The above and other aspects, features and advantages of several embodiments exemplarily described herein will be more apparent from the following more particular description thereof, presented in conjunction with the following drawings.
Corresponding reference characters indicate corresponding components throughout the several views of the drawings. Skilled artisans will appreciate that elements in the figures are illustrated for simplicity and clarity and have not necessarily been drawn to scale. For example, the dimensions of some of the elements in the figures may be exaggerated relative to other elements to help to improve understanding of various embodiments of the present invention. Also, common but well-understood elements that are useful or necessary in a commercially feasible embodiment are often not depicted in order to facilitate a less obstructed view of these various embodiments of the present invention.
The following description is not to be taken in a limiting sense, but is made merely for the purpose of describing the general principles of exemplary embodiments. It is to be understood that other embodiments may be utilized and that logical, mechanical, electrical and other changes may be made without departing from the spirit or scope of the numerous disclosed embodiments of the present invention. The scope of the numerous disclosed embodiments of the present invention should be determined with reference to the claims.
Some portions of the detailed descriptions which follow are presented in terms of algorithms and symbolic representations of operations on data bits within a computer memory. These algorithmic descriptions and representations are the means used by those skilled in the data processing arts to most effectively convey the substance of their work to others skilled in the art. An algorithm is here, and generally, conceived to be a self-consistent sequence of steps leading to a desired result. The steps are those requiring physical manipulations of physical quantities. Usually, though not necessarily, these quantities take the form of electrical or magnetic signals capable of being stored, transferred, combined, compared, and otherwise manipulated.
It has proven convenient at times, principally for reasons of common usage, to refer to these signals as bits, values, elements, symbols, characters, terms, numbers, or the like. All of these and similar terms are to be associated with the appropriate physical quantities and are merely convenient labels applied to these quantities. Unless specifically stated otherwise as apparent from the following discussions, it is appreciated that throughout the present invention, discussions utilizing terms such as processing or computing or calculating or determining or displaying or the like, refer to the action and processes of a computer system, or similar electronic computing device, that manipulates and transforms data represented as physical (electronic) quantities within the computer system's registers and memories into other data similarly represented as physical quantities within the computer system memories or registers or other such information storage, transmission or display devices.
It will be appreciated that more or fewer processes may be incorporated into the methods illustrated in the flow diagrams discussed below without departing from the scope of the numerous disclosed embodiments of the present invention. Describing the methods by reference to a flow diagram enables one skilled in the art to develop such programs including such instructions to carry out the methods on suitably configured hardware platforms and operating environments. The computer-executable instructions may be written in a computer programming language or may be embodied in firmware logic. If written in a programming language conforming to a recognized standard, such instructions can be executed on a variety of hardware platforms and operating environments and for interface to a variety of operating systems. In addition, embodiments of the present invention are not described with reference to any particular programming language. It will be appreciated that a variety of programming languages may be used to implement the teachings of the disclosed embodiments.
As shown in
Categorical data describes multiple attributes or categories. Often the categories are discrete and lack a natural similarity measure between them. The input processing module 109 pre-processes input data into categorical data and load the categorical data 110.
The categorical data 110 is grouped into clusters, and/or classified into folders by the clustering/classification module 120. The category and classification methods described herein can operate on any arbitrary categorical dataset. Generally, clustering of categorical data 110 is the process of identifying a set of underlying patterns in categorical data points within a data set and grouping the data points into one or more groups or clusters based on some measure of similarity. According to numerous embodiments disclosed herein, the clustering/classification module 120 can cluster the categorical data 110 by using seed based clustering techniques. By implementing the seed based clustering techniques disclosed herein, clusters can be organized in a particular manner to reflect, for example, a user's previous knowledge of an exemplar codebook.
The knowledge of category distribution obtained in the process of clustering is used to construct a classifier for some data spaces. Classification is the task of assigning data points to one or more of the identified categories based on some predefined proximity or divergence measure.
According to numerous embodiments described herein, the clustering/classification module 120 uses a folder-based classification system. The folder-based classification system allows a user to build a classifier by creating a group of folders. Each folder within the group is instantiated by a number of exemplary categorical data records.
