JP2001222545A - Similar document search device, similar document search method, and recording medium - Google Patents

Abstract

(57) [Summary] [Problem] To enable similarity document extraction to be performed with high accuracy even if the expression of a word of consent changes by going back to the category in the past. SOLUTION: When classifications such as a new category, an old category [1], and an old category [2] are performed in chronological order, link information between categories and link information of words whose expressions change are stored in a memory. Keep it. Then, after specifying the category to which the document given as the search key belongs,
Based on the information stored in the memory, the old category [1] and the old category [2] related to the specific category are searched, and a document similar to the search key document is extracted using the word link information. Do. As a result, similar documents can be extracted with high accuracy even if the search is performed retroactively in the past category.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a similar document retrieval apparatus for retrieving a similar document from a database,
In particular, the present invention relates to a similar document search device suitable for searching for a similar document targeting documents classified by category, a similar document search method and a recording medium used in this device.

[0002]

2. Description of the Related Art Conventionally, there is a system in which a large number of various documents are stored in a database, and a document similar to a designated document (hereinafter, referred to as a search key document) is automatically searched from the database. In such a system, a word included in a search key document is compared with a word included in a search target document, and a common word type, an appearance location, an appearance frequency, and the like are determined by a vector space method or the like. A similarity is calculated, and a document having a high similarity is output as a search result.

In this type of system, various documents in a database are classified into predetermined categories, a category of a document to be searched is specified, and then a document belonging to the category is searched for a similar document. A search is performed. In this case, the number and definition of categories are not always fixed but are updated according to the year or the like. Accordingly, the present inventors have developed a technique for linking between an updated category and a related category before updating, and using a document included in a plurality of linked categories as a search target document during search. Japanese Patent Application No. 11-328330
issue).

[0004]

The conventional technique is effective in that documents included in a plurality of linked categories before and after updating can be extracted as documents to be searched. However, in many cases, the wording of the word changes as the search target document for calculating the similarity to the search key document is linked back in time to the past category. Since the similarity was calculated from the words included in the search key document and the words included in the search target documents in the past, the accuracy of the extracted similar documents decreases as going back in the past. was there.

The present invention has been made to solve such a problem, and a similar document search apparatus for calculating similarity and improving the accuracy of similar document search when a database whose categories are updated in a time series is used. An object of the present invention is to provide a similar document search method used in this device.

[0006]

Means for Solving the Problems In order to achieve the above object, a similar document search device according to the present invention includes: a database storing a plurality of documents classified into predetermined categories;
Category link information for associating a document belonging to the first category indicating the classification used in the first period with a document belonging to the second category indicating the classification used in the second period in a unit of change of the category A first storage unit that stores word link information for associating a word in which a different expression is used while agreeing between the first category and the second category, and a search key Category specifying means for specifying the category to which the document input as the target belongs for the first category, and the second category related to the first category specified by the category specifying means to the first category. Determining means for determining from the link information stored in the storage means, and the first category and the second category obtained by the determining means A search unit for searching a document belonging to the search key document from the database using the word link information stored in the second storage unit, and a similarity obtained by the search unit; Output means for outputting a document as a search result for the search key document. With such a configuration, when similarity calculation and similar document search are performed using a database in which the categories are updated in a time series, documents in a plurality of linked categories can be easily searched, and the word Even if the wording changes, these words are associated with each other by the word link information, so that similar documents can be searched for documents of all target categories with high accuracy.

[0007]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Before describing embodiments of the present invention, an outline of a similar document search apparatus of the present invention will be described for easy understanding.

The similar document search apparatus of the present invention, when retrieving a similar document from a database in which a plurality of documents classified for each predetermined category is registered, changes the category into a unit of category change and the changed category. The related categories are sequentially searched using link information that associates with the previous category, and a change in the expression of a word between the linked categories is also linked. In other words, if the categories are integrated at a certain point in time, the category before the integration is also searched, and if the category is split, the categories before the split from the splitting destination are also searched. Further, if a plurality of categories are reconstructed at a certain point in time, the corresponding documents are narrowed down from the category before the reconstruction using a search formula or the like, and are set as search targets. Another characteristic is that the expressions of words that change in time series between these categories are also linked.

