Patent Grant
8856096
The present invention relates to methods, systems, and techniques for searching for information in a data set, and, in particular, to enhanced methods, systems, and techniques for syntactically indexing and performing syntactic searching of data sets using relationship queries to achieve greater search result accuracy.
Often times it is desirable to search large sets of data, such as collections of millions of documents, only some of which may pertain to the information being sought. In such instances it is difficult to either identify a subset of data to search or to search all data yet return only meaningful results. The techniques that have been traditionally applied to support searching large sets of data have fallen short of expectations, because they have not been able to achieve a high degree of accuracy of search results due to inherent limitations.
One common technique, implemented by traditional keyword search engines, matches words expected to found in a set of documents through pattern matching techniques. Thus, the more that is known in advance about the documents including their content, format, layout, etc., the better the search terms that can be provided to elicit a more accurate result. Data is searched and results are generated based on matching one or more words or terms that are designated as a query. Results such as documents are returned when they contain a word or term that matches all or a portion of one or more keywords that were submitted to the search engine as the query. Some keyword search engines additionally support the use of modifiers, operators, or a control language that specifies how the keywords should be combined when performing a search. For example, a query might specify a date filter to be used to filter the returned results. In many traditional keyword search engines, the results are returned ordered, based on the number of matches found within the data. For example, a keyword search against Internet websites typically returns a list of sites that contain one or more of the submitted keywords, with the sites with the most matches appearing at the top of the list. Accuracy of search results in these systems is thus presumed to be associated with frequency of occurrence.
One drawback to traditional keyword search engines is that they do not return data that fails to match the submitted keywords, even though the data may be relevant. For example, if a user is searching for information on what products a particular country imports, data that refers to the country as a “customer” instead of using the term “import” would be missed if the submitted query specifies “import” as one of the keywords, but doesn't specify the term “customer.” For example, a sentence such as “Argentina has been the main customer for Bolivia's natural gas” would be missed, because no forms of the word “import” are present in the sentence. Ideally, a user would be able to submit a query and receive back a set of results that were accurate based on the meaning of the query—not just on the specific keywords used in submitting in the query.
Natural language parsing provides technology that attempts to understand and identify the syntactical structure of a language. Natural language parsers (“NLPs”) have been used to identify the parts of speech of each term in a submitted sentence to support the use of sentences as natural language queries against data. However, systems that have used NLPs to parse and process queries against data, even when the data is highly structured, suffer from severe performance problems and extensive storage requirements.
Natural language parsing techniques have also been applied to extracting and indexing information from large corpora of documents. By their nature, such systems are incredibly inefficient in that they require excessive storage and intensive computer processing power. The ultimate challenge with such systems has been to find solutions to reduce these inefficiencies in order to create viable consumer products. Several systems have taken an approach to reducing inefficiencies by subsetting the amount of information that is extracted and subsequently retained as structured data (that is only extracting a portion of the available information). For example, NLPs have been used with Information Extraction engines that extract particular information from documents that follow predetermined grammar rules or when a predefined term or rule is recognized, hoping to capture and provide a structured view of potentially relevant information for the kind of searches that are expected on that particular corpus. Such systems typically identify text sentences in a document that follow a particular part-of-speech pattern or other patterns inherent in the document domain, such as “trigger” terms that are expected to appear when particular types of events are present. The trigger terms serve as “triggers” for detecting such events. Other systems may use other formulations for specified patterns to be recognized in the data set, such as predefined sets of events or other types of descriptions of events or relationships based upon predefined rules, templates, etc. that identify the information to be extracted. However, these techniques may fall short of being able to produce meaningful results when the documents do not follow the specified patterns or when the rules or templates are difficult to generate. The probability of a sentence falling into a class of predefined sentence templates or the probability of a phrase occurring literally is sometimes too low to produce the desired level of recall. Failure to account for semantic and syntactic variations across a data set, especially heterogeneous data sets, has led to inconsistent results in some situations.
It is often desirable to search large sets of unstructured data, such as collections of millions of documents, only some of which may pertain to the information being sought. Traditional search engines approach such data mining typically by offering interactive searches that match the data to one or more keywords (terms) using classical pattern matching or string matching techniques. At the other extreme, information extraction engines typically approach the unstructured data mining problem by extracting subsets of the data, based upon formulations of predefined rules, and then converting the extracted data into structured data that can be more easily searched. Typically, the extracted structured data is stored in a relational database management system and accessed by database languages and tools. Other techniques, such as those offered by Insightful Corporation's INFACT® search engine products, offer greater accuracy and truer information discovery tools, because they employ generalized syntactic indexing with the ability to interactively search for relationships and events in the data, including latent relationships, across the entire data set and not just upon predetermined extracted data that follows particular syntactic patterns. INFACT® search engine product's syntactic indexing and relationship searching uses natural language parsing techniques to grammatically analyze sentences to attempt to understand the meaning of sentences and then applies queries in a manner that takes into account the grammatical information to locate relationships in the data that correspond to the query. Some of these embodiments support a natural language query interface, which parses natural language queries in much the same manner as the underlying data, in addition to a streamlined relationship and event searching interface that focuses on retrieving information associated with particular grammatical roles. Other interfaces for relationship and event searching can be generated using an application programming interface (“API”). Insightful Corporation's syntactic searching techniques are described in detail in U.S. Provisional Application Nos. 60/312,385, 60/620,550, and 60/737,446 and U.S. application Ser. Nos. 10/007,299, 10/371,399, and 11/012,089, which are herein incorporated by reference in their entirety.
