KR102028487B1 - Document topic modeling apparatus and method, storage media storing the same - Google Patents

Abstract

The present invention relates to an apparatus and method for modeling a topic of a document. The present invention relates to a method for generating local topic information by dividing a global document set including a plurality of documents, generating a plurality of local document sets, and performing topical modeling. A local topic information generator, a global topic information generator for generating a global topic information by generating a reduced global document set through documents in each of the plurality of local document sets, and performing global topic modeling; and the local and global topics And a document topic assignment system that assigns topic conversion information generated through document mapping between the information to each of the plurality of documents. Therefore, the present invention can reasonably integrate the results of the segmented topic modeling.

Description

TOPIC MODELING APPARATUS AND METHOD, STORAGE MEDIA STORING THE SAME

The present invention relates to a topic modeling technique of a document, and more particularly, to a topic modeling apparatus and method of a document capable of rationally integrating the results of segmented topic modeling.

Topic modeling can extract key issues from a large volume of documents, identify documents that correspond to each issue, and provide them as a cluster. Topic modeling is a very useful technique because it presents the result reflecting the semantic elements of the document, unlike the general document clustering technique. However, traditional topic modeling is based on the distribution of key terms across the entire document, so a batch analysis of the entire document must be performed to identify the topic of each document. As a result, topic modeling of a large document takes a very long time, and may cause a problem of scalability in which the analysis time increases exponentially as the size of the analysis target increases.

Korean Patent Laid-Open No. 10-2009-0013928 (2009.02.06) relates to a topic extracting apparatus, a social network generating system using the topic extracting apparatus, and a method thereof, which automatically extracts a topic representing a user's characteristics. The present invention provides a method, a ranking method and a management method for improving the accuracy of the extracted topic, and can provide an apparatus and method for automatically forming a social relationship among a plurality of users using the same.

Korean Patent Publication No. 10-2009-0013928 (2009.02.06)

An embodiment of the present invention is to provide a topic modeling apparatus and method of a document that can reasonably integrate the results of partitioned topic modeling.

An embodiment of the present invention is to provide a topic modeling apparatus and method for dividing an entire document into sub-groups and extracting a representative document from each sub-group to form a reduced global document set.

An embodiment of the present invention is to provide a topic modeling apparatus and method of a document that can derive the components of the global topic from the local topic derived from the sub-group through the representative document.

Among the embodiments, a local topic information generator for generating a plurality of local document sets by dividing a global document set including a plurality of documents and generating local topic information by performing local topic modeling. A topic created by a global topic information generator that generates a reduced global document set through documents in each of the local document sets and performs global topic modeling to generate global topic information, and a document mapping between the local and global topic information. A document topic assignment for allocating conversion information to each of the plurality of documents.

The regional topic information generation unit performs the regional topic modeling to extract at least one major regional topic for each of the plurality of regional document sets, and a regional topic between the local document belonging to the regional document set and the at least one major regional topic. You can create a matrix.

The global topic information generation unit may generate the reduced global document set by extracting at least one regional representative document in the order of a high weight for each topic for each of the plurality of local document sets.

The global topic information generator extracts at least one main global topic for the reduced global document set by performing the global topic modeling, and between the local representative document belonging to the reduced global document set and the at least one main global topic. You can create a global topic matrix.

The document topic assignment unit extracts topic component information associated with the local representative document from a local topic matrix generated through the local topic modeling, and multiplies the topic component information and a global topic matrix generated through the global topic modeling. At least one topic transformation matrix may be generated.

The document topic locator may assign global topic weights to the plurality of documents by multiplying each of the local topic matrices with the at least one topic transformation matrix.

Among the embodiments, a topic modeling method of a document may include (a) generating a plurality of local document sets by dividing a global document set including a plurality of documents, and performing local topic modeling to generate local topic information. (B) generating a reduced global document set through a document in each of the plurality of local document sets and performing global topic modeling to generate global topic information; and (c) between the local and global topic information. And assigning the topic transformation information generated through document mapping to each of the plurality of documents.

