JPH11328211A - Automatic sorting method and device for information of large quantity and recording medium recording automatic sorting program of the information of large quantity - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an automatic sorting method/device for the information of a large quantity to shorten a calculation time by previously dividing real categories desired to be sorted into several sets of intermediate categories and calculating a distance between the real and the intermediate categories only when the condition is satisfied for the intermediate category and also to provide a recording medium which records an automatic sorting program of the information of a large quantity. SOLUTION: Categories and their features are inputted for sorting documents and the feature vectors are calculated from those inputted categories and features and then stored. An intermediate category calculation part 107 produces new categories by using the inputted categories and features and also an inputted sorting standard. Then, the similar categories are assigned to the same intermediate category among those inputted categories and the inputted documents are sorted by using the inputted categories and the categories which are acquired at the part 107.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and an apparatus for automatically classifying a large amount of information such as document data into a large number of categories at high speed, and a recording medium on which a program for automatically classifying a large amount of information is recorded.

[0002]

2. Description of the Related Art In recent years, it has become possible to exchange a large amount of electronic documents through a computer network such as the Internet. For this reason, services that allow individuals to search for information they need are implemented on networks. However, the amount of information acquired by the user becomes large, and it becomes difficult to extract characteristics of individual information. Therefore, a technique for classifying and organizing the acquired information is required.

Conventionally, research has been conducted on a method of automatically classifying document information. A typical method is to classify information into categories that are considered appropriate for new information, with known breaks (called categories) for classifying like a library (" Automatic classification of text "Yamamoto, Masuyama (Toyohashi University of Technology), Naito (NTT), 1995), and the category to classify is unknown,
A method in which similar documents are collected from a set of documents to create and assign classification categories ("Automatic segmentation by competitive learning neural network" Kikuchi, Matsuoka et al. (Utsunomiya Univ.)
Others) and 1995). With these technologies, a large number of documents are classified and arranged.

[0004] The classification method to which the present invention is directed is a method in which a category for classification is known in advance. Given a category for pre-classification and a sample document or word to be included in the system, the system calculates the importance of the word from the information, and as a feature of the category, the word and the importance for the category. Produces a pair of vectors. Similarly, for documents to be classified,
Calculate importance for words and documents and generate vectors ("Automatic Text Processing" Gerard Salton, 198
9).

[0005]

(Equation 1) Becomes Incidentally, N represents the number of dimensions. Similarly, a feature vector is generated for a document to be classified.

The classification defines a distance between a feature vector of a category and a feature vector of a document, and assigns each document to a similar category using the value. If it is determined that the distance is very large, that is, it is not similar to any category, it is not assigned to any category.

[0007]

In the above-mentioned conventional method, there is a problem that as the total number of categories to be classified or the number of appearing words increases, the calculation time increases in proportion to the number.

[0008] The present invention has been made in view of the above,
The purpose is to divide the real category to be classified into some intermediate categories in advance, and to calculate the distance from the real category only when the conditions of the intermediate category are satisfied, thereby shortening the calculation time. It is an object of the present invention to provide a method and apparatus for automatically classifying information and a recording medium on which a program for automatically classifying large amounts of information is recorded.

[0009]

According to one aspect of the present invention, there is provided a method for automatically classifying a large amount of information into a large number of categories. A first step of inputting a document, a second step of storing the document input in the first step, a third step of inputting a category for classifying the document and its characteristics, A fourth step of storing the category and the feature input in the third step and calculating and storing a feature vector from the category and the feature, and a fifth step of inputting a classification criterion which is an initial value for the classification process. Step, the sixth step of storing the classification criterion input in the fifth step, the category and its characteristics input in the third step, and the category input in the fifth step. A seventh step of newly creating a category using the classification criteria and assigning similar categories input in the third step to the same intermediate category; and a category input in the third step. And an eighth step of classifying the document input in the first step using the category obtained in the seventh step.

According to the first aspect of the present invention, a category for classifying a document and its features are input, a feature vector is calculated from the input categories and features, stored, and the input category is stored. And create a new category using the features and the input classification criteria,
A similar input category is assigned to the same intermediate category, and the input document is classified using the created category and the input category. Therefore, the documents can be classified at high speed.