The syntactic indexing and relationship and event searching techniques describe therein extend the use of traditional keyword search engines to relationship and event searching of data sets. In summary, the syntactic and semantic information that is gleaned from an enhanced natural language parsing process is stored in an enhanced document index, for example, a form of a term-clause matrix, that is amenable to processing by the more efficient pattern (string) matching capabilities of keyword search engines. Accordingly, traditional keyword search engines, including existing or even off-the-shelf search engines, can be utilized to discover information by pattern (or string) matching the terms of a relationship query, which are inherently associated with syntactic and semantic information, against the syntactically and semantically annotated terms of sentence clauses (of documents) stored in the enhanced document index. As another benefit, the additional capabilities of such search engines, such as the availability of Boolean operations, and other filtering tools, are automatically extended to relationship and event searching.
Relationship and event searching, also described as “syntactic searching” in U.S. application Ser. Nos. 60/312,385, 10/007,299, 10/371,399, and 60/620,550 supports the ability to search a corpus of documents (or other objects) for places, people, or things as they relate to other places, people, or things, for example, through actions or events. Such relationships can be inferred or derived from the corpus based upon one or more “roles” that each term occupies in a clause, sentence, paragraph, document, or corpus. These roles may comprise grammatical roles, such as “subject,” “object,” “modifier,” or “verb;” or, these roles may comprise other types of syntactic or semantic information such as an entity type of “location,” “date,” “organization,” or “person,” etc. The role of a specified term or phrase (e.g., subject, object, verb, place, person, thing, action, or event, etc.) is used as an approximation of the meaning and significance of that term in the context of the sentence (or clause). In this way, a relationship or syntactic search engine attempts to “understand” the sentence when a query is applied to the corpus by determining whether the terms in sentences or clauses of the corpus are associated with the roles specified in the pending query. For example, if a user of the search engine desires to determine all events in which “Hillary Clinton” participated in as a speaker, then the user might specify a relationship query that instructs a search engine to locate all sentences/documents in which “Hillary Clinton” is a source entity and “speak” is an action. In response, the syntactic search engine will determine and return indicators to all sentences/clauses in which “Hillary Clinton” has the role of a subject and with some form of the word “speak” (e.g., speaking, spoke) or a similar word in the role of a verb.
For example,
In the example user interface shown in
As an alternative to specifying the relationship query using IQL, the user can use a graphical/form-based interface, termed here a “query generator,” for example, by selecting the “Try your own Fact Search” link 110. In response, a form such as that illustrated in
More details of the relationship query language and of an example user interface provided by Insightful Corporation's products are described in co-pending patent application Ser. No. 11/012,089.
Embodiments described herein provide enhancements to the methods, systems, and techniques for syntactically indexing and searching data sets to achieve more accurate search results with greater flexibility and efficiency than previously available. Example embodiments provide improvements to an enhanced Syntactic Query Engine (“SQE”) that parses, indexes, and stores a data set, as well as performs syntactic searching in response to queries submitted against the data set. In one embodiment, the SQE includes, among other components, a data set repository and an Enhanced Natural Language Parser (“ENLP”). The ENLP parses each object in the data set (typically a document) and transforms it into a canonical form (also termed a normalized form) that can be searched efficiently using techniques of the present invention. To perform this transformation, the ENLP determines the syntactic structure of the data by parsing (or decomposing) each data object into syntactic units, determines the grammatical roles and relationships of the syntactic units, associates recognized entity types if configured to do so, and represents these relationships in a normalized form. The normalized data are then stored and/or indexed as appropriate for efficient searching. Thus, the SQE can provide a unified knowledge representation for both structured and unstructured data.
In example embodiments of the SQE described herein, the normalized data, including the grammatical role and other tag information that can be used to discover or explore relationships, are integrated into enhanced versions of document indexes that are typically used by traditional keyword search engines to index the terms of each document in a corpus. A traditional keyword search engine can then search the enhanced indexing information that is stored in these document indexes for matching relationships in the same manner the search engine searches for keywords. That is, the search engine looks for pattern/string matches to terms associated with the desired tag information as specified (explicitly or implicitly) in a query. A detailed description of an example SQE that provides enhanced indexing information according to these techniques is described in U.S. patent application Ser. No. 11/012,089. In one such example system, the SQE stores the relationship information that is extracted during the parsing and data object transformation process (the normalized data) in an annotated “term-clause matrix,” which stores the terms of each clause along with “tagged terms,” which include the syntactic and semantic information that embodies relationship information. Other example embodiments may provide different levels of organizing the enhanced indexing information, such as an annotated “term-sentence matrix” or an annotated “term-document matrix.” Other variations of storage organization are possible, including that each matrix may be comprised of a plurality of other data structures or matrices.
The integration of the enhanced indexing information into traditional keyword search engine type document indexes (for example, an inverted index) is what supports the use of standard keyword search techniques to find a new type of document information—that is, relationship information—easily and quickly. An end user, such as a researcher, can pose simple Boolean style queries to the SQE yielding results that are based upon an approximation of the meaning of the indexed data objects. Because traditional search engines do not pay attention to the actual contents of the indexed information (they just perform string matching or pattern matching operations without regard to the meaning of the content), the SQE can store all kinds of relationship information in the indexed information and use a keyword search engine to quickly retrieve it. In addition, standard document searches can be combined with more specialized relationship searches to generate many types of results.
The SQE processes each query by translating or transforming the query into component keyword searches that can be performed against the indexed data set using, for example, an “off-the-shelf” or existing keyword search engine. These searches are referred to herein for ease of description as keyword searches, keyword-style searches, or pattern matching or string matching searches, to emphasize their ability to match relationship information the same way search terms can be string- or pattern-matched against a data set using a keyword search engine. The SQE then combines the results from each keyword-style search into a cohesive whole that is presented to the user.