In the step (a), the local topic modeling is performed to extract at least one main regional topic for each of the plurality of regional document sets, and the local topic between the local document belonging to the regional document set and the at least one main regional topic. It may be a step of generating a matrix.

The step (b) may be performed to generate the reduced global document set by extracting at least one regional representative document in the order of increasing weight for each topic for each of the plurality of regional document sets.

In the step (b), the global topic modeling is performed to extract at least one main global topic for the reduced global document set, and between the local representative document belonging to the reduced global document set and the at least one main global topic. It may be a step of generating a global topic matrix.

Step (c) extracts the topic component information associated with the regional representative document from the local topic matrix generated through the regional topic modeling, and performs a multiplication operation between the topic component information and the global topic matrix generated through the global topic modeling. It may be a step of generating at least one topic transformation matrix.

Step (c) may be a step of assigning global topic weights to the plurality of documents by multiplying each of the local topic matrices with the at least one topic transformation matrix.

In one or more embodiments, a computer-executable recording medium may generate a plurality of local document sets by dividing a global document set including a plurality of documents, and generate local topic information by performing local topic modeling. The topic transformation information generated by generating a reduced global document set through a document in each of the plurality of local document sets, performing global topic modeling to generate global topic information, and mapping a document between the local and global topic information. And assigning to each of the plurality of documents.

The disclosed technique can have the following effects. However, since a specific embodiment does not mean to include all of the following effects or only the following effects, it should not be understood that the scope of the disclosed technology is limited by this.

The apparatus and method for modeling a topic of a document according to an embodiment of the present invention may divide the entire document into sub-groups and extract a representative document from each sub-group to form a reduced global document set.

An apparatus and method for modeling a topic of a document according to an embodiment of the present invention may derive a component of a global topic from a local topic derived from a lower cluster through a representative document.

1 is a diagram illustrating a topic modeling system of a document according to an embodiment of the present invention.
FIG. 2 is a block diagram illustrating the topic modeling apparatus in FIG. 1.
3 is a flowchart illustrating a topic modeling process performed in the topic modeling apparatus of FIG. 1.
4 is a diagram showing an overall overview of a topic modeling system of a document according to an embodiment of the present invention.
FIG. 5 is an exemplary diagram illustrating an embodiment of regional topic information generated through regional topic modeling performed by the regional topic information generator of FIG. 2.
FIG. 6 is an exemplary diagram illustrating an embodiment of global topic information generated through global topic modeling of a reduced global document set performed by the global topic information generator of FIG. 2.
FIG. 7 is an exemplary diagram illustrating an embodiment of a process of generating topic transformation information through document mapping performed in the document topic assignment in FIG. 2.
FIG. 8 is an exemplary diagram illustrating an embodiment of a process of allocating topic transformation information to each of a plurality of documents in the document topic assignment in FIG. 2.

Description of the present invention is only an embodiment for structural or functional description, the scope of the present invention should not be construed as limited by the embodiments described in the text. That is, since the embodiments may be variously modified and may have various forms, the scope of the present invention should be understood to include equivalents capable of realizing the technical idea. In addition, the objects or effects presented in the present invention does not mean that a specific embodiment should include all or only such effects, the scope of the present invention should not be understood as being limited thereby.

On the other hand, the meaning of the terms described in the present application should be understood as follows.

Terms such as "first" and "second" are intended to distinguish one component from another component, and the scope of rights should not be limited by these terms. For example, the first component may be named a second component, and similarly, the second component may also be named a first component.

When a component is referred to as being "connected" to another component, it should be understood that there may be other components in between, although it may be directly connected to the other component. On the other hand, when a component is referred to as being "directly connected" to another component, it should be understood that there is no other component in between. On the other hand, other expressions describing the relationship between the components, such as "between" and "immediately between" or "neighboring to" and "directly neighboring to", should be interpreted as well.