In one embodiment, one or more of the exemplary records may be input by the user. Accordingly, a group of folders created by the user may act as a classifier such that new categorical data records are compared against the user-created group of folders and automatically sorted into the most appropriate folder.
The output of the clustering/classification module 120 is an organizational data structure 130, such as a dendrogram, a cluster tree, or a matrix, collectively referred to herein as a taxonomy. A cluster tree may be used as an indexed organization of the categorical data or to select a suitable cluster of the data.
Many clustering applications require identification of a specific layer within a cluster tree that best describes the underlying distribution of patterns within the categorical data. In one embodiment, an output of the clustering/classification module 120 includes an optimal layer that contains a unique cluster group containing an optimal number of clusters.
A data analysis module 140 may use the folder-based classifiers and/or classifiers generated by clustering operations for automatic recommendation or selection of content. The data analysis module 140 may automatically recommend or provide content that may be of interest to a user or may be similar or related to content selected by a user. In one embodiment, the data analysis module 140 assigns categorical data records for new content items with the appropriate folders based on similarity.
A user interface 150 also shown in
Although shown in
Referring to
In one embodiment, the input data received at 202 includes a plurality of records electronically input into the system 100. As used herein, a record can be a document, a video, photograph, music, or the like, or any combination thereof.
An exemplary process flow of pre-processing the dataset at 204 is discussed below with respect to
Referring to
In one embodiment, records within the received dataset can be mapped onto predetermined fields of a system ontology at 302. The format of the data, the transformations expected, and the format of the output data are specified in a language designed to describe metadata such as RDF. RDF employs a graph-based data model with a well-defined entailment relation. RDF is expressed in an XML format, the syntax of which explicitly specifies the parameters required and simplifies specification of the transformation process.
As shown in
As shown in
Records are represented within the system as vectors. The dimension of each vector corresponds to the total number of terms characterizing all attributes found in all records processed by the system (i.e., the global vocabulary of the system). Values assigned to components of a vector represent the presence of a term within a corresponding record. For example, a vector component can be binarily represented as either a 0 (indicating the absence of a term from a record) or a 1 (indicating the presence of a term in a record).
For example, suppose that the vocabulary for the entire collection of records and attributes (i.e., the global vocabulary) consists of only six terms: A, B, C, D, E, and F. Let a first record d1 include the terms A, B, D, and E and a second record d2 include the terms C, D, and F. Accordingly, the first record d1 can be binarily represented by a six-dimensional vector d1={1, 1, 0, 1, 1, 0} and the second record d2 can be binarily represented by a six-dimensional vector d2={0, 0, 1, 1, 0, 1}. As shown above, the global vocabulary consists of only six terms. In practice, however, the global vocabulary includes thousands of terms, wherein any particular record includes only a relatively small number terms. Accordingly, only a relatively few number of vector components are assigned non-zero values. In practice, therefore, records are typically represented as highly “sparse” vectors that can be difficult to process to efficiently generate taxonomies and yield inaccurate results. To increase the efficiency with which taxonomies are generated, and to produce more accurate results, each record is “cleaned” at 304.
In other embodiments, vector components can be numerically represented (as opposed to binarily represented) by a value corresponding to the number of times that a term occurs in a record (i.e., the frequency of the term in the record) or by a normalized value corresponding to the number of times that a term occurs in a record in addition to the total number of terms the record contains (i.e., the normalized frequency of the term in the record). An exemplary process flow of performing the cleaning at 304 is discussed below with respect to
Referring to
In one embodiment, terms can be removed by removing terms from attributes that are not generally useful in generating a folder set or in classifying records based on the generated folder set. Non-useful terms can be identified as those that occur infrequently (e.g., once) within a particular attribute of a record. By removing terms from each record, the total number of terms within the attribute vocabulary can be significantly reduced. For example, suppose the system receives 2154 records describing television programs playing over the course of a week. Such a collection of records may include 3016 description keywords, wherein each record contains an average of 5.37 keywords, wherein each keyword occurs an average of 4.29 times, and wherein 1113 keywords occur just once. By removing the keywords occurring just once, the size of the description keyword attribute space can be reduced by about 33%.