Specifically, in step 401, the SQE receives a data set, for example, a set of documents. The documents may be received electronically, scanned in, or communicated by any reasonable means. In step 402, the SQE preprocesses the data set to ensure a consistent data format. In step 403, the SQE parses the data set, identifying entity type tags and the syntax and grammatical roles of terms within the data set as appropriate to the configured parsing level. For the purpose of extending keyword searching to syntactically and semantically annotated data, parsing sufficient to determine at least the subject, object, and verb of each clause is desirable to perform syntactic searches in relationship and event queries. However, as described elsewhere in co-pending patent applications, subsets of the capabilities of the SQE could be provided in trade for shorter corpus ingestion times if full syntactic searching is not desired. In step 404, the SQE transforms the each parsed clause (or sentence) into normalized data by applying various linguistic normalizations and transformations to map complex linguistic constructs into equivalent structures. Linguistic normalizations include lexical normalizations (e.g., synonyms), syntactic normalizations (e.g., verbalization), and semantic normalizations (e.g., reducing different sentence styles to a standard form). These heuristics and rules are applied when ingesting documents and influence how well the stored sentences eventually will be “understood” by the system. In step 405, the SQE stores the parsed and transformed sentences in a data set repository. As described above, when the SQE is used with a keyword search engine, the normalized data is stored in (used to populate) an enhanced document index such as the term-clause matrix.
After storing the data set, the SQE can process relationship queries against the data set. In step 406, the SQE receives a relationship query, for example, through a user interface such as that shown in
Although the techniques are described primarily with reference to text-based languages and collections of documents, similar techniques may be applied to any collection of terms, phrases, units, images, or other objects that can be represented in syntactical units and that follow a grammar that defines and assigns roles to the syntactical units, even if the data object may not traditionally be thought of in that fashion. Examples include written or spoken languages, for example, English or French, computer programming languages, graphical images, bitmaps, music, video data, and audio data. Sentences that comprise multiple words are only one example of a phrase or collection of terms that can be analyzed, indexed, and searched using the techniques described herein. One can modify the structures and program flow exemplified herein to account for differences in types of data being indexed and retrieved. The concepts and techniques described are applicable to any environment where the keyword style searching is contemplated.
Also, although certain terms are used primarily herein, one skilled in the art will recognize that other terms could be used interchangeably to yield equivalent embodiments and examples. In addition, terms may have alternate spellings which may or may not be explicitly mentioned, and one skilled in the art will recognize that all such variations of terms are intended to be included. For example, the terms “matrix” and “index” are used interchangeably and are not meant to imply a particular storage implementation. Also, a document may be a single term, clause, sentence, or paragraph or a collection of one or more such objects. Also, when referring to various data, aspects, or elements in the alternative, the term “or” is used in its plain English sense, unless otherwise specified, to mean one or more of the listed alternatives.
As additional examples, the term “query” is used herein to include any form of specifying a desired relationship query, including a specialized syntax for entering query information, a menu driven interface, a graphical interface, a natural language query, batch query processing, or any other input (including API function calls) that can be transformed into a Boolean expression of terms and annotated terms. Annotated terms are terms associated with syntactic or semantic tag information, and are equivalently referred to as “tagged terms.” Semantic tags include, for example, indicators to a particular node or path in an ontology or other classification hierarchy. “Entity tags” are examples of one type of semantic tag that points, for example, to a type of ENTITY node in an ontology. In addition, although the description is oriented towards parsing and maintaining information at the clause level, it is to be understood that the SQE is able to parse and maintain information in larger units, such as sentences, paragraphs, sections, chapters, documents, etc., and the routines and data structures are modified accordingly. Thus, for ease of description, the techniques are described as they are applied to a term-clause matrix. One can equivalently apply these techniques to a term-sentence matrix or a term-document matrix.
In the following description, numerous specific details are set forth, such as data formats and code sequences, etc., in order to provide a thorough understanding of the described techniques. The embodiments described also can be practiced without some of the specific details described herein, or with other specific details, such as changes with respect to the ordering of the code flow, different code flows, etc. Thus, the scope of the techniques and/or functions described are not limited by the particular order, selection, or decomposition of steps described with reference to any particular routine.
As described in co-pending U.S. patent application Ser. No. 11/012,089, published as U.S. Patent Publication No. 2005/0267871, many different interfaces are possible for specifying relationship queries, including form based (template-style) interfaces, relationship search strings, visual GUI drag and drop interfaces, a query language called IQL, an application programming interface (“API”) for creating application access to relationship query information or for creating customized (also alternative) user interfaces etc. Appendix A, incorporated herein by reference in its entirety, describes several such interfaces and enhancements made to the SQE example embodiments that were previously described in U.S. patent application Ser. No. 11/012,089. Many of these enhancements are exemplified in Insightful Corporation's INFACT® System Release 3.1 search engine product.
Embodiments of the interfaces described in Appendix A include various enhancements to the extended keyword searching methods, systems, and techniques described herein, including, amongst other features, support for a new context operator; improved metadata searching performance, support for ontology (entity tag) searches at the document search level, an improved interface for exporting relationship data, data sorting across an entire result, and automatically generated navigation tips.
Context Operator
Embodiments of the enhanced SQE provide a context operator for determining in a relationship search whether the prescribed relation might be satisfied by searching surrounding sentences (clauses, or any other granularity desired, etc.). In some embodiments, the number of surrounding sentences/clauses searched is a predetermined number “n.” In other embodiments, it is contemplated that this number can be set dynamically, even from within the query itself. Combinations and permutations (such as permitting entry of any number less than “n”) are also possible.