Singular expressions should be understood to include plural expressions unless the context clearly indicates otherwise, and terms such as "comprise" or "have" refer to a feature, number, step, operation, component, part, or feature thereof. It is to be understood that the combination is intended to be present and does not exclude in advance the possibility of the presence or addition of one or more other features or numbers, steps, operations, components, parts or combinations thereof.

In each step, an identification code (e.g., a, b, c, etc.) is used for convenience of description, and the identification code does not describe the order of the steps, and each step clearly indicates a specific order in context. Unless stated otherwise, they may occur out of the order noted. That is, each step may occur in the same order as specified, may be performed substantially simultaneously, or may be performed in the reverse order.

The present invention can be embodied as computer readable code on a computer readable recording medium, and the computer readable recording medium includes all kinds of recording devices in which data can be read by a computer system. . Examples of computer-readable recording media include ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical data storage device, and the like. The computer readable recording medium can also be distributed over network coupled computer systems so that the computer readable code is stored and executed in a distributed fashion.

All terms used herein have the same meaning as commonly understood by one of ordinary skill in the art unless otherwise defined. Generally, the terms defined in the dictionary used are to be interpreted to coincide with the meanings in the context of the related art, and should not be interpreted as having ideal or excessively formal meanings unless clearly defined in the present application.

Topic Modeling is one of the statistical models for discovering the abstract 'topics' of a document set of documents, and may correspond to one of the text mining techniques used to discover hidden semantics of text bodies. Topic modeling is used for text structuring and main topic extraction in text analysis studies.

Topic modeling generally uses the concept of a bag of words, which recognizes each document as a set of terms that appear in that document. Since each document contains a large number of words, it can be represented as an appropriate number of word clusters through the dimension reduction process, and the number of dimensions derived from this process can correspond to the number of topics. As a result, each document has a Document Topic Weight, which is a correspondence to an individual topic, and each document's topic through the Document Cutoff, which is generally calculated through the “average + 1σ” of the document weight. Determine whether it is included.

That is, it is interpreted that a document having a document weight equal to or greater than the document threshold value includes the corresponding topic. In this process, the document weight of each document and topic is expressed as a two-dimensional matrix called a document / topic matrix. In the present invention, a local topic matrix or a global topic matrix may be used as an embodiment of the document / topic matrix.

1 is a diagram illustrating a topic modeling system of a document according to an embodiment of the present invention.

Referring to FIG. 1, a topic modeling system of a document 100 may include a topic modeling apparatus 110 and a database 130.

The topic modeling apparatus 110 may be implemented as a server corresponding to a computer or a program capable of performing topic modeling on a document set to extract a topic for the document set. The topic modeling apparatus 110 may use the document set data stored in the internal memory, and receive the document set data from an independently implemented database 130 to perform topic modeling.

The topic modeling apparatus 110 may be implemented including a database 130 and may be implemented independently of the database 130. When implemented independently of the database 130, the topic modeling device 110 may be connected to the database 130 by wire or wirelessly to exchange data.

The database 130 is a storage device that stores various information necessary for performing topic modeling of a document. The database 130 may store information about a document set for the topic modeling device 110 to perform topic modeling, and may store topic information, local topic information, and global topic information on a document calculated during the topic modeling process. However, the present invention is not limited thereto and may store information collected or processed in various forms in connection with topic modeling of a document. Here, geographic and global topic information can include geographic and global topic matrices that are produced by modeling geographic and global topics for a set of geographic and global documents, and geographic and global topic matrices correspond to weight matrices between documents and topics. can do.

The database 130 may be composed of at least one independent sub-databases that store information belonging to a specific range, and may be configured as an integrated database in which at least one independent sub-databases are integrated into one. When composed of at least one independent sub-database, each sub-database may be wirelessly connected through Bluetooth, WiFi, and the like, and may exchange data with each other through a network. When the database 130 is configured as an integrated database, the database 130 may include a control unit for integrating respective sub-databases into one and managing data exchange and control flow between them.

FIG. 2 is a block diagram illustrating the topic modeling apparatus in FIG. 1.

Referring to FIG. 2, the topic modeling apparatus 110 may include a local topic information generator 210, a global topic information generator 230, a document topic assigner 250, and a controller 270.