For example, as shown in FIG.
It is assumed that classifications such as the old category [1] and the old category [2] are performed in chronological order. The new category is the latest category, and here has categories [11] to [15]. The old category [1] is the previous category, and here has categories [21] to [25]. Further, the old category [2] is the category of the previous one (the category of the previous two), and here, the categories [31] to [3]
3] and [25].

In this example, category [11] and category [12] are the same category [2] in old category [1].
1]. Furthermore, this category [21]
Are integrated into category [31] together with category [22] in old category [2]. Similarly, for the other categories [13] to [15], the old category [1],
In [2], they are integrated into the same category or divided into different categories.

As a specific example, suppose that category [11] in the new category indicates a field such as "soft tennis" and category [12] indicates a field such as "rigid tennis". ], And integrated in fields such as "tennis", and the old category [2]
In this case, category [31] has been integrated in a field such as "sports".

In another example, if the category [14] in the new category indicates a field such as "traffic", in the old category [2], the category [23] indicates "automobile" and the category [24]. Was divided into fields such as "bicycles". Here, in the age of this new category in the category [14], an item often expressed by the word “4WD” is the category [2].
[4] In the age of the old category [1] with [3]
Is often expressed by the word In such a case, since it is clear that both words point to the same item, both words are linked and treated as the same word.

By doing so, even if the category is updated, a document similar to the search key document can be extracted from the old category [1] and the old category [2] linked to the new category. Similarity calculation and similar document search accuracy can be improved.

An embodiment of the present invention for realizing such a similar document search will be described below.

FIG. 2 is a diagram showing a hardware configuration of a similar document search apparatus according to one embodiment of the present invention. This device is constructed as one function on a computer having a general architecture.

As shown in FIG. 2, the present apparatus includes a control device 1, an input device 2 including a keyboard, a pointing device, and a scanner, a display device 3 for displaying similar document search results, and an external storage device 4. Be composed. The control device 1 includes a control unit 1a including a CPU and a memory unit 1b, and controls the entire device according to a predetermined program. The input device 2 is used for inputting search conditions and the like. The display device 3 is, for example, a CRT (Ca
thode-ray tube) and LCD (Liquid Crystal Display)
Display similar document search results. The external storage device 4 is composed of, for example, a magnetic disk device or an optical disk device, and holds various data handled by this device.

The external storage device 4 has a document database 4a in which a plurality of documents are registered for each predetermined category, and a category identification in which the ID of a document corresponding to each category classified as a new category (current category) is registered. Document database 4b, a search target category, a search target category document information database 4c in which information such as a previous time category name group, a document ID group, and a document registration date is registered, a word ID in a specific category and a word corresponding thereto , A word conversion information database 4d registered as a pair, a word weight table database 4e registered with a weight value of a word appearing in a category unit, and a word ID and a word corresponding to a word ID in a category that goes back in time series from a specific category And a time-sequential word conversion information database 4f registered as a pair is constructed. There.

FIG. 3 is a diagram showing the internal configuration of the control device 1. The control unit 1a is constituted by a CPU and a ROM, and expresses each function executed by these individually. On the other hand, the memory unit 1b is constituted by a RAM, and expresses each content of data stored therein individually.

The control unit 1a includes an initialization unit 201, an input unit 2
03, output unit 205, search key document input unit 207, search key document item cutout unit 209, search key word extraction unit 2
11. Search key word appearance frequency calculation unit 213, similar category identification unit 215, search target category document information reading unit 2
17, specific category word conversion information reading section 219, time-series word weight reading section 221, search target document reading section 2
23, search target document item cutout unit 225, search target document word extraction unit 227, search target word appearance frequency calculation unit 22
9, common word extraction unit 231, similarity calculation unit 233, previous category search unit 235, time-sequential word conversion information reading unit 2
37, an inter-category word conversion information generation unit 239, and a search result output unit 241.