The present invention described in claim 2 is the same as the claim 1.
In the invention described in the above, in the seventh step, a distance function according to the classification criterion is created by combining distance functions between a plurality of categories based on the classification criterion input in the fifth step, and this created The gist is to calculate the distance between the categories input in the third step using the distance function to generate an intermediate category.

According to the second aspect of the present invention, a distance function according to the classification criterion is created by combining distance functions between a plurality of categories based on the input classification criterion, and the created distance function is created. Is used to calculate the distance between the input categories to generate intermediate categories.

Further, the present invention according to claim 3 provides the invention according to claim 1.
In the invention according to the second aspect, in the eighth step, classification into the intermediate category generated in the seventh step is performed, and a third step is performed based on the classification result and the classification criterion input in the fifth step. The point is to determine the necessity of classification into the category input in.

According to the third aspect of the present invention, the necessity of classification into the input category is determined based on the generated classification result into the intermediate category and the input classification standard.

According to a fourth aspect of the present invention, there is provided an automatic mass information classifying apparatus for classifying a large amount of information into a large number of categories, comprising: a document input unit for inputting a document described in a natural language; A document storage unit that stores the document input in step S, a category information input unit that inputs a category for classifying the document and its characteristics, and stores the category and characteristic input in the category information input unit. A category information storage unit that calculates and stores a feature vector from a category and a feature, an initial value input unit that inputs a classification criterion that is an initial value for a classification process, and a classification criterion input by the initial value input unit. An initial value storage unit for storing;
Using the category and its characteristics input in the category information input unit and the classification criterion input in the initial value input unit, a new category is created, and the category input in the category information input unit is similar. Classifying a document input by the document input unit by using an intermediate category calculation unit that assigns objects to the same intermediate category; and a category input by the category information input unit and a category obtained by the intermediate category calculation unit. And a real category calculation unit that performs

According to a fourth aspect of the present invention, a category for classifying a document and its features are input, a feature vector is calculated from the input categories and the features and stored, and the input categories and Using the characteristics and the input classification criteria, a new category is created in the intermediate category calculation unit, and a similar one of the input categories is assigned to the same intermediate category, and the input category and the intermediate category calculation are performed. Since the input document is classified using the category obtained by the section, the document can be classified at high speed.

The present invention according to claim 5 provides the present invention according to claim 4.
In the described invention, the intermediate category calculation unit combines a distance function between a plurality of categories based on the classification criterion input by the initial value input unit to create a distance function according to the classification criterion, It is essential to have means for calculating the distance between the categories input by the category information input unit using the distance function obtained and generating an intermediate category.

According to the present invention, a distance function corresponding to a classification criterion is created by combining distance functions between a plurality of categories based on the input classification criterion, and the created distance function is created. Is used to calculate the distance between the input categories to generate intermediate categories.

Further, the present invention according to claim 6 provides the invention according to claim 4.
In the invention according to the fifth aspect, the real category calculation unit performs classification into the intermediate categories generated by the intermediate category calculation unit, and based on the classification result and the classification criteria input by the initial value input unit, The gist of the present invention is to have a means for determining whether or not classification into the category input by the information input unit is necessary.

According to the present invention, the necessity of classification into the input category is determined based on the classification result into the intermediate category generated by the intermediate category calculation unit and the input classification standard. doing.

According to a seventh aspect of the present invention, there is provided a recording medium recording a large-volume information automatic classification program for classifying a large amount of information into a large number of categories, and a first step of inputting a document described in a natural language. A second step of storing the document input in the first step, a third step of inputting a category for classifying the document and its characteristics, and a third step of inputting the category. A fourth step of storing a category and a feature and calculating and storing a feature vector from the category and the feature; a fifth step of inputting a classification criterion which is an initial value for a classification process; A sixth step of storing the classification criterion input in the step, a category and its characteristics input in the third step, and the classification criterion input in the fifth step. Te, to create a category to a new, third
A seventh step of assigning similar categories among the categories input in step to the same intermediate category, and a first step using the categories input in the third step and the categories obtained in the seventh step. The gist of the invention is to record a mass information automatic classification program having an eighth step of classifying a document input in the step on a recording medium.