In one embodiment, the attributes are split by splitting large record attributes (i.e., attributes characterized by a large number of unique terms) into a plurality of smaller sub-attributes characterized by a smaller number of unique terms. By splitting a large record attribute into small sub-attributes, probabilities can be normalized over smaller, less sparse domains and yield more accurate results. Further, each sub-attribute can be weighted to reflect its relative importance within the record.
In one embodiment, a large attribute can be split into a plurality of smaller sub-attributes by grouping terms within the large attribute based on their frequency of occurrence within the large attribute. In this case, it is assumed that terms occurring more frequently are more relevant, and thus more important, than terms occurring less frequently). As a result, information indicating relevance levels of particular sub-attributes is imparted and used as a weighting value to reflect the relative importance of each sub-attribute. The size and number of the sub-attributes created as a result of the splitting can be controlled, for example, by specifying the proportions of frequencies or value of frequencies desired in each sub-attribute.
In another embodiment, a large attribute can be split into a plurality of sub-attributes by grouping semantically-related terms. In this case, meanings of words may be internally generated or obtained from a lexical reference system such as WordNet, a lexical reference system developed at the Cognitive Science Laboratory at Princeton University. As a result, information indicating a certain degree of “common-sense” is imparted and used as a weighting value to reflect the relevance of each sub-attribute. The size and number of the sub-attributes created as a result of the splitting can be controlled using statistics about particular terms and term frequencies in the attribute. In one embodiment, the total number of occurrences of terms within each sub-attribute can be maintained within a predetermined range to control the degree of relatedness between terms in a particular sub-attribute or to control the abstractness of relationships between terms within a sub-attribute.
In one embodiment, terms that can be grouped together into a sub-attribute according to their semantic interrelatedness are identified using an “is-a” (hypernym) chain relation as applied to nouns.
Consider, for example, the following hypernym chains:
1. cat, feline, carnivore, placental, mammal, vertebrate, chordate, animal, organism, living thing, object, entity
2. dog, canine, carnivore, placental, mammal, vertebrate, chordate, animal, organism, living thing, object, entity
3. umbrella, canopy, shelter, protective covering, covering, artifact, object, entity
Based on the first hypernym chain, a cat is a feline; a feline is a carnivore, etc. Based on the second hypernym chain, a dog is a canine; a canine is a carnivore, etc. Based on the third hypernym chain, an umbrella is a canopy; a canopy is a shelter, etc. As shown, the terms ‘cat’ and ‘dog’ share the ‘carnivore’ hypernym relation much earlier in the chain than they share the ‘object’ hypernym relation with the term ‘umbrella’. Accordingly, the term ‘cat’ is more semantically related to the term ‘dog’ than it is related to the term ‘umbrella’. As a result, a large attribute containing the terms ‘cat,’ ‘dog,’ and ‘umbrella’ will be split into a first sub-attribute containing the terms ‘cat’ and ‘dog’ and a second sub-attribute containing the term ‘umbrella.’ As will be appreciated, semantic relationships between terms changes as the structural taxonomy of the lexical reference system changes or as classes are inserted into or removed from a particular relation path. It will be appreciated, however, that terms other than the nouns appearing in WordNet can be grouped in a separate attribute space and downweighted if necessary.
In the example provided above, where the received records identify television programs, one large attribute may, for example, be divided into two smaller sub-attributes, wherein the first sub-attribute, related to recreation, contains the terms ‘Recreation’, ‘Pachinko’, ‘Hobby’, ‘Fun’, ‘Entertainment’, ‘Encore’, ‘Swimming’, ‘Skating’, ‘Gymnastics’, ‘Hunting’, ‘Fishing’, ‘Tennis’, ‘Basketball’, ‘Golf’, ‘Soccer’, ‘Baseball’, and ‘Athletics’ while the second sub-attribute, related to food, contains the terms ‘Tofu’, ° Food’, ‘Diet’, ‘Vitamin’, ‘Sushi’, ‘Soup’, ‘Pudding’, ‘Dessert’, ‘Chocolate’, and ‘Beverage’. In one embodiment, each of the terms identified above can be further divided into smaller sub-attributes according to a semantic relatedness between the terms.