The memory section 1b includes a search key document storage buffer section 251 and a search key document item storage buffer section 25.
3. Search key word information storage buffer unit 255, specific category storage buffer unit 257, search target category document information storage buffer unit 259, previous category search buffer unit 26
1. Specific category word conversion information storage buffer unit 263,
A time-series word weight reading buffer 265, a search target document storage buffer 267, a search target document item storage buffer 269, a search target word information storage buffer 271,
Common word information storage buffer 273, inter-category word conversion information storage buffer 275, similarity storage buffer 2
77, a time-sequential word conversion information storage buffer unit 279, and a search result output buffer unit 281.

Here, the initialization section 201 initializes each buffer section. The input unit 203 inputs search key document settings and the like in response to a user's designation operation using the input device 2. The output unit 205 outputs the content of the search key document and the search result performed by the input unit 203 to the display device 3.

The search key document input unit 207 stores the text information of the search key document input from the input device in the search key document storage buffer unit 251.

The search key document item cutout unit 209 analyzes the sentence structure of the search target document stored in the search key document storage buffer unit 251, cuts out the unit of contents as a unit of contents, and searches the unit of the search key. It is stored in the document item storage buffer unit 253. Search key word extraction unit 21
1 performs word segmentation of the text document information stored in the search key document item storage buffer unit 253. further,
A key word in representing the contents of the document is extracted, and the extracted word type is stored in a search key word information storage buffer unit 25.
5 is stored. For example, word segmentation is performed using morphological analysis or the like, and words that are key in representing the contents of the document are performed using part-of-speech information (for example, “noun” or “sa-variable noun”) of each word.

The search key word appearance frequency calculation unit 213 stores the key word extracted by the search key word extraction unit 211 in the search target document item storage buffer unit 25.
3 is calculated for each word type in the text document stored in No. 3 and stored in the search key word information storage buffer unit 255.

The similar category specifying unit 215 stores the category specifying document database 4 stored in the external storage device.
b, the search key and the similarity of each document are calculated from the word information of the category-added document and the word information of the search key document stored in the search key word information storage buffer 255. An arbitrary number of search key document categories are specified based on the degree and the category assigned to the document, and the specified categories are stored in the specific category storage buffer unit 257.

Search target category document information reading section 217
Calls the search target category document information stored in the external storage device 4 from the search target category stored in the specific category storage buffer unit 257 and stores it in the search target category document information storage buffer unit 259. The search target category document information includes a category name, a previous time category name group, a document identification number (document ID) group corresponding to this category, and a field code of time-sequential word conversion information belonging to this category (category code and ID of a word included in the category). The number of documents, the period, and the number of previous time categories registered in one category are arbitrary for each category.

Specific category word conversion information reading section 219
Reads word conversion information corresponding to a specific category similar to the search key document from the external storage device 4 and stores it in the specific category word conversion information storage buffer unit 263. The specific category document word conversion information includes a field code and a word character string corresponding to the code.

The time-series word weight reading unit 221 calls the word weight table corresponding to the category stored in the search target category document information storage buffer unit 259 from the external storage device 4 and is included in the documents belonging to the category. The weight value of the word is stored in the time-series word weight reading buffer unit 265. Since the word weight changes depending on the category and its time series, the word weight table is stored in the external storage device 4 for each category and each time series.

The search target document reading unit 223 reads a document to be converted into a document database from the external storage device in order to convert the information related to the document stored in the external storage device into a document database, and converts the text document information into the search target document. The data is stored in the storage buffer unit 267.

The search target document item cutout unit 225 analyzes the sentence structure of the search target document stored in the search target document storage buffer unit 267, cuts out the content in units of items, and searches the target unit for each item. It is stored in the document item storage buffer unit 269. Search target document word extraction unit 22
Reference numeral 7 performs word segmentation of the text document information stored in the search target document item storage buffer unit 269. further,
A key word for representing the contents of the document or the item is extracted, and the extracted word is stored in the search target word information storage buffer unit 271.

The search target word appearance frequency calculation unit 229 stores the key words of the key words extracted by the search target document word extraction unit 227 in the search target document storage buffer unit 267 or the search target document item storage unit 269. The appearance frequency in the text document is calculated for each word, and stored in the search target word information storage buffer unit 271.