According to the present invention, a category for classifying documents and its features are input, a feature vector is calculated from the input categories and features, stored, and the input category is stored. And create a new category using the features and the input classification criteria,
By assigning similar input categories to the same intermediate category, using the created category and the input category, storing a mass information automatic classification program for classifying the input document in a recording medium, Therefore, the distribution property can be improved by using the recording medium.

The present invention described in claim 8 provides the present invention according to claim 7.
In the invention described in the above, in the seventh step, a distance function according to the classification criterion is created by combining distance functions between a plurality of categories based on the classification criterion input in the fifth step, and this created The gist of the present invention is to record a mass information automatic classification program for calculating a distance between categories input in the third step using a distance function to generate an intermediate category on a recording medium.

According to the present invention, a distance function according to the classification criterion is created by combining distance functions between a plurality of categories based on the input classification criterion, and the created distance function is created. , The mass information automatic classification program for calculating the distance between the input categories and generating the intermediate category is recorded on the recording medium, so that the distribution can be improved by using the recording medium. .

Further, the present invention described in claim 9 is the same as claim 7.
In the eighth aspect, in the eighth step, classification into the intermediate category generated in the seventh step is performed, and a third step is performed based on the classification result and the classification criterion input in the fifth step. The gist is to record a mass information automatic classification program for determining whether or not to classify into the category input in the recording medium on the recording medium.

According to the ninth aspect of the present invention, based on the generated classification result to the intermediate category and the input classification criterion, it is possible to determine whether or not the classification into the input category is necessary. Since the classification program is recorded on the recording medium, it is possible to use the recording medium to improve the distribution.

[0027]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing a configuration of a mass information automatic classification device according to an embodiment of the present invention. The mass information automatic classification device shown in FIG. 1 includes a document input unit 101 for inputting a document described in a natural language, and a document storage unit 10 for storing the document input by the document input unit 101.
2. A category information input unit 103 for inputting a category for classifying the document and its characteristics, storing the categories and characteristics input by the category information input unit 103, and calculating a feature vector from the categories and characteristics. A category information storage unit 104 for storing, an initial value input unit 1 for inputting a classification criterion which is an initial value for classification processing
05, an initial value storage unit 106 for storing the classification criteria input by the initial value input unit 105, and the category information input unit 1
A new category is created using the category and its characteristics input in step 03 and the classification criteria input in the initial value input unit 105, and similar categories are input in the category information input unit 103. An intermediate category calculation unit 107 for allocating to the same intermediate category, using the category input by the category information input unit 103 and the category obtained by the intermediate category calculation unit 107 to convert the document input by the document input unit 101 It comprises a real category calculation unit 108 for classifying and a display unit 109 for displaying the result of classification by the real category calculation unit 108.

Before describing the operation of the thus configured automatic mass information classification apparatus, the basic principle of the automatic mass information classification apparatus of the present embodiment will be described first. The mass information automatic classification device of the present embodiment divides a real category, which is a category to be classified, into some intermediate groups in advance, and calculates a distance from the real category only when a condition in this intermediate category is satisfied. By doing so, the calculation time is reduced.

More specifically, a distance is introduced between categories by using a feature vector of a real category. The distance between each real category is calculated, similar real categories are extracted, and two or more similar categories are assigned to a newly created intermediate category.

As a distance function between categories, an inner product of feature vectors, a co-occurrence relationship of words between categories, and a combination thereof are used. For example, the co-occurrence relation I _ij between words of category i and category j is

(Equation 2) Is defined. The co-occurrence relation I _ij is a function indicating how many words are common to categories i and j.
Also, the degree of overlap R _ij of the words of category i and category j,

(Equation 3) Such functions are also conceivable. The similarity between categories is defined by a combination of values of inner products of feature vectors and values of co-occurrence relations between categories.

Next, the actual classification processing will be described.
To classify a document, first, the distance between the document and the intermediate category is calculated. If it is determined that the distances are more similar to each other, it is determined whether or not the distance to the actual category belonging to the intermediate category is calculated and classified.