The local topic information generator 210 may generate a plurality of regional document sets by dividing a global document set including a plurality of documents, and perform regional topic modeling to generate local topic information. Here, the number of local document sets to be divided may be determined by various criteria such as the total number of documents, the agency managing the documents, and the site. In addition, the topic and timing of each regional document set may be the same or different.

In one embodiment, the regional topic information generation unit 210 performs regional topic modeling to extract at least one major regional topic for each of a plurality of regional document sets, and includes a local document and at least one major topic belonging to the regional document set. You can create a local topic matrix between local topics. Regional topic modeling means topic modeling performed on a local document set, and general topic modeling techniques may be applied. The regional topic matrix may correspond to a 2-dimensional matrix of document weights of each regional document and a main regional topic included in the regional document set.

The global topic information generation unit 230 may generate global topic information by generating a reduced global document set through a document in each of a plurality of local document sets and performing global topic modeling. More specifically, the global topic information generator 230 extracts some of the documents included in the local document set for each of the plurality of local document sets divided by the local topic information generator 210, and then merges them into one to reduce them. Generated global document sets. The global topic information generator 230 may perform global topic modeling on the reduced global document set, and global topic modeling may correspond to a topic modeling performed on the global document set as a general topic modeling.

In one embodiment, the global topic information generation unit 230 may generate a reduced global document set by extracting at least one local representative document in the order of the weight of each topic for each of the plurality of local document sets in order. . More specifically, the global topic information generator 230 may perform descending sorting on documents included in a local document set according to weights for each topic, and documents within a specific ranking among documents having a higher weight for each topic. Can be extracted as a regional representative document.

The topic modeling apparatus 110 may preset specific ranking information necessary for extracting the local representative document so that the global topic information generator 230 may be used to extract the local representative document. In another embodiment, the topic modeling apparatus 110 may generate a reduced global document set by extracting a local representative document through simple random sampling on the documents included in the local document set.

In one embodiment, the global topic information generation unit 230 performs global topic modeling to extract at least one main global topic for the reduced global document set, and includes at least one local representative document belonging to the reduced global document set. You can create global topic matrices between major global topics. The global topic matrix may correspond to a 2-dimensional matrix representing document weights of respective regional representative documents and major global topics included in the reduced global document set.

The document topic locator 250 may assign at least one topic transformation information generated through document mapping between local and global topic information to each of the plurality of documents. Here, the document mapping may correspond to linking the local topic information and the global topic information through a local representative document, which is a document involved in both local topic information and global topic information generation. The topic transformation information is a two-dimensional matrix generated as a result of transforming local topic information through document mapping and may correspond to a weight transformation matrix between local topic information and global topic information.

In one embodiment, document topic assignment 250 extracts topic component information associated with the regional representative document from the regional topic matrix generated through regional topic modeling and between the topic component information and the global topic matrix generated through global topic modeling. At least one topic transformation matrix may be generated through a multiplication operation. Here, the topic component information is a two-dimensional matrix generated by extracting the parts corresponding to the local representative documents from the local topic matrix generated through the regional topic modeling and integrating them into one, which corresponds to the weight matrix between the main regional topics and the local representative documents. Can be. The topic transformation matrix may be included in the topic transformation information and may correspond to a weight transformation matrix between local topic information and global topic information. In another embodiment, the document topic assignment 250 may generate a topic transformation matrix by calculating the similarity between the regional and global topic information associated with the regional representative document.

In one embodiment, the document topic assignment 250 may assign global topic weights to a plurality of documents by multiplying each topic topic matrix with at least one topic transformation matrix. The document topic assignment 250 can generate a weight matrix between key global topics extracted by global topic modeling for all local documents and a reduced set of global documents by assigning document weights for key global topics for all local documents. have. As a result, the document topic assignment 250 can achieve similar results to performing global topic modeling for all local documents.