In one embodiment, terms can be replaced by replacing specific terms with equivalent, more abstract terms. In this case, terms found, for example, in WordNet can be replaced with equivalent, more abstract terms. As a result, the number of unique terms characterizing an attribute of a record can be reduced because several specific terms can be mapped onto the same abstract term. Moreover, vector representations of records become much less sparse because each abstract term appears in more records and there are proportionately more abstract terms appearing in each record. The size and number of the sub-attributes created as a result of the term replacements can be controlled using statistics about particular terms and term frequencies in the attribute.
In the example provided above, where the received records identify television programs, the specific terms ‘Brother’, ‘Sister’, ‘Grandchild’, ‘Baby’, ‘Infant’, ‘Son’, ‘Daughter’, ‘Husband’, ‘Mother’, ‘Parent’, and ‘Father’ of an attribute can all be mapped onto an equivalent abstract term ‘relative’. Similarly, the specific terms ‘Hunting’, ‘Fishing’, ‘Gymnastics’, ‘Basketball’, ‘Tennis’, ‘Golf’, ‘Soccer’, ‘Football’, and ‘Baseball’ of an attribute can all be mapped onto an equivalent abstract term ‘sport’.
By cleaning each record as exemplarily described above with respect to
The dataset D can be represented as an m×n matrix. Accordingly, each row of the m×n matrix corresponds to a vector-space representation of a particular record.
Once the categorical data has been pre-processed, a plurality of exemplars (e.g., x exemplars, where x<<m) are received at 206. In one embodiment, and as similarly discussed with respect the records received at 202, each exemplar can be represented as a vector having the same dimension as each record pre-preprocessed at 204, wherein values assigned to each component of the exemplar vector represent the presence of a term within the particular exemplar. In another embodiment, each exemplar can be labeled with a predetermined category label.
In one embodiment, the dataset D can be augmented with the received exemplars at 208 simply by treating the exemplars as records. Accordingly, the dataset D obtained as a result of augmenting the dataset D with the received exemplars at 208 contains m+x records {d1, d2, . . . , dm, dm+1, . . . , dm+x}.
Once the input data has been pre-processed and augmented, a taxonomy can be generated at 210. In accordance with numerous embodiments, the taxonomy, Γ, generated at 210 is a partial taxonomy of the dataset D resulting from the augmentation at 208. A given layer Γt within the partial taxonomy Γ contains a set of k(t)clusters nested within a hierarchically higher layer Γt+1, containing a set of k(t+1) clusters (k(t+1)<k(t)), such that Γt ⊂ Γt+1. Thus, each cluster in Γt is a subset of a cluster in Γt+1 and at least one cluster in Γt is a proper subset of a cluster in Γt+1. In one embodiment, the partial taxonomy Γ comprises g nested layers Γ1 ⊂ Γ2 ⊂ . . . ⊂ Γg. The uppermost layer Γg of the partial taxonomy Γ contains x clusters, collectively containing all of the m records and x exemplars. The bottommost layer Γ1 of the partial taxonomy contains m+x singleton leaf clusters, each corresponding to one of the m records or x exemplars (i.e., seeds) within the dataset D. An exemplary process flow of a taxonomy generation process is discussed below with respect to
Referring to
In one embodiment, the hierarchy index identifies layers within the taxonomy and, as shown above, is represented by the hierarchy index variable, t. Referring to
After initializing the hierarchy index at 602, an internal representation of the leaf clusters is created at 604.
In one embodiment, a generic cluster Cti ε Γt can be internally represented by its conditional probability distribution P(A|Cti), where P(A=vj|Cti) is the normalized probability of occurrence of the term vj in the cluster Cti. Similarly, a generic record di ε D can be internally represented by its conditional probability distribution P(A|di), where P(A=Vj|di) is the normalized probability of occurrence of the term vj in di. The internal representation scheme described above is herein referred to as a globally normalized (i.e. g-norm) representation because the term-frequency vectors of all records and clusters are normalized across the global attribute space A to generate the probability mass function representation.
In another embodiment, each generic record di ε D or cluster Cti ε Γt can be internally represented as a set of d probability mass functions, one for each attribute. Accordingly, the internal representation of every record di ε D is the disjunction of d locally normalized conditional probability distributions, {P(A1|di)P(A2|di)
. . .