The criterion at the time of generating and classifying the intermediate category depends on the classification criterion of the actual category. For example, when a document is classified so as to always belong to one of the real categories, the real category is classified so as to belong to any one of the intermediate categories. In addition, if a document that is not classified anywhere, that is, a document that does not have an appropriate real category, is not to be classified, the judgment of “not classified into any category” can be made without calculating the actual category once, depending on the judgment of the intermediate category. Can be done. In addition, if the co-occurrence relation is used to generate the intermediate category and whether or not a word present in the document to be classified exists in the intermediate category as a criterion, the classification speed can be easily increased without lowering the classification accuracy.

Further, the intermediate category is not limited to one layer. By introducing a distance between the intermediate categories, creating an intermediate category of the intermediate category, and judging whether or not to classify the intermediate category in multiple stages, a higher speed can be achieved. Classification can be performed.

Next, the mass information automatic classification device shown in FIG. 1 will be described. In FIG. 1, a document to be subjected to classification processing is input from the document input unit 101. Documents to be subjected to this classification processing include all documents input to the computer. For example, newspaper articles, HTML files on the Internet, net news, teletext, F
There are M multiplex broadcasting and broadcast manuscripts on television. Further, the document storage unit 102 stores the information input in the document input unit 101 so that the document of each information transmission medium, some of the information transmission media, or all the information transmission media can be retrieved.

In the category information input section 103, the system user inputs a category to be classified and sample document information and keywords included in the category. For example, to collect information about sports soccer, a category related to “soccer” and newspaper articles and documents and keywords related to soccer collected from the Internet are input.

The category information storage unit 104 stores information input by the category information input unit 103, calculates a feature vector from the input information, and stores the result.

The initial value input unit 105 inputs an initial value when the system performs a calculation. For example, a classification error criterion or a classification condition such as "classify all documents into one similar real category" or "classify one document into a plurality of real categories" is input.

The intermediate category calculation unit 107 creates an intermediate category from the information input by the category information input unit 103 and the initial value input unit 105. A feature vector is generated from the information of the real category input by the category information input unit 103, and a distance function between the real categories is determined using the information input by the initial value input unit 105 to generate an intermediate category. At the same time, the criterion for determining whether or not to classify the document into the actual category belonging to the intermediate category is determined.

The real category calculation unit 108 performs a process of assigning a document input by the document input unit 101 to a real category. First, the distance between the intermediate category and the document is calculated by using the intermediate category and the judgment criterion, which are the processing results of the intermediate category calculation unit 107. Using the result and the criterion for the intermediate category determined by the intermediate category calculation unit 107, it is determined whether to calculate the distance between the actual category belonging to the intermediate category and the document. If you need to calculate
The distance between the document and the actual category is actually calculated, and the document is assigned to the actual category according to the classification criterion input by the initial value input unit 105.

Next, the operation of the present embodiment will be described in detail with reference to FIGS. A document set to be input to the system is a newspaper article of a newspaper company in 1998, and this document set is input from the document input unit 101.

Next, from the category information input unit 103, a category to be collected and a sample document or word included therein are input. For example, input "category category to collect articles about" Traffic accidents "", "category category to collect articles about" electric vehicles "" and "classification category to collect articles about" Earthquake "". As a document, some articles about past "traffic accidents" are classified into "category categories to collect articles about" traffic accidents ", and some articles about" electric cars "are classified into" category categories to collect articles about "electric cars". , Input that some articles relating to “earthquake” are to be assigned to “a classification category for collecting articles relating to“ earthquake ””.

The category information storage unit 104 creates and stores a feature vector of each category from the information input by the category information input unit 103. For example, in the "category category that collects articles about" traffic accidents "", a feature vector is obtained from the input sample document as a feature vector.

((Car, 0.23), (speed, 0.13), (night, 0.08), (dead, 0.05),...) (5) are generated and stored. In the “category category for collecting articles related to“ electric vehicles ”,” a feature vector is obtained from the input sample document as a feature vector.