The common word extraction unit 231 includes a search key word information storage buffer unit 255 and the word information of the search key document and the search target word information storage buffer unit 271.
Using the word information of the search target document stored in the buffer and using the inter-category word conversion information in the inter-category word conversion information storage buffer unit 275, the words stored in both buffers and the information on the frequency thereof are shared. The information is stored in the information storage buffer unit 273. That is, when the word stored in the search key word information storage buffer 255 and the word stored in the search target word information storage buffer 271 are different, both words are stored in the inter-category word conversion information storage buffer 275. Is stored in the common word information storage buffer unit 273 taking this into account.

The similarity calculation unit 233 receives a word of the word from the search key word information storage buffer unit 255, the search target word information storage buffer unit 271, the common word information storage buffer unit 273, and the time-series word weight read buffer unit 265. By adding the word weight of the word to the appearance frequency, the similarity between the search key document and one search target document is calculated by a method such as the word vector space method, and the similarity value is stored in the similarity storage buffer unit 2.
77.

The previous category search unit 235 activates the search target category document information reading unit 217 and, if there is a previous category stored in the previous category search buffer unit 261, externally stores the search target category document information of the previous category. It is called from the device 4 and stores the search target category document information of the previous time category in the search target category document information storage buffer unit 259. This search target category document information is category link information for associating a previous category in a specific category.

Time-sequential word conversion information reading unit 237
Reads the time-series word conversion information stored in the external storage device 4 and stores it in the time-series word conversion information storage buffer unit 279. The time-series word conversion information includes a field code (category code and I of a word included in the category).
D) and a word character string corresponding to the field code.

The inter-category word conversion information generation section 239
The field code included in the similar category specified as similar to the search key document stored in the specific category word conversion information storage buffer unit 263 and the time series word conversion information storage buffer unit 279 Each word of the same field code with reference to the field code (words with the same field code but different word character strings)
In the same language, and the information is stored in the inter-category word conversion information storage buffer unit 275. This inter-category word conversion information is word link information for associating words with different notations between categories, despite consent.

The search result output unit 241 stores, in the search result output buffer unit 281, document information (for example, a document ID) of a higher similarity based on the similarity for each search target document stored in the similarity storage buffer unit 277. I do. Then, the contents of the search result output buffer unit 281 are output to the display device 3 via the output unit 205.

Next, the operation of the similar document search apparatus of the present embodiment will be described with reference to FIG. The operation described here is performed by the CPU of the control unit 1a of the control device 1 using a program in the ROM and a storage area in the RAM described as the memory unit 1b.

First, the initialization unit 201 is activated and the memory unit 1
b is cleared (step 400). Subsequently, when the search key document input unit 207 is activated and a search key document is input via the input device 2, the search key document is stored as a text document in the search key document storage buffer unit 251 (step 401). . Here, it is assumed that a patent application specification is used as a search key document to be input, and an example thereof is shown in FIG.
Shown in The search key document item cutout unit 209 analyzes the sentence structure of the search key document stored in the search key document storage buffer unit 251 and cuts out the content in units of items, which are illustrated in FIG. It is stored in the search key document item storage buffer unit 253 in such a format. In the storage example of FIG. 6, “[Title of Invention] [Claims]
[Summary] "to extract items and store them in the buffer for each item. Next, the search key word extraction unit 211 is activated and cuts the text into words using morphological analysis or the like.
A key word representing the content of the document, such as "a" or "sa noun", is extracted and stored in the search key word information storage buffer unit 255 as shown in FIG. For example, "information / database / storage / road / information /.../ 4WD" is extracted from "road information stored in the information database ...". Subsequently, the search key word appearance frequency calculation unit 213 is activated, and the words stored in the search key word information storage buffer unit 255 are searched for in the text document of the items stored in the search key document item storage buffer unit 253. The appearance frequency is calculated and stored in the search key word information storage buffer 255 (step 402). For example, as shown in FIG. 8, words and frequencies are described side by side, and when "4WD" appears twice, it is stored as "4WD = 2". This completes the retrieval key document information extraction.