P(Ad|di)} and the internal representation of every cluster Cti ε Γt is the disjunction of d locally normalized conditional probability distributions, {P(A1|Cti)
P(A2|Cti)
. . .
P(Ad|Cti)}. The internal representation scheme described above is herein referred to as locally normalized (i.e., l-norm) representation.
Compared to g-norm, 1-norm representation provides d degrees of freedom, as there are d entropy/information components corresponding to d probability distributions. The entropy/information components can be linearly combined in any manner, and therefore any type of clustering can be obtained varying the weight distribution across attributes. Moreover, g-norm representation propagates the effect of sparsity within one attribute to all other attributes whereas l-norm representation confines attribute-local sparsities to the same local attribute space.
Therefore, and as shown in
After creating the internal representation of the leaf clusters, the truncated hierarchy can be built at 606. In one embodiment, the truncated hierarchy is generated according to a constrained hierarchical agglomerative clustering process. In one embodiment, the constrained hierarchical agglomerative clustering process is a clustering process that partitions the dataset D into k mutually exclusive clusters (where k≦m), C1, C2, . . . , Ck, such that the following conditions are satisfied:
1. For all i=1, . . . , k, Ci is a non-null subset;
2. For all i, j=1, . . . , k, i≠j, Ci and Cj are non-overlapping, i.e., Ci ∩ Cj is null; and
3. All the clusters, when combined together, are exhaustive, i.e., ∪i=1k Ci=D.
Referring to
Upon initially performing 902, the value of the cluster number variable, k, is inherited from 602 and the value of the taxonomy layer variable, t, is inherited from 802 and incremented by one. Subsequent values of the cluster number variable, k, and the taxonomy layer variable, t, are based on the number of clusters contained within each new taxonomy (i.e., dendrogram) layer after cluster pairs have been merged and non-merged clusters have been isomorphically mapped from preceding taxonomy layers.
In one embodiment, the distance matrix Mt-1 created at 904 is an i×j matrix (i=j=1, . . . , k) listing all possible combinations of clusters present within the taxonomy layer Γt-1. In another embodiment, each entry within the distance matrix Mt-1 identifies a degree of similarity between a cluster identified in the ith row and a cluster identified in the jth column of the distance matrix Mt-1.
Where clusters are internally represented under the l-norm representation schema, the degree of similarity can be represented by an entropic distance metric, μ that characterizes a measure of the divergence (or proximity) between every possible pair-wise combination of clusters present within the taxonomy layer Γt-1. In one embodiment, values for the entropic distance metric, μ, can be obtained using principles related to information-theoretic data compression.
According to information-theoretic data compression, if the value-set of a random variable X is compressed into a smaller-sized value-set of another random variable Y, and if both X and Y contain information about a third random variable Z, then. The information content of Y is always upper-bounded by (i.e., less than or equal to) the information content of X, which means there is always some finite amount of information loss when X is compressed to Y.
Applying the aforementioned information-theoretic compression principles to the context of generating a taxonomy of categorical data, the information content of Γt-1, about the global vocabulary A, is defined as:
I(A, Γt-1)=H(A)−H(A|Γt-1),
where H(A) is the absolute entropy and H(A|Γt-1) is the Γt-1-conditional entropy of A. Similarly, the information content of Γt, about the global vocabulary A, is I(A,Γt).
During the hierarchical agglomerative clustering process, Γt is created from Γt-1 and the loss of information content (i.e., information loss or differential mutual information) incurred when Γt is generated from Γt-1 is defined as:
δI(Γt-1,Γt)=I(A,Γt-1)−I(A,Γt)=H(A|Γt)−H(A|Γt-1).
As shown above, the amount of information loss incurred when Γt is generated from Γt-1 is equal to the increase in the conditional uncertainty when Γt is generated from Γt-1. Accordingly, the equivalence between differential mutual information δI(Γt-1,Γt) and μ is as follows:
δI(Γt-1,Γt)=μ(P(A|Ct-1i), P(A|Ct-1j))
In one embodiment, the differential mutual information δI(Γt-1,Γt) can be determined based on the well known Jensen-Shannon (JS) divergence measure, quantifying the difference between two or more probability distributions, which are transformed representations of categorical data vectors. The JS measure is built upon the Kullback-Leibler (KL) divergence measure and, therefore, inherits all the properties of KL divergence. However, the JS measure does not suffer from the singularity problem. Moreover, the JS divergence can be generalized to more than two distributions and argument distributions can be weighted.
