresponding to a destination navigation state, wherein a series of one or more transitions provides a path between any two navigation states, there being more than one path between at least a first of the navigation states and a second of the navigation states; providing a display of transitions for selection; accepting a selection of a transition; obtaining a result navigation state associated with the selected transition; and providing a revised display of transitions in the result navigation state for selection.

59. The method of claim 58, wherein the step of providing a display of transitions for selection includes at least one transition that corresponds to conjunctive selection of an attribute-value pair.

60. The method of claim 58, wherein the step of providing a display of transitions for selection includes at least one transition that corresponds to disjunctive selection of an attribute-value pair.

61. The method of claim 58, wherein the step of providing a display of transitions for selection includes at least one transition that corresponds to negational selection of an attribute-value pair.

62. The method of claim 58, wherein the step of providing a display of transitions for selection includes providing a display of attribute-value pairs, wherein attribute-value pairs in the display can be related by Boolean operators.

63. The method of claim 62, wherein the step of providing a display includes organizing the attribute-value pairs by attribute.

64. The method of claim 58, wherein the step of accepting a selection of a transition includes accepting a selection of an attribute-value pair.

65. The method of claim 64, wherein the step of accepting the selection of an attribute-value pair triggers the step of obtaining a result navigation state.

66. The method of claim 58, wherein the step of providing a revised display of transitions includes providing a display of a revised set of attribute-value pairs, wherein attribute-value pairs in the revised set can be related by conjunction, disjunction, and negation.

67. The method of claim 66, wherein the step of providing a display of a revised set of attribute-value pairs includes providing a display of, for the previously selected attribute-value pair, a set of refining attribute-value pairs.

68. The method of claim 66, wherein the step of providing a display of a revised set of attribute-value pairs includes organizing the attribute-value pairs by more general attribute-value pairs.

69. The method of claim 66, wherein the step of providing a revised display includes adding an attribute previously unrepresented in the display.

70. The method of claim 58, further including the steps of accepting an additional selected attribute-value pair and obtaining an additional result navigation state associated with the selected attribute-value pair and the additional selected attribute-value pair.

71. The method of claim 70, wherein the additional selected attribute-value pair is associated with a different attribute than the selected attribute-value pair.

72. The method of claim 58, further including the step of displaying a list of previously selected attribute-value pairs.

73. The method of claim 72, further including the steps of accepting the deselection of a selected attribute-value pair and obtaining a revised result navigation state.

74. A method for enabling a user to browse a set of materials, a plurality of attributes characterizing the materials, a plurality of values describing the materials, each of the values having an association with at least one of the attributes, each association defining an attribute-value pair, some of the attribute-value pairs refining other of the attribute-value pairs, a plurality of navigation states, each navigation state corresponding to an expression of particular attribute-value pairs and to a particular subset of materials, an interface providing a plurality of transitions, each transition providing a direct path between two of the navigation states, wherein each transition represents a change from the expression of attribute-value pairs corresponding to an originating navigation state to the expression of attribute-value pairs corresponding to a destination navigation state, wherein a series of one or more transitions provides a path between any two navigation states, there being more than one path between at least a first of the navigation states and a second of the navigation states, comprising the steps of: displaying a free-text search box; accepting a search term entered in the free-text search box; determining the attribute-value pairs that match the search term; displaying the matching attribute-value pairs for the search term; accepting a selection of a matching attribute-value pair; and entering a navigation state having a corresponding expression that relates to the selection of the matching attribute-value pair.

75. A method for providing an information navigation system, the information navigation system including a set of materials, a plurality of attributes characterizing the materials, a plurality of values describing the materials, each of the values having an association with at least one of the attributes, each association defining an attribute-value pair, some of the attribute-value pairs refining other of the attribute-value pairs, a plurality of navigation states, each navigation state corresponding to a particular expression of attribute-value pairs and to a particular subset of the materials, comprising the steps of: computing the navigation states; storing the navigation states in a data structure; providing an interface to the information navigation system, the interface providing a plurality of transitions, each transition providing a direct path between two of the navigation states, wherein each transition represents a change from the expression of attribute-value pairs corresponding to an originating navigation state to the expression of attribute-value pairs corresponding to a destination navigation state, wherein a series of one or more transitions provides a path between any two navigation states, there being more than one path between at least a first of the navigation states and a second of the navigation states; accepting a query to the navigation system; and returning a responsive navigation state by retrieving a responsive precomputed navigation state or computing a responsive navigation state.

76. The method of claim 75, wherein the data structure is a graph data structure including nodes and edges between nodes, the nodes representing navigation states, the edges representing transitions.

77. A method of providing an information navigation system, the information navigation system including a set of materials, a plurality of attributes characterizing the materials, a plurality of values describing the materials, each of the values having an association with at least one of the attributes, each association defining an attribute-value pair, some of the attribute-value pairs refining other of the attribute-value pairs, a plurality of navigation states, each navigation state corresponding to a particular expression of attribute-value pairs and to a particular subset of the materials, comprising the steps of: generating a partial set of pre-computed navigation states; storing the partial set of pre-computed navigation states; providing an interface to the information navigation system, the interface providing a plurality of transitions, each transition providing a direct path between two of the navigation states, wherein each transition represents a change from the expression of attribute-value pairs corresponding to an originating navigation to the expression of attribute-value pairs corresponding to a destination navigation state, wherein a series of one or more transitions provides a path between any two navigation states, there being more than one path between at least a first of the navigation states and a second of the navigation states; accepting a query to the navigation system; and returning a responsive navigation state by retrieving a responsive precomputed navigation state or computing a responsive navigation state.