In the user interface described in Appendix A, the context operator (also termed a context constraint) can be specified in a query as:
The context operator is particularly beneficial to catch related text when two separate clauses (sentences) that have no inter-clause connections contain information that is related to what the researcher is searching for. For example, the sentences:
In order to support context operators, the data structures used to stored the normalized (extended keyword) data have been modified. In summary, the context, stored as a “bag of words,” is stored in the term-clause index for fast access to context. In addition, since context information is now searched for surrounding sentences, the ontology paths present in the surrounding sentences (all entities) are stored for quick comparisons. See the “context” and “context_ont_path” fields in the modified term-clause index described in Table 1 below. The term-clause index and other relevant data structures were described in detail in co-pending U.S. patent application Ser. No. 11/012,089.
Metadata Constraint Performance
Embodiments of an enhanced SQE provide improved techniques for efficiently handling meta data used in relationship searches. Metadata filtering allows one to constrain a search based on document level metadata constraints. Metadata may or may not be available for a particular document corpus. (A user can determine whether such metadata is available and the various types of information available by selecting a link to the “Corpus” page as shown in Appendix A and available through some embodiments of Insightful Corporation's SQE.) For example, the following (equivalent) queries, search for all documents having at least one sentence indicating Clinton in a “visits” relationship to China, where the author of the document is “John Smith.”
In the enhanced SQE, the data structures used to stored the normalized (extended keyword) data have been modified to more efficiently store metadata, and specific metadata such as a document date, for fast retrieval. In one example embodiment, the term-document index is modified to contain additional fields such as “docdate” and “author.” In some embodiments, all of the metadata is available as an inverted index as well. See also, Table 2 described below.
Ontology Specifications in Document Level Searches
Embodiments of the SQE have been enhanced to support ontology path specifications in document level searches, regardless of whether the search is performed as part of a document (i.e., keyword) search or relationship search. For example, one can search for all documents which contain a specific entity type or ontology path the same way one specifies a keyword search. For example, the general document search query:
One can also use relationship searches with document constraints (also called document specifications) to search for documents with a specific keyword in it. For example, the relationship search exemplified in Query 2 above can be modified to find documents where a “DC-8” is described somewhere in the document (and not just in the surrounding sentences) as:
The data structures used to stored the normalized (extended keyword) data have been modified to efficiently support document level searches for ontology path information by including such information in specific fields in, for example, the term-document index. In one example embodiment, all ontology paths (and sub-paths) of each available entity are stored in the term-document index. See Tables 1-3 below.
Note as well that the ontologies themselves can be stored in inverted indices for easy searching and access by the SQE. Specifically, ontology entity data can be stored in keyword search engine compatible indices such as those illustrated in Appendix D, which is incorporated herein by reference in its entirety.
Data Sorting Across Entire Result
Another feature of the enhanced SQE is that it can provide search results organized by data across the whole result, and not just on a per page basis.
When the user selects the retrieve by date field 501, the results are displayed organized by date.
Exporting Relationship Data
The ability to share relationship data with other applications is useful, especially when there are large amounts of data to be analyzed and statistical tools could be applied to model the data. The enhanced SQE provides an improved interface for exporting relationship query results in a (character delimited) format that can be input to analytic tools such as those provided by S-PLUS or Excel.
When the data is exported, it is typically stored as a tab-delimited file. Different delimiters can be incorporated through configuration parameters.
Automatic Guided Navigation Tips
Embodiments of syntactic query engines have been enhanced to incorporate automated tips for users to increase the ability of a user to specify a desired search. Based upon the inverted indexes used to represent the corpus, the ontologies used in the system and other rules (such as popularity of term choice, etc.) these tips act as “suggestions” to the user to specify with greater particularity a possibly desired search.
Behind the scenes, when a query (e.g., either a document level search such as one or more keywords or a relationship search, including portions of a or an entire relationship query using IQL) is entered, the SQE attempts to determine what relationship searches might be desired based upon a set of rules parsed and evaluated by the SQE. Then, the SQE attempts to run some number of these potential searches in the background (up to all such possibilities). Thereafter, depending upon the particular tip user interface, some indication of the navigation tips and/or search results of the tips are presented to the user. Once displayed, the user can then choose one or more of these preformed searches (by selecting the navigation tips) to quickly see results.
The query may take the form of either one or more keywords, an IQL expression, or components of an IQL expressed query. For example, a user may enter the keywords “Japan China” to try to determine relationship and event information relating to both countries, and the SQE will respond with appropriate navigation tips to assist the user to discover more information. As another example, the user might type in a partial IQL expression such as “China< >*˜Japan”, and the SQE will attempt to recognize the input as particular components of IQL such as a source entity and a context operator expression.
The rules used to determine the navigation tips are specified, typically in a configuration file, by an administrator when the searching system is configured (or at other times). Each rule specifies some sort of template or specification for determining what relationship query to execute based upon the recognition of particular input. For example, a rule might specify that when a single entity is recognized as input, a rule that treats it as a source entity and looks for related organizations is fired. Such a rule might be expressed, for example, by
Different heuristics and/or rules can be used to determine which potential search alternatives to suggest, the order they are suggested, etc. For example, the number of results returned, the popularity of a particular search (e.g., how many times it has been executed against that corpus), or other measurements of value or interest can be incorporated. Note that other types of rules and an entire programming or scripting language for running and combining potential queries can be defined for use by the SQE.
In one embodiment, the searches are run to validate whether there are any results for a particular tip before showing them to a user. In another embodiment, the searches are run (typically in the background and even in parallel), the results cached, and indications to some number of them presented to the user. In some embodiments, the cached tip results are first consulted to speed up response time. The cached tip results can also be useful to return tips to a user when an SQE facility is under heavy load and desires to suspend tip rule processing. Also, depending upon the architecture and infrastructure used to evaluate and perform the tip searches (which execute typically in the background), the SQE can perform load balancing, parallel processing, etc.