For any two arbitrary, mutually exclusive clusters, Ct-1i, Ct-1j ε Γt-1, with respective likelihood probabilities πi and πj, let the union be Ct-1u=Ct-1i∪Ct-1j, with corresponding likelihood probability πu=πi+πj. The probability mass function representation of the merged cluster Ct-1u over the attribute space can be defined as the weighted mean of the probability functions of Ct-1i and Ct-1j:
and the JS-divergence between Ct-1i and Ct-1j:
Using this JS-divergence function, the differential mutual information between Ct-1i and Ct-1j can be represented as a linear function of their JS-divergence:
μ(P(A|Ct-1i), P(A|Ct-1j))=(πi+πj)JS(P(A|Ct-1i)∥P(A|Ct-1j)).
Thus, for each i=1, . . . , m, and for each j=1, . . . , k, the entry M[i, j] created at 904 denotes μ(P(A|Ct-1i),P(A|Ct-1j)) (i.e., the entropic distance between the ith and jth clusters in Γt-1).
As discussed above, the l-norm representation schema transforms each record and cluster into a set of d locally normalized probability distributions, one normalized probability distribution for each of the d attributes. Therefore, the entropic proximity or divergence measure between two records or clusters is a weighted summation of the entropic similarity or dissimilarity measure between respective intra-attribute probability functions. Assuming that the weight distribution over the attributes A1, A2, . . . , Ad is {ω1, ω2, . . . , ωd}, each individual weight can be used to regulate the contribution of an entropic proximity or divergence measure specific to a particular attribute. In one embodiment, the entropic divergence measure is obtained as follows:
and the entropic proximity measure as follows:
where λ is a gradient-controlling parameter of the entropic measure μ.
In many embodiments, less weight may be given to the divergence between two probability components if the corresponding attribute is of a higher weight while more weight may be given the proximity between two probability components if the corresponding attribute is of a higher weight.
In view of the above, the predetermined entropic similarity condition, from which a cluster pair is at least partially identified at 906, represents a minimal amount of information loss incurred as a result of generating a new taxonomy layer Γt from a previous taxonomy layer Γt-1. Referring back to the principles of information-theoretic data compression, a high-quality data compression is achieved when the amount of information loss is minimized. Accordingly, and as applied to the present case of building the hierarchy, pairs of clusters listed in the distance matrix Mt-1 are identified at 906 at least partially by identifying a cluster pair listed within the distance matrix Mt-1 that, when merged, will minimize the amount of information loss δI(Γt-1,Γt) incurred when Γt is generated from Γt-1. Stated another way, pairs of clusters listed in the distance matrix Mt-1 are identified at 906 at least partially by identifying a cluster pair listed within the distance matrix Mt-1 which, when merged, will maximize the reduction of uncertainty as numerous, smaller-sized clusters are grouped into fewer, larger-sized clusters.
In an embodiment where the entropic distance metric, μ, represents a measure of proximity (i.e., closeness) between clusters Ct-1i and Ct-i j, clusters Ct-1r and Ct-1s are at least partially identified for which μ(P(A|Ct-1i),P(A|Ct-1j)) is a maximum (i.e., a maximum entropic proximity), resulting in the equivalence:
In an embodiment where the entropic distance metric, μ, represents a measure of divergence (i.e., farness) between clusters Ct-1i and Ct-1j, clusters Ct-1r and Ct-1s are at least partially identified for which μ(P(A|Ct-1i),P(A|Ct-1j) is a minimum (i.e., a minimum entropic divergence), resulting in the equivalence:
In one embodiment, the predetermined constraint condition, from which a cluster pair is at least partially identified at 906, represents whether or not the cluster pair contains more than one exemplar {dm+1, . . . , dm+x}. Accordingly, pairs of clusters listed in the distance matrix Mt-1 are identified at 906 at least partially by identifying a cluster pair listed within the distance matrix Mt-1 that, when merged, will contain less than two exemplars {dm+1, . . . , dm+x}.