78. The method of claim 77, wherein the step of returning a responsive navigation state includes returning a navigation state computed by a plurality of servers acting in parallel.

79. The method of claim 78, further including the step of partitioning the materials among the plurality of servers.

80. The method of claim 79, wherein the plurality of servers are nested hierarchically.

81. The method of claim 80, wherein a root server of the plurality of servers acts as a master server and some of the plurality of servers act as slave servers, further including the steps of the master server distributing a request for a navigation state on to a plurality of slave servers, the slave servers computing navigation states for those requests and returning the results to the master server, and the master server combining the results from the slave servers to obtain a navigation state corresponding to the request.

82. The method of claim 81, wherein the combining step includes computing least common ancestors of attribute-value pairs corresponding to the navigation states returned by the slave servers.

83. A method of providing an information navigation system, the information navigation system including a set of materials, a plurality of attributes characterizing the materials, a plurality of values describing the materials, each of the values having an association with at least one of the attributes, each association defining an attribute-value pair, some of the attribute-value pairs refining other of the attribute-value pairs, a plurality of navigation states, each navigation state corresponding to a particular expression of attribute-value pairs and to a particular subset of the materials, comprising the steps of: storing the navigation states in an implicit representation; providing an interface to the navigation system, the interface providing a plurality of transitions, each transition providing a direct path between two of the navigation states, wherein each transition represents a change from the expression of attribute-value pairs corresponding to an originating navigation state to the expression of attribute-value pairs corresponding to a destination navigation state, wherein a series of one or more transitions provides a path between any two navigation states, there being more than one path between at least a first of the navigation states and a second of the navigation states; accepting a query to the navigation system; generating a responsive navigation state from the implicit representation; and presenting the responsive navigation state.

84. The method of claim 83, wherein the step of generating a responsive navigation state includes computing a responsive navigation state using a plurality of servers acting in parallel.

85. The method of claim 84, further including the step of partitioning the materials among the plurality of servers.

86. The method of claim 85, wherein the plurality of servers are nested hierarchically.

87. The method of claim 86, wherein a root server of the plurality of servers acts as a master server and some of the plurality of servers act as slave servers, further including the steps of the master server distributing a request for a navigation state on to a plurality of slave servers, the slave servers computing navigation states for those requests and returning the results to the master server, and the master server combining the results from the slave servers to obtain a navigation state corresponding to the request.

88. The method of claim 87, wherein the combining step includes the computation of least common ancestors of attribute-value pairs corresponding to the navigation states returned by the slave servers.

89. The method of claim 83, further including the step of storing the responsive navigation state in a cache.

90. A method for profiling a user of a navigation system, comprising the steps of: providing a navigation system, the navigation system including a set of materials, a plurality of attributes characterizing the materials, a plurality of values describing the materials, each of the values having an association with at least one of the attributes, each association defining an attribute-value pair, some of the attribute-value pairs refining other of the attribute-value pairs, a plurality of navigation states, each navigation state corresponding to a particular expression of attribute-value pairs and to a particular subset of the materials, and an interface providing a plurality of transitions, each transition providing a direct path between two of the navigation states, wherein each transition represent a change from the expression of attribute-value pairs corresponding to an originating navigation state, wherein a series of one or more transitions provides a path between any two navigation states, there being more than one path between at least a first of the navigation states and a second of the navigation states, the transitions being selectable by attribute-value pairs; accepting the user's selection of attribute-value pairs to navigate through the plurality of navigation states; and tracking the user's selection of attribute-value pairs.

91. The method of claim 90, further including the step of developing a user profile based on the user's selection of attribute-value pairs.

92. The method of claim 91, wherein the step of developing a user profile further includes the step of supplementing the user profile based on the attribute-value pairs corresponding to the navigation states visited by the user.

93. A computer program product, residing on a computer readable medium, for use in navigating a set of materials, in which the materials are characterized by a plurality of attributes, and the materials are described by a plurality of values, each of the values having an association with at least one of the attributes, each association defining an attribute-value pair, some of the attribute-value pairs refining other of the attribute value pairs, the computer program product comprising instructions for causing a computer to: define a plurality of navigation states, each navigation state corresponding to a particular expression of attribute-value pairs and to a particular subset of the materials; provide an interface, the interface providing a plurality of transitions, each transition providing a direct path between two of the navigation states, wherein each transition represents a change from the expression of attribute-value pairs corresponding to an originating navigation state to the expression of attribute-value pairs corresponding to a destination navigation state, wherein a series of one or more transitions provides a path between any two navigation states, there being more than one path between at least a first of the navigation states and a second of the navigation states; accept a selection of a transition; and provide a result navigation state associated with the selected transition.