According to one example embodiment, which can be accessed via url “www.globalsecurity.org,” the SQE provides navigation tips to a user based upon whether it can recognize the specification of a (single) entity that is part of a configured ontology. More specifically, the SQE looks at the user specified entity and determines whether there are one of more ontology paths that include the user-specified term as a “leaf” node (an entity). If not, no tips are suggested. If so, then the SQE runs one or more appropriate relationship searches that corresponds to each possible ontology path, in the background using rules that define which searches to run in order to “validate” the searches against the particular corpus. In some instances, only searches that result in matches are then presented to the user.
In some implementations, a configuration file containing rules is supplied (for example by a system administrator) and a mapping (or other means for specifying a set of rules) is stored between entities to be recognized and potential searches to execute. The configuration file may be parsed and the mappings stored when the system comes up or at other times. Appendix B contains one example of a set of ontology path specifications and mappings to corresponding queries as specified for such a configuration file. Also, depending upon the implementation, the SQE may implement different orders or precedence for applying recognized rules. In one case, the rules are applied in the order that they are encountered in the configuration file, so if an entered query matches multiple rules in the file, the first encountered set of rules will apply. In another case, all of the matched rules are applied and all of the potential searches performed in the order specified. Other orders of precedence can be similarly incorporated.
To use a tip, the user selects one of the tips from the tip link area 1402.
Alternatively, suppose that the user wanted to see who President Bush chose as a replacement Secretary of Defense—the user knew the person came from another government agency, but couldn't remember which agency.
Note as well that different user interfaces can be used to display the tips. In one such interface, a menu of possible search specifications is provided. In another interface, links are presented that can be selected to show “fast search” results.
According to a second example embodiment, which can be accessed via url “books.infact.com,” the SQE provides navigation tips to a user based upon whether it can recognize one or more keywords (as in a document level search) or the specification of (any part of) a relationship query using IQL. More specifically, the SQE looks at the user specified entry and determines whether there are rules that map additional relationship queries to the keywords and/or recognized (portion of a) relationship query. In the case of keyword input, the SQE first parses the input (using the extended natural language parser of the SQE) to transform the input into a relationship query structure (an enhanced or normalized data structure of the SQE) that can be compared with the rules. Appendix C contains one example of a configuration file having a set of mappings from a relationship query structure (which may contain a single entity or action) to corresponding additional queries. Note that the configuration file of Appendix C specifies rules using XML tag definitions; however, as indicated above, other languages, specifications, and mappings could be used. Based upon evaluating the corresponding rules, the SQE then display the potential other relationship queries that might be of interest to the user.
Note as well that, in some embodiments, synonyms for tips are collapsed and other normalizations of the tips are performed. For example, if during the tip construction process the SQE determines that relationship queries yield valid results against the corpus for “Bill Clinton,” “William Clinton,” and “President Clinton,” then the tip subsystem may display a single deep tip “Clinton.” Similarly, synonyms such as “UK,” “Great Britain,” and “United Kingdom” will be coalesced into one tip. Other normalizations, such as upper and lower case transformations may also be performed.
In the books corpus example, a user can also peruse individual books and quickly get more information using relationship searching and navigation tips.
When the user selects an entity type in the book index area 2003, the various instances of entities that appear in the book are displayed.
There are any number of techniques an SQE can use to provide the interfaces and navigation tips described in
Depending upon the architectural implementation, including the various parallel processing and load balancing techniques and components available, one or more components may actually be invoked to perform tip processing. In one implementation (not shown) a tip searcher manager is invoked by the tip engine to properly load balance and distribute requests to some number of tip searcher components. In addition, as will be described below, a tip searcher may further distribute execution of various relationship queries so that they can be performed in parallel. Other architectures are also possible.
As an example, the rule file may specify a more general relationship query that involves an entity type or an action type. Once the rule is executed and the general relationship query evaluated, the tip searcher (or its components) can evaluate (weigh, score, filter, etc.) the particular entities/actions discovered as results from evaluating the general relationship query and then execute specific relationship queries for those entities/actions. The specific relationship queries that correspond to entities or actions (as opposed to types of entities or actions) can be used to provide the deep tips described in the screen displays above.
Assuming, for example, that the deep tips are determined asynchronously, at some point the results from executing the general relationship query are made available and the tip searcher is notified. In step 2608, the tip searcher determines from these results whether the general tip has generated results that satisfy the tip system rules and heuristics, and if so, presents them to the user. For example, if a (general) tip generates less than some threshold number of results, the tip system may not present a general tip. In other embodiments, the tip searcher always presents the general tip as it corresponds to an available rule. The tip searcher then returns to step 2605 to process the next rule, if any. If not, the tip searcher process is done (until invoked again).
Specifically, in step 2701, the routine receives the general relationship query to execute and evaluate. (Note that, in other embodiments, the routine might just receive the rule and be responsible for determining the corresponding relationship query.) In step 2702, the routine determines whether a cached result of the general relationship query is already available and, if so, retrieves it (step 2703); otherwise in step 2603 executes the general relationship query that corresponds to the rule to determine a result. In step 2604, the routine returns the results to the tip searcher (or stores them and notifies the tip searcher) for presentation of a “general” tip, and in steps 2705-2707 processes and evaluates the specific entities and/or actions found in the results.
More specifically, in step 2705 the routine determines whether there is another entity/action to process and, if so, continues in step 2706, other continues in step 2708. In step 2706, the routine processes the next entity/action. In step 2707, the routine evaluates the search results that correspond to this entity/action to determine whether they meet the criteria associated with presenting a deep tip. This evaluation, as mentioned, may rank all of the results to determine the most frequently appearing entities/actions and/or may determine whether the number of times an entity/verb appears reaches or surpasses some threshold. Other evaluation criteria may of course be incorporated. In any case, in step 2707 once the routine determines whether an entity/action qualifies for a deep tip, it determines a corresponding relationship query using that entity/action and stores the corresponding RQ result. In step 2708 the routine returns relationship queries that correspond to the stored (or “n” number of them) results and causes them to be presented as deep tips, and then finishes processing. Note that the number of results stored/returned may be a predetermined number, a settable number, a default, etc. In addition, a preference variable may be available to change the number while the SQE is running.