Thus, clusters Ct-1i and Ct-1j are identified and marked as Ct-1r and Ct-1s, respectively, at 906 if they have a maximum amount of overlap in the information that they contain about A and do not both contain an exemplar {dm+1, . . . , dm+x}.
In one embodiment, clusters can be merged at 908 by performing a union operation of all identified clusters C t-1r and Ct-1s to form Ct-1u (i.e., Ct-1u←Ct-1r∪Ct-1s). Accordingly, each merged cluster Ct-1u represents the union of a plurality of clusters. As mentioned above each leaf cluster C1i is identified by an underlying vector. In one embodiment, Ct-1u can be identified by an underlying vector representing the centroid of the plurality of clusters it contains. For the sake of ease of exposition, no formal distinction will be drawn between a merged cluster and the underlying vector identifying the merged cluster.
In one embodiment, the new taxonomy layer can be created at 912 by isomorphically mapping the non-merged clusters onto the new taxonomy layer.
An exemplary method of identifying a pair of clusters listed in the distance matrix Mt-1 according to predetermined constraint and similarity conditions as discussed above with respect to 906 to facilitate merging identified clusters into a single cluster at 908 will now be discussed in greater detail with respect to
Referring to
when the entropic distance metric, μ, represents a measure of divergence (1002), setting a distance matrix row counter variable, i, equal to 1 (1004), setting a distance matrix column counter variable, j, equal to 1 (1006), and determining whether the
If, as a result of 1008, it is determined that
the process flow is directed to 1010 where it is determined whether both clusters Ct-1i and Ct-1j contain an exemplar {dm+1, . . . , dm+x}. If, as a result of 1010, it is determined that both Ct-1i and Ct-1j do not contain an exemplar (e.g., that Ct-1i contains an exemplar but that Ct-1j does not contain an exemplar, that Ct-1j contains an exemplar but that Ct-1i does not contain an exemplar, or that neither Ct-1j nor Ct-1i contain an exemplar), then the process flow proceeds to 1012 where Ct-1i and Ct-1j are marked Ct-1r and Ct-1s, respectively, and are subsequently merged into a single cluster Ct-1u at 1014. The process flow then continues to 910 discussed above.
If, as a result of 1008, it is determined that
(or if, as a result of 1010, it is determined that both Ct-1i and Ct-1j contain an exemplar), then the process flow proceeds directly to 1016 where the distance matrix column counter variable, j, is incremented by 1 and, subsequently at 1018, it is determined whether the incremented distance matrix column counter variable, j, exceeds the total number of clusters, k, within the taxonomy layer Γt-1.
If, as a result of 1018, it is determined that the incremented distance matrix column counter variable, j, is less than or equal to the total number of clusters, k, the process flow is directed back to 1008. If, as a result of 1018, it is determined that the incremented distance matrix column counter variable, j, is greater than the total number of clusters, k, the process flow continues to 1020 where the distance matrix row counter variable, i, is incremented by 1 and, subsequently at 1022, it is determined whether the incremented distance matrix column row variable, i, exceeds the total number of clusters, k, within the taxonomy layer Γt-1.
If, as a result of 1022, it is determined that the incremented distance matrix column row variable, i, is less than or equal to the total number of clusters, k, the process flow is directed back to 1008. If, as a result of 1022, it is determined that the incremented distance matrix row counter variable, i, is greater than the total number of clusters, k, the process flow continues to 910 discussed above.
Referring back to
As discussed above, the partial taxonomy generated as discussed above with respect to 606 provides a truncated hierarchy that is created according to the constrained agglomeration process exemplarily discussed with respect to
For example, the partial taxonomy generated at 210 can be completed by representing the uppermost layer of the partial taxonomy as an intermediate taxonomy layer and appending a non-truncated hierarchy to the partial taxonomy in an unconstrained hierarchical agglomerative clustering process. In one embodiment, the unconstrained hierarchical agglomerative clustering process is a clustering process that partitions the dataset D into k mutually exclusive clusters (where k≦x), C1,C2, . . . , Ck, such that the following conditions are satisfied:
Referring to
Upon initially performing 1102, the value of the cluster number variable, k, is set to x and the value of the taxonomy layer variable, t, is set to g and incremented by one. Subsequent values of the cluster number variable, k, and the taxonomy layer variable, t, are based on the number of clusters contained within each new taxonomy layer after cluster pairs have been merged and non-merged clusters have been isomorphically mapped from preceding taxonomy layers.