The controller 270 controls the overall operation of the topic modeling device 110, and controls a control flow or data flow between the local topic information generator 210, the global topic information generator 230, and the document topic distributor 250. Can manage

3 is a flowchart illustrating a topic modeling process performed in the topic modeling apparatus of FIG. 1.

Referring to FIG. 3, the topic modeling apparatus 110 generates a plurality of regional document sets by dividing a global document set including a plurality of documents through the regional topic information generation unit 210, and performs local topic modeling. Topic information may be generated (step S310).

The topic modeling apparatus 110 may generate global topic information by generating a reduced global document set and global topic modeling through documents in each of a plurality of local document sets through the global topic information generator 230. (Step S330).

The topic modeling apparatus 110 may allocate at least one topic transformation information generated through document mapping between local and global topic information to each of the plurality of documents through the document topic assignment unit 250 (step S350).

4 is a diagram showing an overall overview of a topic modeling system of a document according to an embodiment of the present invention.

Referring to FIG. 4, the topic modeling apparatus 110 divides a global set, which is a document set collected for analysis, into sub-local clusters (Local Set, ex. L _A , L _B ) (410). In operation 420, the main topics of the regional communities are extracted from these. Next, some documents are randomly selected from each local community, and designated as regional representative documents (Delegate), and these are combined to generate a reduced global set that can represent the characteristics of all documents (440). Then, by extracting a global topic from the reduced global set (450) and comparing it with local topic information of the local representative document (460), a rule for extracting a component of the global topic from the local topic is derived (470). Finally, this rule is applied to the local topic weights of each document to assign a global topic for each document (480).

FIG. 5 is an exemplary diagram illustrating an embodiment of regional topic information generated through regional topic modeling performed by the regional topic information generator of FIG. 2.

Referring to FIG. 5, the topic modeling apparatus 110 may perform local topic modeling on a local document set generated by dividing an entire document through the regional topic information generator 210. In FIG. 5, the entire document is divided into two local document sets. However, the topic modeling apparatus 110 may operate in the same manner with respect to the plurality of local document sets.

The regional topic information generation unit 210 performs regional topic modeling on each of two regional document sets 510 and 530 to generate the main regional topics 511 and 531 and the local topic matrix 513 and 533 as regional topic information. Each can be created. In FIG. 5, it is assumed that the subjects of two local document sets 510 and 530 are different, and each local document set Local A 510 and Local B 530 are composed of political and life / culture related documents, respectively. . The regional topic matrices 513 and 533 may correspond to weight matrices between a plurality of regional documents belonging to each regional document set 510 and 530 and main regional topics 511 and 531.

FIG. 6 is an exemplary diagram illustrating an embodiment of global topic information generated through global topic modeling of a reduced global document set performed by the global topic information generator of FIG. 2.

Referring to FIG. 6, the topic modeling apparatus 110 may generate a reduced global document set (RGS) by integrating local representative documents extracted from each local document set into one through the global topic information generator 230. . The global topic information generator 230 may generate a global topic model 610 and a global topic matrix 630 as global topic information by performing global topic modeling on the reduced global document set.

In one embodiment, the global topic information generator 230 may extract the local representative document from each local document set through a simple random extraction method. In another embodiment, the global topic information generation unit 230 may sort a plurality of regional documents belonging to respective regional document sets in a weighted order per topic, and localize only the upper specific number of local documents according to the sorted order. Can be extracted as a representative document.

The global topic information generator 230 is a local representative document of the local document set Local A 510 and Local B 530 in FIG. 5, respectively. 1, 3 and 5 and Doc. 10, 12 and 14 can be extracted. The global topic matrix 630 may represent a correspondence between the regional representative document and the main global topic 610.

FIG. 7 is an exemplary diagram illustrating an embodiment of a process of generating topic transformation information through document mapping performed in the document topic assignment in FIG. 2.

Referring to FIG. 7, the topic modeling apparatus 110 may generate topic transformation information through a document mapping between local topic information and global topic information through the document topic assignment 250. The document topic assignment unit 250 may extract topic component information 710 associated with the regional representative document from the local topic matrix generated through the regional topic modeling, and may be generated through the topic component information 710 and the global topic modeling. At least one topic transformation matrix 750 may be generated through a multiplication operation between the global topic matrices 730. The topic component information 710 may correspond to a two-dimensional weight matrix between main regional topics and regional representative documents.