Enhanced SQE Architecture
Details of the operation of an SQE are provided in co-pending U.S. patent application Ser. No. 11/012,089. In summary, the enhanced SQE 2801 receives as input a data set 2802 to be indexed and stored. The Data Set Preprocessor 2803 prepares the data set for parsing by assigning a Document ID to each document that is part of the received data set (and sentence and clause IDs as appropriate), performing OCR processing on any non-textual entities that are part of the received data set, and formatting each sentence according to the Enhanced Natural Language Parser format requirements. The Enhanced Natural Language Parser (“ENLP”) 2804 parses the data set, identifying for each sentence, a set of terms, each term's tags, including potentially part of speech and associated grammatical role tags and any associated entity tags or ontology path information, and transforms this data into normalized data. The Data Set Indexer 2807 indexes and stores the normalized data output from the ENLP in the data set repository 2808. The data set repository 2808 represents whatever type of storage along with the techniques used to store the enhanced document indexes. For example, the indexes may be stored as sparse matrix data structures, flat files, etc. and reflect whatever format corresponds to the input format expected by the keyword search engine. After a data set (or a portion of a data set) is indexed, a Relationship Query 2809 may be submitted to the enhanced SQE 2801 for processing. The Relationship Query Processor 2810 prepares the query for parsing, for example by splitting the Relationship Query 2809 into sub-queries that are executable directly by the Keyword Search Engine 2811. As explained elsewhere, a Relationship Query 2809 is typically comprised of a syntactic search along with optional constraint expressions. Also, different system configuration parameters can be defined that influence and instruct the SQE to search using particular rules, for example, to include synonyms, related verbs, etc. Thus, the Relationship Query Processor 2810 is responsible for augmenting the specified Relationship Query 2809 in accordance with the current SQE configured parameters. To do so, the Relationship Query Processor 2810 may access the ontology information which may be stored in Data Set Repository 2808 or some other data repository. The Relationship Query Processor 2810 splits up the query into a set of Boolean expression searches that are executed by the Keyword Search engine 2811 and causes the searches to be executed. The Relationship Query Processor 2810 then receives the result of each search from the Keyword Search Engine 2811 and combines them as indicated in the original Relationship Query 2809 (for example, using Boolean operators). Note that the Relationship Query Processor 2810 may be comprised of multiple subcomponents that each execute a portion of the work required to preprocess and execute a relationship query and combine the results for presentation. The results (in portions or as required) are sent to the User Interface/API component 2813 to produce the overall Query Result 2812. The User Interface Component 2813 may interface to a user in a manner similar to that shown in the display screens of
The functions of data set processing (data object ingestion) and relationship query processing can be practiced in any number of centralized and/or distributed configurations of client—server systems. Parallel processing techniques can be applied in performing indexing and query processing to substantial increase throughput and responsiveness. Representative configurations and architectures are described in detail in co-pending U.S. patent application Ser. No. 11/012,089; however, a variety of other configurations could equivalently perform the functions and capabilities identified herein.
As mentioned, the data structures used to store relationship data have been modified to support the enhancements described herein. Specifically, the data set indexer 2807 in
Table 1 below conceptually illustrates the modifications made to the information that is maintained in an example term-clause index to support the enhanced SQE.
Table 2 below conceptually illustrates the modifications made to the information that is maintained in an example sentence index to support the enhanced SQE.
Table 3 below conceptually illustrates the modifications made to the information that is maintained in an example document index to support the enhanced SQE.
All of the above U.S. patents, U.S. patent application publications, U.S. patent applications, foreign patents, foreign patent applications and non-patent publications referred to in this specification and/or listed in the Application Data Sheet, including but not limited to U.S. patent application Ser. No. 10/007,299, filed on Nov. 8, 2001, entitled “Method and System for Enhanced Data Searching,” and published as U.S. Patent Publication No. 2004/0221235; U.S. patent application Ser. No. 10/371,399, filed on Feb. 19, 2003, entitled “Method and System for Enhanced Data Searching”, and published as U.S. Patent Publication No. 2003/0233224; U.S. patent application Ser. No. 11/012,089, filed on Dec. 13, 2004, entitled “Method and System for Extending Keyword Searching to Syntactically and Semantically Annotated Data,” and published as U.S. Patent Publication No. 2005/0267871; U.S. Provisional Application No. 60/312,385, filed on Aug. 14, 2001, entitled “Methods and Systems for Enhanced Indexing and Syntactic Searching”; U.S. Provisional Application No. 60/620,550, filed on Oct. 20, 2004, entitled “Method and System for Relationship Searching”; and U.S. Provisional Application No. 60/737,446, filed on Nov. 16, 2005, entitled “Method and System for Extending Keyword Searching to Syntactically and Semantically Annotated Data” are incorporated herein by reference, in their entirety.