The distance matrix Mt-1 created at 1104 is created in the same manner as previously discussed with respect to 904. Moreover, the predetermined entropic similarity condition, from which a cluster pair is identified at 1106, represents a minimal amount of information loss incurred as a result of generating a new taxonomy layer Γt from a previous taxonomy layer Γt-1 as similarly discussed with respect to 906. Accordingly, a pair of clusters listed in the distance matrix Mt-1 is identified at 1106 by identifying a cluster pair listed within the distance matrix Mt-1 that, when merged, will minimize the amount of information loss δI(Γt-1, Γt) incurred when Γt is generated from Γt-1. Stated another way, a pair of clusters listed in the distance matrix Mt-1 is identified at 1106 by identifying a cluster pair listed within the distance matrix Mt-1 which, when merged, will maximize the reduction of uncertainty as numerous, smaller-sized clusters are grouped into fewer, larger-sized clusters.
In one embodiment, clusters can be merged at 1108 in the same manner as previously described at 908. Moreover, a new taxonomy layer can be created at 1112 in the same manner as previously described at 912.
An exemplary method of identifying pairs of clusters listed in the distance matrix Mt-1 according to a predetermined entropic similarity condition as discussed above with respect to 1106 to facilitate merging identified clusters into a single cluster at 1108 will now be discussed in greater detail with respect to
Referring to
wherein the entropic distance metric, μ, represents a measure of divergence (1202), setting a distance matrix row counter variable, i, equal to 1 (1204), setting a distance matrix column counter variable, j, equal to 1 (1206), and determining whether the
If, as a result of 1208, it is determined that
then the process flow proceeds to 1210 where Ct-1i and Ct-1j are marked Ct-1r and Ct-1s, respectively, and are subsequently merged into a single cluster Ct-1u at 1212. The process flow then continues to 1110 discussed above.
If, as a result of 1208, it is determined that
then the process flow proceeds directly to 1214 where the distance matrix column counter variable, j, is incremented by 1 and, subsequently at 1216, it is determined whether the incremented distance matrix column counter variable, j, exceeds the total number of clusters, k, within the taxonomy layer Γt-1.
If, as a result of 1216, it is determined that the incremented distance matrix column counter variable, j, is less than or equal to the total number of clusters, k, the process flow is directed back to 1208. If, as a result of 1216, it is determined that the incremented distance matrix column counter variable, j, is greater than the total number of clusters, k, the process flow continues to 1218 where the distance matrix row counter variable, i, is incremented by 1 and, subsequently at 1220, it is determined whether the incremented distance matrix column row variable, i, exceeds the total number of clusters, k, within the taxonomy layer Γt-1.
If, as a result of 1220, it is determined that the incremented distance matrix column row variable, i, is less than or equal to the total number of clusters, k, the process flow is directed back to 1208. If, as a result of 1220, it is determined that the incremented distance matrix row counter variable, i, is greater than the total number of clusters, k, the process flow continues to 1110 discussed above.
As discussed above, the numerous embodiments discussed above provide a taxonomy generated in accordance with a seed based clustering system and method. The taxonomy that has been generated from the categorical dataset can be used in various ways. For example, the taxonomy can be used as an indexed organization of the categorical dataset or a layer within the taxonomy can be selected to characterize and classify the categorical dataset. An exemplary process of classifying records within the categorical dataset can include treating the selected taxonomy layer as a nearest neighbor classifier.
While the invention herein disclosed has been described by means of specific embodiments, examples and applications thereof, numerous modifications and variations could be made thereto by those skilled in the art without departing from the scope of the invention set forth in the claims.
This application is a continuation-in-part of U.S. application Ser. No. 11/436,142, filed May 16, 2006, entitled “CLUSTERING AND CLASSIFICATION OF CATEGORY DATA” to Acharya et al., (Attorney Docket No. 080398.P649) which is incorporated in its entirety herein by reference.
Number | Date | Country | |
---|---|---|---|
Parent | 11436142 | May 2006 | US |
Child | 11457115 | US |