In Fig. 7, “Topic j” (major local topic j) of “Local i” (local document set i) is L _i _T _j , and “Topic k” (major global topic k) of the reduced global document set (RGS). ) Is _denoted by RGS_T _k .

In the case of the regional representative document, since it participates in the global topic modeling in FIG. 6 as well as the regional topic modeling in FIG. For example, Doc. 1 can have a document weight of (0.013, 0.009, 0.048, 0.022, 0.021) for five major regional topics from L _A _T ₁ to L _A _T ₅ as a result of regional topic modeling, while at the same time As a result, it may have a document weight of (0.312, 0.024, 0.003, 0.004, 0.050) for five major global topics of RGS_T ₁ to RGS_T ₅ .

7 is L _A _T ₁ to L _A _T ₅ from ₁ to rules for predicting the value of RGS_T ₅ (Rule A) RGS_T and L _B _T ₁ to L _B _T ₅ from the rules for predicting the value of RGS_T ₁ to RGS_T ₅ The derivation process for (Rule B) is shown. The document topic assignment 250 can extract only the portion 710 corresponding to the regional representative document from the local topic matrix resulting from each regional topic modeling, and then global topic modeling for the reduced global document set (RGS). It can be compared with the global topic matrix 730 derived through.

The document topic locator 250 may generate a topic transformation matrix 750 from local document sets Local A and Local B. For example, Figure 7 (a) shows a process of deriving a weight of from ₁ RGS_T L _A L _A _T _T ₁ ~ _{5. A} value of '0.021', indicated by dotted rectangles, in the topic transformation matrix 750 indicates that an increase of '1' of the value of L _A _T ₂ results in an increase of '0.021' of the value of RGS_T ₁ . In other words, RGS_T ₁ receives a lot of influence of L _A _T _2, compared to other major regional topics, these results are representative of local documents Doc. 1, Doc. 3, Doc. The higher the value of L _A _ T ₂ in 5, the higher the value of RGS_T ₁ reflects the phenomenon.

FIG. 8 is an exemplary diagram illustrating an embodiment of a process of allocating topic transformation information to each of a plurality of documents in the document topic assignment in FIG. 2.

Referring to FIG. 8, the topic modeling apparatus 110 may assign the topic transformation information to each of a plurality of documents included in the global document set through the document topic assignment 250. Here, the topic transformation information may correspond to at least one topic transformation matrix 830 corresponding to the plurality of local topic matrices 810. Also, in FIG. 8A, the local topic matrix 810 may correspond to the local topic matrix 513 of FIG. 5, and the topic transformation matrix 830 may correspond to the topic transformation matrix 750 of FIG. 7. Can be.

Table 1 below shows some notation.

TABLE 1

Using the notation in Table 1 above, the local document Doc. A rule for deriving the global topic weight from the local topic weight of i is defined as follows (where N is the total number of local topics in Local A).

For example, Doc that belongs to the local document set Local A. The document weights for global topic RGS_T ₁ of 2 are as follows.

d ₂ (RGS_T ₁ ) = (0.006 * 0.020) + (0.021 * 0.062) + (0.016 * 0.030) + (0.007 * 0.022) + (0.011 * 0.045) = 0.003

Doc. The document weight for global topic RGS_T ₁ of 2 is calculated as 0.003. The document topic assignment 250 may assign global topic weights to the local representative document as well as all local documents not included in the representative in the same manner, and the two-dimensional weight matrix 850 between each regional document and the global topic weights. Can be generated as a result.

Although described above with reference to a preferred embodiment of the present invention, those skilled in the art will be variously modified and changed within the scope of the invention without departing from the spirit and scope of the invention described in the claims below I can understand that you can.