Next, the process proceeds to step 403 for specifying a similar category. The similar category specifying unit 215 uses the word information of the category-added document in the category specifying document database 4b of the external storage device 4 and the word information of the search key document stored in the search key word information storage buffer unit 255. Then, the similarity between the search key document and each category-added document is calculated, and an arbitrary number of search key document categories (for example, the top two Category) Specify. The category specified in this way is stored in the specific category storage buffer unit 257. The category-added document information used in step 403 includes, for example, a new category and a document ID to which the category is assigned, as shown in FIG. Figure 10 shows the identified categories.
The category name is stored as shown in FIG.

When the specification of the similar category in step 403 is completed, the process proceeds to step 404 to read the category. The search target category document information reading unit 217 acquires the search target category document information in the category stored in the specific category storage buffer unit 257 from the search target category document information database 4c of the external storage device 4, and retrieves the search target category document. Information storage buffer unit 259
To be stored. For example, as shown in FIG. 11, the category name “Category ID = 14 Traffic” and the storage period “1999.1 to 1999.
2), the previous category 1 “category ID = 23 car”, the previous category 2 “category ID = 24 bicycle”, the corresponding document ID and the document date are stored. This is because the category "Transportation" currently exists,
Documents from January 1999 to February 1999 are stored.
The category classification before January 1999 indicates that the current "traffic" was classified into the categories "automobile" and "bicycle". If there is no update in the category for the category "traffic", "category" and
Will be represented by the same category name. The previous category names (here, “automobile” and “bicycle”) included in the search target category document information are also stored in the previous category search buffer unit 261 as shown in FIG.

Next, in step 405, specific category word conversion information is read. The specific category word conversion information reading unit 219 includes a specific category storage buffer unit 25.
7, the word conversion information corresponding to the specific category is read from the word conversion information database 4d of the external storage device 4, and stored in the specific category word conversion information storage buffer unit 263. The specific category word conversion information is composed of a field code (category code and ID of a word included in the category) and a word character string corresponding to the code. For example, as shown in FIG. 13, a category ID, a word ID within the category, and the word
A word character string corresponding to the ID is stored, which indicates that “word ID = 101, word = 4WD” is stored. The word ID numbers are assigned in the order of appearance of the new words. As will be described later, an operator corrects the IDs of the words to be associated between the categories so that they are common.

Following step S405, step S40
In step 6, word weight reading is performed. The time-series word weight reading unit 221 stores a word weight table corresponding to the category stored in the search target category document information storage buffer unit 259, that is, a unique word weight table for each category.
1 is read from the word weight table database 4e of FIG.
4, the weight value of the word is stored in the time-sequential word weight reading buffer unit 265. In this example, "Category =
Indicates the weight value of the words included in the search target documents belonging to "traffic" and "storage period = 1999.1-1999.2", and associates the words with the weight values such as "car = 10" and "motor = 5" Stored. Here, words with a large number of appearing documents are
The “weight” is increased, and conversely, words appearing with a small number of documents are reduced in weight. That is, if the number of documents in which the target word appears is large, it means that the contribution ratio of the word to the similarity determination is low, and the above-described weighting is performed.

In this step 406, if the previous category of the specific category is stored in the search target category document information storage buffer section 259, the word weight table is unique to the category as described above, so that it corresponds to the previous category. Is stored in the time-series word weight reading buffer unit 265.

Subsequently, in step 407, a search target document is extracted. The search target document reading unit 223 reads a text document of one search target document from the external storage unit 4,
As illustrated in FIG. 15, the search target document storage buffer unit 2
67.