((Car, 0.20), (electricity, 0.12), (fuel consumption, 0.10), (speed, 0.04),...) (6) are generated and stored. In the “category category that collects articles about“ earthquake ””, a feature vector is obtained from the input sample document as a feature vector.

[Formula 6] ((earthquake, 0.15), (seismic intensity, 0.10), (dead, 0.07), (tsunami, 0.04),...) (7) are generated and stored.

Next, an initial value for classification is input from the initial value input unit 105. Here, it is assumed that a condition of “assignment to any one of the actual categories” and a condition of “permit that there is a document that cannot be assigned to any of the actual categories” are given.

From the above input processing, first, the processing of the intermediate category calculation unit 107 is performed. For example, “Category category that collects articles about“ traffic accidents ”” and “Electric vehicles”
When the word overlapping concrete R _ij is used for the classification category that collects articles about “earthquake” and the classification category that collects articles about “earthquake”, similar categories “classification category that collects articles about“ traffic accident ”” and “ A classification category for collecting articles related to “electric vehicles” is assigned to a certain intermediate category. In this case, the generated intermediate category name will be referred to as an “intermediate category of“ car ”” to make the story easier to understand.
The feature vector of "intermediate category of" automobile ""
When generated as “words that are commonly present in the feature vectors of the real categories in the intermediate category”, for example,

## EQU00007 ## A feature vector of ((automobile, 0.215), (speed, 0.065),...) (8) is generated. The weight of each word was the average value of the original feature vector. The judgment criterion is “whether any word existing in the document exists in any of the feature vectors of the intermediate category”.

Next, the actual category calculation unit 108 performs a classification process of the documents into the actual categories. For example, assume that a newspaper article in 1998 includes an article “A car collides with a guardrail due to excessive speed” as shown in FIG. The feature vector for this article is

(8) ((collision, 0.25), (automobile, 0.19), (guardrail, 0.12), (speed, 0.10),...) (9) This feature vector is compared with the word in the document, “the middle category of“ automobile ””. in this case,
Since the words "automobile" and "speed" exist in both the feature vector of the article and that of the intermediate category, the criteria for the intermediate category are satisfied. For this reason, the actual categories "category category for collecting articles on" traffic accidents "" and "category category for collecting articles on" electric vehicles ""
And finally assign it to the “category category that collects articles about“ traffic accidents ””.

Also, as shown in FIG. 3, it is assumed that there is an article "A high school wins in a high school soccer game." The feature vector for this article is

(9) ((high school student, 0.18), (soccer, 0.15), (winning, 0.10), (A high school, 0.09),...) (10) In this case, when compared with the feature vector of the “intermediate category of“ automobile ””, no common word exists and the criterion is not satisfied. For this reason, the actual categories “classification category for collecting articles related to“ traffic accident ”” and “classification category for collecting articles related to“ electric vehicle ”” are not classified.

The display unit 109 displays the classification result.
Documents that have been classified into real categories or not,
Information on the generated intermediate category, information on the feature vector in the actual category, information on the time required for classification, and the like are displayed.

[0048]

As described above, according to the present invention,
A category for classifying a document and its features are input, a feature vector is calculated and stored from the input categories and features, and the category is calculated using the input category and its features and the input classification criteria. Is newly created, and similar input categories are assigned to the same intermediate category, and the input documents are classified using the generated categories and the input categories. Can be quickly classified into a large number of classification categories.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a mass information automatic classification device according to an embodiment of the present invention.

FIG. 2 is a diagram for explaining a specific example of the mass information automatic classification device shown in FIG. 1;

FIG. 3 is a diagram for explaining a specific example of the mass information automatic classification device shown in FIG. 1;

[Explanation of symbols]

 Reference Signs List 101 Document input unit 102 Document storage unit 103 Category information input unit 104 Category information storage unit 105 Initial value input unit 106 Initial value storage unit 107 Intermediate category calculation unit 108 Real category calculation unit 109 Display unit

 ────────────────────────────────────────────────── ─── Continued on the front page (72) Inventor Kazuo Tanaka Nippon Telegraph and Telephone Corporation 3-19-2 Nishishinjuku, Shinjuku-ku, Tokyo

Claims (9)