From the foregoing it will be appreciated that, although specific embodiments have been described herein for purposes of illustration, various modifications may be made without deviating from the spirit and scope of the present disclosure. In addition, those skilled in the art will understand how to make changes and modifications to the methods and systems described to meet their specific requirements or conditions. The methods and systems discussed herein are applicable to differing protocols, communication media (optical, wireless, cable, etc.) and devices (such as wireless handsets, electronic organizers, personal digital assistants, portable email machines, game machines, pagers, navigation devices such as GPS receivers, etc.). For example, the methods and systems described herein can be applied to any type of search tool or indexing of a data set, and not just the enhanced SQE described. In addition, the techniques described may be applied to other types of methods and systems where large data sets must be efficiently reviewed. For example, these techniques may be applied to Internet search tools implemented on a PDA, web-enabled cellular phones, or embedded in other devices. Furthermore, the data sets may comprise data in any language or in any combination of languages. In addition, the user interface and API components described may be implemented to effectively support wireless and handheld devices, for example, PDAs, and other similar devices, with limited screen real estate. These and other changes may be made in light of the above-detailed description.
This invention was made with government support under Contract No. DAAH01-00-C-R168, awarded by Defense Advanced Research Project Agency; Contract Nos. W74Z8H-04-P-0104 and W74V8H-05-C-0016, awarded by the Office of the Secretary of Defense, U.S. Army; and Contract No. FA8650-05-C-6500, awarded by the Office of the Secretary of Defense, U.S. Air Force. The government has or may have certain rights in this invention.
| Number | Name | Date | Kind |
|---|---|---|---|
| 4839853 | Deerwester et al. | Jun 1989 | A |
| 5301109 | Landauer et al. | Apr 1994 | A |
| 5317507 | Gallant | May 1994 | A |
| 5325298 | Gallant | Jun 1994 | A |
| 5331556 | Black, Jr. et al. | Jul 1994 | A |
| 5377103 | Lamberti et al. | Dec 1994 | A |
| 5619709 | Caid et al. | Apr 1997 | A |
| 5778362 | Deerwester | Jul 1998 | A |
| 5794050 | Dahlgren et al. | Aug 1998 | A |
| 5794178 | Caid et al. | Aug 1998 | A |
| 5799268 | Boguraev | Aug 1998 | A |
| 5857179 | Vaithyanathan et al. | Jan 1999 | A |
| 5884302 | Ho | Mar 1999 | A |
| 5933822 | Braden-Harder et al. | Aug 1999 | A |
| 5950189 | Cohen et al. | Sep 1999 | A |
| 6006221 | Liddy et al. | Dec 1999 | A |
| 6006225 | Bowman et al. | Dec 1999 | A |
| 6026388 | Liddy et al. | Feb 2000 | A |
| 6061675 | Wical | May 2000 | A |
| 6064951 | Park et al. | May 2000 | A |
| 6122647 | Horowitz et al. | Sep 2000 | A |
| 6185550 | Snow et al. | Feb 2001 | B1 |
| 6192360 | Dumais et al. | Feb 2001 | B1 |
| 6202064 | Julliard | Mar 2001 | B1 |
| 6246977 | Messerly et al. | Jun 2001 | B1 |
| 6363373 | Steinkraus | Mar 2002 | B1 |
| 6411962 | Kupiec | Jun 2002 | B1 |
| 6460029 | Fries et al. | Oct 2002 | B1 |
| 6484162 | Edlund et al. | Nov 2002 | B1 |
| 6510406 | Marchisio | Jan 2003 | B1 |
| 6584464 | Warthen | Jun 2003 | B1 |
| 6601026 | Appelt et al. | Jul 2003 | B2 |
| 6728707 | Wakefield et al. | Apr 2004 | B1 |
| 6732097 | Wakefield et al. | May 2004 | B1 |
| 6732098 | Wakefield et al. | May 2004 | B1 |
| 6738765 | Wakefield et al. | May 2004 | B1 |
| 6741988 | Wakefield et al. | May 2004 | B1 |
| 6745161 | Arnold et al. | Jun 2004 | B1 |
| 6757646 | Marchisio | Jun 2004 | B2 |
| 6859800 | Roche et al. | Feb 2005 | B1 |
| 6862710 | Marchisio | Mar 2005 | B1 |
| 6910003 | Arnold et al. | Jun 2005 | B1 |
| 6996575 | Cox et al. | Feb 2006 | B2 |
| 7051017 | Marchisio | May 2006 | B2 |
| 7054854 | Hattori et al. | May 2006 | B1 |
| 7171349 | Wakefield et al. | Jan 2007 | B1 |
| 7283951 | Marchisio et al. | Oct 2007 | B2 |
| 7398201 | Marchisio et al. | Jul 2008 | B2 |
| 7403938 | Harrison et al. | Jul 2008 | B2 |
| 7526425 | Marchisio et al. | Apr 2009 | B2 |
| 20020007267 | Batchilo et al. | Jan 2002 | A1 |
| 20020010574 | Tsourikov et al. | Jan 2002 | A1 |
| 20020059161 | Li | May 2002 | A1 |
| 20020078041 | Wu | Jun 2002 | A1 |
| 20020078045 | Dutta | Jun 2002 | A1 |
| 20020091671 | Prokoph | Jul 2002 | A1 |
| 20030004716 | Haigh et al. | Jan 2003 | A1 |
| 20030101182 | Govrin et al. | May 2003 | A1 |
| 20030115065 | Kakivaya et al. | Jun 2003 | A1 |
| 20030115191 | Copperman et al. | Jun 2003 | A1 |