After the extraction of the search target document, in step 408, the search target document information is extracted. First, the search target document item extraction unit 225 is activated, and the search key document item extraction unit 2
As in the case of the search key document 09, the sentence structure is analyzed for the search target document stored in the search target document storage buffer unit 267, the content is cut out in units of items, and the item unit is extracted. Is stored in the search target document item storage buffer unit 269. In the example of FIG. 16, items are cut out based on “name of invention”, “claims”, “summary”, and the like, and stored in the buffer unit 269 in item units. Thereafter, the search target word extraction unit 227 cuts the text into words using morphological analysis or the like, extracts a key word representing the content of the document such as “noun” or “sa-variable noun”, and searches for the word type. It is stored in the word information storage buffer unit 271. FIG.
In the example of No. 7, "information / database / accumulation / road /.../ four-wheel drive vehicle" is extracted from "road information stored in the information database ..." and stored in the buffer 271. . Furthermore, the search target word appearance frequency calculation unit 22
9 starts, calculates the appearance frequency of the words stored in the search target word information storage buffer unit 271 in the text document, and stores the words in the search target word information storage buffer unit 271 again. For example, as shown in FIG. 18, words and frequencies are described side by side, and when "four-wheel drive vehicle" appears twice, it is stored as "four-wheel drive vehicle = 2".

When the retrieval target document information extraction is completed, the process proceeds to the similarity calculation in step 409, and the similarity between the retrieval key document and each retrieval target document is calculated. First, the common word extraction unit 231 is activated, and the search target word information storage buffer unit 2 is started.
71 and the search key word information storage buffer unit 255 are stored in common word information storage buffer unit 2
73.

In this step 409, when the similarity of the search target document of the target category is first obtained and then the similarity of the search target document of the previous category is obtained, the inter-category word conversion information storage buffer is used. The conversion information between words between another category that has already been implemented and the category may be stored in the unit 275, and a common word is extracted in consideration of this. For example, if there is information “4WD... Four-wheel drive vehicle”, “4WD” and “four-wheel drive vehicle” are regarded as the same word, and “information / database / storage / road /.. ./4WD” and “4WD” The "information / database /.../ 4WD" which is a common word of "information / database / accumulation / road /.../ four-wheel drive vehicle" is stored in the common word information storage buffer unit 273 as shown in FIG. . A method of storing the inter-category word conversion information in the inter-category word conversion information storage buffer unit 275 will be described later.

Next, the similarity calculating section 233 is activated, and the search target word information storage buffer section 271, the search key word information storage buffer section 255, the time-sequential word weight reading buffer 265, and the common word information storage buffer section 273 are provided. Is calculated using the vector space method or the like from each piece of stored information, and the similarity value is stored in the similarity storage buffer unit 277. The similarity is set so that the document to be searched and the similarity value form a pair, for example, as shown in FIG. 20, “Document ID = 10 Similarity = 0.540
4, document ID = 13 similarity = 0.2351 ".

After step 409, the search target document reading unit 233 starts up and determines whether or not there is a search target document in the external storage unit 4 that has not been processed yet (step 41).
0) If there is a document to be searched, the process returns to step 407. On the other hand, if there is no search target document that has not been processed in step 410, the process proceeds to step 411.
In step 411, the previous category search unit 235 determines whether there is category document information stored in the previous category search buffer unit 261. If there is a previous category, the process proceeds to step 412.

In step 412, the previous category search unit 2
35 activates the search target category document information reading unit 217, calls the search target category document information of the category stored in the previous category search buffer unit 261 from the search target category document information database 4c of the external storage device 4, and retrieves the search target category document information. Category document information storage buffer unit 25
9 is stored.

Next, the routine proceeds to step 413. The time-series word conversion information reading unit 237 reads the time-series word conversion information stored in the time-series word conversion information database 4f of the external storage device 4, and stores the time-series word conversion information as shown in FIG. The data is stored in the buffer unit 279. The time-series word conversion information includes a field code (category code and ID of a word included in the category) and a word character string corresponding to the code. In the example of FIG. 23, in the category ID = 23 “car”, “word ID = 101, word =
Four-wheel drive vehicle "is stored. The word ID numbers used here are also assigned in the order of appearance of new words, similarly to the specific category word conversion information. And between a specific category, the previous category, and other categories,
The operator corrects the word IDs of the words to be associated between the categories so that they are common.