100: document modeling system
110: topic modeling device 130: database
210: local topic information generation unit 230: global topic information generation unit
250: document topic assignment 270: control unit
510: Local Document Set Local A 511: Main Regional Topics
513: Local Topic Matrix 530: Local Document Set Local B
531: Major Regional Topics 533: Regional Topic Matrix
610: main global topic 630: global topic matrix
710: topic component information 730: global topic matrix
750: topic transformation matrix 810: local topic matrix
830: topic transformation matrix
850: Global topic weighting matrix for geographic documents

Claims (13)

A local topic information generation unit for generating a plurality of local document sets by dividing a global document set including a plurality of documents, and generating local topic information by performing local topic modeling;
A global topic information generator configured to generate a global document set reduced through a representative document in each of the plurality of local document sets and perform global topic modeling to generate global topic information; And
And a document topic allocation unit for allocating the topic transformation information generated through the document mapping between the local and global topic information to each of the plurality of documents.
The method of claim 1, wherein the local topic information generation unit
Extracting at least one major regional topic for each of the plurality of regional document sets by generating the regional topic modeling, and generating a regional topic matrix between the local document belonging to the regional document set and the at least one major regional topic Topic modeling device for documents.
The method of claim 1, wherein the global topic information generation unit
The topic modeling apparatus of claim 1, wherein the reduced global document set is generated by extracting at least one local representative document in the order of the weight of each topic for each of the plurality of local document sets.
The method of claim 1, wherein the global topic information generation unit
Extracting at least one main global topic for the reduced global document set by performing the global topic modeling, and generating a global topic matrix between a local representative document belonging to the reduced global document set and the at least one major global topic Topic modeling apparatus of a document, characterized in that.
The method of claim 1, wherein the document topic allocation
At least one topic transformation matrix by extracting topic component information associated with a local representative document from a local topic matrix generated through the regional topic modeling and multiplying the topic component information and a global topic matrix generated through the global topic modeling Topic modeling apparatus of a document, characterized in that for generating.
The method of claim 5, wherein the document topic allocation
And assigning a global topic weight to the plurality of documents by multiplying each of the local topic matrices with the at least one topic transformation matrix.
In the topic modeling method performed in the topic modeling apparatus of a document,
(a) generating a plurality of local document sets by dividing a global document set including a plurality of documents, and performing local topic modeling to generate local topic information;
generating global topic information by generating a reduced global document set through representative documents in each of the plurality of local document sets and performing global topic modeling; And
(c) assigning to each of the plurality of documents topic translation information generated through document mapping between the local and global topic information.
The method of claim 7, wherein step (a)
Extracting at least one major regional topic for each of the plurality of regional document sets by generating the regional topic modeling, and generating a regional topic matrix between the local document belonging to the regional document set and the at least one major regional topic; Topic modeling method of a document, characterized in that.
The method of claim 7, wherein step (b)
And extracting at least one regional representative document in order of increasing weight for each topic for each of the plurality of regional document sets to generate the reduced global document set.
The method of claim 7, wherein step (b)
Extracting at least one main global topic for the reduced global document set by performing the global topic modeling, and generating a global topic matrix between a local representative document belonging to the reduced global document set and the at least one major global topic A topic modeling method for a document, characterized in that the step.
The method of claim 7, wherein step (c)
At least one topic transformation matrix by extracting topic component information associated with a local representative document from a local topic matrix generated through the regional topic modeling and multiplying the topic component information and a global topic matrix generated through the global topic modeling The topic modeling method of the document, characterized in that for generating a step.
The method of claim 11, wherein step (c)
And assigning a global topic weight to the plurality of documents by multiplying each of the local topic matrices with the at least one topic transformation matrix.
A computer-readable recording medium for recording a topic modeling method performed by a topic modeling apparatus of a document,
Dividing a global document set including a plurality of documents to generate a plurality of local document sets and performing local topic modeling to generate local topic information;
Generating global topic information by generating a reduced global document set through representative documents in each of the plurality of local document sets and performing global topic modeling; And
And assigning to each of the plurality of documents topic translation information generated through document mapping between the local and global topic information.

Images