| 20030233224 | Marchisio et al. | Dec 2003 | A1 |
| 20040064447 | Simske et al. | Apr 2004 | A1 |
| 20040103090 | Dogl et al. | May 2004 | A1 |
| 20040125877 | Chang et al. | Jul 2004 | A1 |
| 20040167870 | Wakefield et al. | Aug 2004 | A1 |
| 20040167883 | Wakefield et al. | Aug 2004 | A1 |
| 20040167884 | Wakefield et al. | Aug 2004 | A1 |
| 20040167885 | Wakefield et al. | Aug 2004 | A1 |
| 20040167886 | Wakefield et al. | Aug 2004 | A1 |
| 20040167887 | Wakefield et al. | Aug 2004 | A1 |
| 20040167907 | Wakefield et al. | Aug 2004 | A1 |
| 20040167908 | Wakefield et al. | Aug 2004 | A1 |
| 20040167909 | Wakefield et al. | Aug 2004 | A1 |
| 20040167910 | Wakefield et al. | Aug 2004 | A1 |
| 20040167911 | Wakefield et al. | Aug 2004 | A1 |
| 20040221235 | Marchisio et al. | Nov 2004 | A1 |
| 20040243388 | Corman et al. | Dec 2004 | A1 |
| 20050027704 | Hammond et al. | Feb 2005 | A1 |
| 20050108001 | Aarskog | May 2005 | A1 |
| 20050138018 | Sakai et al. | Jun 2005 | A1 |
| 20050144064 | Calabria et al. | Jun 2005 | A1 |
| 20050149494 | Lindh et al. | Jul 2005 | A1 |
| 20050177805 | Lynch et al. | Aug 2005 | A1 |
| 20050197828 | McConnell et al. | Sep 2005 | A1 |
| 20050210000 | Michard | Sep 2005 | A1 |
| 20050216443 | Morton et al. | Sep 2005 | A1 |
| 20050234879 | Zeng et al. | Oct 2005 | A1 |
| 20050267871 | Marchisio et al. | Dec 2005 | A1 |
| 20060149734 | Egnor et al. | Jul 2006 | A1 |
| 20080005651 | Grefenstette et al. | Jan 2008 | A1 |
| 20080059456 | Chowdhury et al. | Mar 2008 | A1 |
| 20080288456 | Omoigui | Nov 2008 | A1 |
| 20090019020 | Dhillon et al. | Jan 2009 | A1 |
| 20090076886 | Dulitz et al. | Mar 2009 | A1 |
| 20090228439 | Manolescu et al. | Sep 2009 | A1 |
| 20130124510 | Guha | May 2013 | A1 |
| Number | Date | Country |
|---|---|---|
| 0 280 866 | Sep 1988 | EP |
| 0 597 630 | Jul 2002 | EP |
| 20080111822 | Dec 2008 | KR |
| WO 0014651 | Mar 2000 | WO |
| WO 0057302 | Sep 2000 | WO |
| WO 0122280 | Mar 2001 | WO |
| WO 0180177 | Oct 2001 | WO |
| WO 0227536 | Apr 2002 | WO |
| WO 0233583 | Apr 2002 | WO |
| WO 2004053645 | Jun 2004 | WO |
| Entry |
|---|
| Abraham, “FoXQ—Xquery by Forms,” Human Centric Computing Languages and Environments, Proceedings 2003 IEEE Symposium, Oct. 28-31, 2003, Piscataway, New Jersey, pp. 289-290. |
| Cass, “A Fountain of Knowledge,” IEEE Spectrum Online, URL: http://www.spectrum.ieee.org/WEBONLY/publicfeature/jan04/0104comp1.html, download date Feb. 4, 2004, 8 pages. |
| Feldman et al., “Text Mining at the Term Level,” Proc. of the 2nd European Symposium on Principles of Data Mining and Knowledge Discover, Nantes, France, 1998. |
| Ilyas et al., “A Conceptual Architecture for Semantic Search Engine,” IEEE, INMIC, 2004, pp. 605-610. |
| Jayapandian et al., “Automating the Design and Construction of Query Forms,” Data Engineering, Proceedings of the 22nd International Conference IEEE, Atlanta, Georgia, Apr. 3, 2006, pp. 125-127. |
| Kaiser, “Ginseng—A Natural Language User Interface for Semantic Web Search,” University of Zurich, Sep. 16, 2004, URL=http://www.ifi.unizh.ch/archive/mastertheses/DA—Arbeiten—2004/Kaiser—Christian.pdf, pp. 1-84. |
| Liang et al., “Extracting Statistical Data Frames from Text,” SIDKDD Explorations, Jun. 2005, vol. 7, No. 1, pp. 67-75. |
| Littman et al., “Automatic Cross-Language Information Retrieval using Latent Semantic Indexing,” In Grefenstette, G., editor, Cross Language Information Retrieval. Kluwer, 1998. |
| Nagao et al., “Semantic Annotation and Transcoding: Making Web Content More Accessible,” IEEE Multimedia, IEEE Computer Society, US. 8(2):69-81, Apr. 2001. |
| Nguyen et al., “Accessing Relational Databases from the World Wide Web,” SIGMOD Record, ACM USA, Jun. 1996, vol. 25, No. 2, pp. 529-540. |
| Pohlmann et al., “The Effect of Syntactic Phrase Indexing on Retrieval Performance for Dutch Texts,” Proceedings of RIAO, pp. 176-187, Jun. 1997. |
| Rasmussen, “WDB-A Web Interface to Sybase,” Astronomical Society of the Pacific Conference Series, Astron. Soc. Pacific USA, 1995, vol. 77, pp. 72-75. |
| Sneiders, “Automated Question Answering Using Question Templates That Cover the Conceptual Model of the Database,” Natural Language Processing and Information Systems, 6th International Conference on Applications of Natural Language to Information Systems, Revised Papers (Lecture Notes in Computer Science vol. 2553), Springer-Verlag, Berlin, Germany, 2002, vol. 2553, pp. 235-239. |
| Dekai Wu, A Stacked, Voted, Stacked Model for Named Entity Recognition, 2003, pp. 1-4. |
| Number | Date | Country | |
|---|---|---|---|
| 20070156669 A1 | Jul 2007 | US |
| Number | Date | Country | |
|---|---|---|---|
| 60737446 | Nov 2005 | US |