When reading of the time-series word conversion information is completed, the flow advances to step 414 to generate inter-category word conversion information. Inter-category word conversion information generation unit 239
The field code (category code and word ID) included in the similar category specified as similar to the search key document stored in the specific category word conversion information storage buffer unit 263, Referring to the field code (category code and word ID) stored in the word conversion information storage buffer unit 279, each word having the same word ID already prepared as described above (that is, the word ID is the same) The word information is stored in the inter-category word conversion information storage buffer unit 275 in order to catch words having different word character strings) in the same language. For example, as shown in FIG. 22, from the contents of the specific category word conversion information storage buffer 263 shown in FIG. 13 and the contents of the time-sequential word conversion information storage buffer 279 shown in FIG. By storing "4WD" and "four-wheel drive vehicle" in association with each other, the two words are associated with each other so as to be regarded as the same language. When the generation of the inter-category word conversion information ends in step 414, step 406
Returning to step 410, the process is repeated in the same manner as described above.

If it is determined in step 411 that all search target category document information stored in the previous category search buffer unit 261 has been processed, the process proceeds to step 415. The search result output unit 241 calculates the similarity of each search target document stored in the similarity storage buffer unit 277,
A search target document having a high degree of similarity is extracted, and document information (for example, a document ID) serving as a search key is stored in the search result output buffer unit 281 as illustrated in FIG. Then, the contents of the search result output buffer unit 281 are output to the display device 3 via the output unit 205 as illustrated in FIG. In the present embodiment, as described above, by generating inter-category word conversion information from the specific category word conversion information and the time-series word conversion information, a common word when calculating the similarity between the search key document and the search target document is generated. At the time of extraction, it is possible to handle words as common words regardless of changes in word phrases over time. Therefore, it is necessary to improve the accuracy of similarity calculation and similar document search when using a database whose categories are updated in chronological order. Can be.

The present invention is not limited to the above embodiment as long as it does not depart from the gist of the present invention. And
The present invention can be widely applied to a document creation support device, an information search device, an information management device, an information filtering device, and the like.

[0056]

As described above in detail, according to the present invention, a similar document retrieval apparatus for improving the accuracy of similarity calculation and similar document retrieval when a database whose categories are updated in time series is used, and this apparatus. Can be provided.

[Brief description of the drawings]

FIG. 1 is a diagram for explaining an outline of a similar document search device according to the present invention;

FIG. 2 is a block diagram illustrating a configuration of an embodiment of a similar document search device according to the present invention.

FIG. 3 is a block diagram showing internal functions of a control device of the similar document search device according to the embodiment;

FIG. 4 is a view showing an operation of a similar document search process according to the embodiment;

FIG. 5 is a diagram showing an example of storing a search key document.

FIG. 6 is a diagram showing an example of retrieval key document item storage;

FIG. 7 is a diagram showing a storage example of search key word information.

FIG. 8 is a diagram showing an example of storing search key word information with appearance frequency information;

FIG. 9 is a diagram showing an example of similar category specifying document information.

FIG. 10 is a diagram showing a specific category storage example.

FIG. 11 is a diagram showing a storage example of search target category document information.

FIG. 12 is a diagram showing a previous category search storage example.

FIG. 13 is a diagram showing an example of storing specific category word conversion information.

FIG. 14 is a diagram showing an example of reading and storing word weights by time series;

FIG. 15 is a diagram illustrating an example of storing a search target document.

FIG. 16 is a diagram showing an example of storing search target document items.

FIG. 17 is a diagram showing an example of storing search target word information.

FIG. 18 is a diagram showing a storage example of search target document items with appearance frequency information.

FIG. 19 is a diagram showing an example of storing common word information.

FIG. 20 is a diagram showing a similarity storage example.

FIG. 21 is a diagram showing an example of storing word conversion information by time series.

FIG. 22 is a diagram showing an example of storing word conversion information between categories.

FIG. 23 is a diagram showing a similarity storage example.