[Claims] 1. A mass information automatic classification method for classifying a large amount of information into a large number of categories, comprising: a first step of inputting a document described in a natural language; and a document input in the first step. A third step of inputting a category for classifying the document and its characteristics, and storing the category and characteristics input in the third step and from the categories and characteristics. A fourth step of calculating and storing the feature vector; and a fifth step of inputting a classification criterion which is an initial value for the classification process.
And a sixth step of storing the classification criterion input in the fifth step.
And a new category is created using the category and its characteristics input in the third step and the classification criterion input in the fifth step, and a similarity is created in the category input in the third step. Classifying the document input in the first step using the seventh step of allocating the extracted data to the same intermediate category, and the category input in the third step and the category obtained in the seventh step. An automatic mass information classification method, comprising the steps of: 2. In the seventh step, a distance function corresponding to a classification criterion is created by combining distance functions between a plurality of categories based on the classification criterion input in the fifth step, and the created distance function is created. 2. The method according to claim 1, wherein a distance between the categories input in the third step is calculated using a distance function to generate an intermediate category. 3. In the eighth step, classification into the intermediate category generated in the seventh step is performed, and input is performed in the third step based on the classification result and the classification criterion input in the fifth step. 3. The method according to claim 1, wherein it is determined whether or not the classification into the determined category is necessary. 4. A mass information automatic classification apparatus for classifying a large amount of information into a large number of categories, comprising: a document input unit for inputting a document described in a natural language; and a document input by the document input unit. A document storage unit, a category information input unit for inputting a category for classifying the document and its features, and a category and feature input by the category information input unit, and a feature vector from the categories and features. A category information storage unit for calculating and storing; an initial value input unit for inputting a classification criterion which is an initial value for classification processing; an initial value storage unit for storing the classification criterion input in the initial value input unit A new category is created by using the category and its characteristics input in the category information input unit and the classification criteria input in the initial value input unit, An intermediate category calculator that allocates similar categories to the same intermediate category in the category input in the information input unit, using the category input in the category information input unit and the category obtained in the intermediate category calculator. And a real category calculating unit for classifying the document input by the document input unit. 5. The intermediate category calculation unit creates a distance function according to the classification criterion by combining distance functions between a plurality of categories based on the classification criterion input by the initial value input unit. The mass information automatic classification device according to claim 4, further comprising: means for calculating a distance between the categories input by the category information input unit using the obtained distance function to generate an intermediate category. 6. The real category calculation unit performs classification into the intermediate categories generated by the intermediate category calculation unit, and inputs the category information based on the classification result and the classification criterion input by the initial value input unit. 6. The mass information automatic classification device according to claim 4, further comprising means for determining whether or not the classification into the category input by the section is necessary. 7. A recording medium recording a mass information automatic classification program for classifying a large amount of information into a large number of categories, the first step of inputting a document described in a natural language, and the first step A second step of storing the document input in step 3; a third step of inputting a category for classifying the document and its characteristics; and storing the category and characteristic input in the third step. A fourth step of calculating and storing a feature vector from the category and the feature, and a fifth step of inputting a classification criterion which is an initial value for a classification process.
And a sixth step of storing the classification criterion input in the fifth step.
And a new category is created using the category and its characteristics input in the third step and the classification criterion input in the fifth step, and a similarity is created in the category input in the third step. Classifying the document input in the first step using the seventh step of allocating the extracted data to the same intermediate category, and the category input in the third step and the category obtained in the seventh step. A recording medium having recorded thereon a mass information automatic classification program, comprising: an eighth step. 8. In the seventh step, a distance function according to the classification criterion is created by combining distance functions between a plurality of categories based on the classification criterion input in the fifth step, and the created distance function is created. 8. The recording medium according to claim 7, wherein an intermediate category is generated by calculating a distance between the categories input in the third step using a distance function. 9. In the eighth step, classification into the intermediate category generated in the seventh step is performed, and the classification is performed in the third step based on the classification result and the classification criterion input in the fifth step. 9. A recording medium recording a mass information automatic classification program according to claim 7 or 8, wherein it is determined whether or not the classification into the determined category is necessary.