FIG. 24 is a diagram showing a search result output example.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 ... Control device 1a ... Control part 1b ... Memory part 2 ... Input device 3 ... Display device 4 ... External storage device 4b ... Category specification document database 4c ... Search target category document information database 4d ... Word conversion information database 4f ... Time series Another word conversion information database 215 ... Similar category specifying unit 217 ... Search target category document information reading unit 219 ... Specific category word conversion information reading unit 221 ... Time series word weight reading unit 235 ... Previous category search unit 237 ... Time series word Conversion information reading unit 239 ... inter-category word conversion information generation unit 257 ... specific category storage buffer unit 259 ... search target category document information storage buffer unit 261 ... previous category search buffer unit 263 ... specific category word conversion information storage buffer unit 265 ... hour Word weight reading buffer by sequence Part 275 ... category between the words conversion information storage buffer section 279 ... time series by word conversion information stored in the buffer unit

 ──────────────────────────────────────────────────続 き Continuing on the front page (72) Inventor Takuya Nishina 1-3-3-1 Shinmachi, Ome-shi, Tokyo (72) Inventor Hiroshi Yamazaki 1-3-3-1 Shinmachi, Ome-shi, Tokyo 1 (72) Inventor Tsuyoshi Matsukuma 3F-3 Shinmachi, Ome-shi, Tokyo 1F term (reference) 5B075 ND03 NK02 NK32 NK44 NR06 NR12 PP25 PQ02 PR06 QM08 UU06

Claims (7)

[Claims] 1. A database storing a plurality of documents classified by a predetermined category, a document belonging to a first category indicating a classification used in a first period as a unit of change of a category and a document belonging to a second period. A first storage unit for storing category link information for associating a document belonging to a second category indicating a classification to be used; and a different but consensus expression between the first category and the second category. Second storage means for storing word link information for associating words to be used; category specifying means for specifying the category to which the document input as the search key belongs for the first category; Determining the second category associated with the first category specified from the link information stored in the first storage means. From the database using the word link information stored in the second storage means, as a search target, and the documents belonging to the first category and the second category obtained by the determination means. A similar document search comprising: searching means for searching for a document similar to the search key document; and output means for outputting the similar document obtained by the searching means as a search result for the search key document. apparatus. 2. The method according to claim 1, wherein the search means includes means for creating the word link information by comparing a word used in the first category with a word used in the second category. The similar document search device according to claim 1, wherein 3. The method according to claim 1, wherein said search means calculates a similarity using the words linked by the word link information as the same word, and extracts a similar document based on the similarity. Similar document search device. 4. The search means calculates a similarity of a word included in the document belonging to the first category in consideration of a first weight value, and calculates a word included in the document belonging to the second category. 2. A similar document search apparatus according to claim 1, wherein the similarity is calculated by taking into account a second weight value, and a similar document is extracted based on the calculated similarity. 5. The method according to claim 1, wherein the first category and the second category are created in chronological order.
A similar document search device described in the section. 6. A similar document retrieval method for retrieving similar documents from a database in which a plurality of documents classified for each predetermined category are registered, wherein a classification used in a first period for a category change unit Category link information for associating a document belonging to a first category indicating a category with a document belonging to a second category indicating a classification used in a second period, and the first category and the second category. The word link information for associating words using different expressions while agreeing is stored in the memory, and the category to which the document given as the search key belongs is specified for the first category, and the specified Said second category associated with said first category
Is determined from the category link information stored in the memory, documents belonging to the first and second categories are searched, and similar to the search key document from the database using the word link information. A similar document search method comprising: searching for a document; and outputting the similar document as a search result for the search key document. 7. A computer provided with a database in which a plurality of documents classified for each predetermined category are registered, a document belonging to a first category indicating a classification used in a first period as a unit of change of a category, Category link information for associating a document belonging to a second category indicating a classification used in the second period, and a word in which the first category and the second category agree but different expressions are used A function of storing word link information for associating a document with a memory, a function of specifying a category to which a document given as a search key belongs for the first category, and a function of relating to the specified first category. The second
And a function of judging the category of the document from the link information stored in the memory. The documents belonging to the first category and the second category are searched, and the search key is searched from the database using the word link information. A computer-readable recording medium storing a program for executing a function of searching for a document similar to a document and a function of outputting the similar document as a search result for the search key document.