JP2002014970A - Language transverse type concept retrieving system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a language transverse type concept retrieving system by which documents in various kinds of languages can be obtained precisely and simultaneously if a user has knowledge at the word level of a foreign language except for one's mother tongue by simultaneously retrieving documents in many languages by only preparing an inquiry expression in a single language. SOLUTION: The language transverse type concept retrieving system which is constructed so as to retrieve respectively different language documents and includes plural similar document retrieving systems (21, 23), in which documents of the same contents described in respective languages are scattered is provided with a corresponding document retrieving device (6) for extracting the documents of the same contents described in each language corresponding to the document existing in the result of retrieving with respect to a retrieving question in one language and for outputting the documents of the same contents as a retrieving kind document for each language and a similar document retrieving system 23 for simultaneously retrieving the similar document in the plural languages by using the outputted retrieving kind document.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention includes a plurality of similar document retrieval systems constructed to retrieve documents in different languages, and each similar document retrieval system includes documents having the same contents described in each language. Is a cross-language concept retrieval system in which is scattered.

[0002]

2. Description of the Related Art In order to obtain a foreign patent document, a method of accessing a foreign language patent document database, creating a query (question sentence) in a foreign language, and obtaining the foreign patent document, has been prepared in a native language. There is a method in which a query is translated into a foreign language query using an automatic translator and searched to obtain a foreign patent document.

The former requires not only knowledge of foreign languages but also knowledge of characteristics of a search system to be used to create a query including synonyms and related words. It is said that you cannot get it. In the latter case, in the native language, for example, Japanese and English, the meaning of the words is not one-to-one, and in many cases it cannot be translated into an English query that matches the Japanese query, and a satisfactory answer cannot be obtained. There was a problem.

[0004]

SUMMARY OF THE INVENTION A first object of the present invention is to provide a cross-language concept search system capable of simultaneously searching documents in many languages simply by creating a query formula in one language. is there.

[0005] A second object of the present invention is to provide a cross-language concept retrieval system capable of acquiring documents in various languages more precisely and simultaneously if there is knowledge at the word level of a foreign language other than the native language. It is to be.

[0006]

The above objects can be attained by the following means. That is, the cross-language concept search system according to the first aspect of the present invention includes a plurality of similar document search systems constructed to search for different language documents, and each similar document search system includes a language-specific document. In a cross-language concept search system in which documents having the same content described are scattered, documents having the same content described in each language corresponding to the document present in a search result for a search query in one language are searched for. Corresponding document search device for extracting documents having the same content as a search type document for each language, and a similar document search system for simultaneously searching for similar documents in a plurality of languages using the output search type documents Is provided.

A cross-language concept search system according to a second aspect of the present invention includes a plurality of similar document search systems constructed to search different language documents.
In each of the similar document search systems, in a cross-language concept search system in which documents of the same content described in each language are scattered, in the cross-language concept search system, A corresponding document search device that extracts documents of the same content described in a language and outputs the same document as a search species document for each language, and a similar document in a plurality of languages using the output search species document. A similar document search system for simultaneous searching; and a seed document meaning adjusting device for adjusting the meaning of the document selected from the search result from the similar document search system and performing a similar document search again. .

[0008] A cross-language concept search system according to a third aspect of the present invention includes a plurality of similar document search systems constructed to search for different language documents, respectively.
In each of the similar document search systems, in a cross-language concept search system in which documents of the same content described in each language are scattered, in the cross-language concept search system, A corresponding document search device that extracts documents of the same content described in a language and outputs the same document as a search species document for each language, and a similar document in a plurality of languages using the output search species document. A similar document search system configured to simultaneously search, and a re-search instruction device configured to select an optimal seed document from search results from the similar document search system and perform similar document search again. is there.

According to a fourth aspect of the present invention, there is provided a cross-language concept search system comprising the cross-language concept search system according to the first to third inventions. A text display device for displaying the result is provided.

According to a fifth aspect of the present invention, there is provided a cross-language concept search system according to the first to fourth inventions, wherein the cross-language concept search system further includes an optional language. An automatic translator for translating from one language to any other language is provided.

[0011]

DESCRIPTION OF THE PREFERRED EMBODIMENTS In the field of patents, the right to claim priority granted by the Parry Alliance Convention, the Patent Cooperation Treaty, etc. (within one year from the filing date of the first country for the same invention) The right to be filed on the same date as the filing date of the first country).

An application filed in each country claiming these priorities can be regarded as a group of application inventions for each country in which the same technical content is described using the priority number as a key. This is what is usually called a patent family. And these patent applications are concentrated in globalized technical fields.

[0013] In other words, the patent literature enemy databases in each country contain, at a fixed rate, patent documents having exactly the same technical contents, and the patent documents forming the patent family include the application number of one country. This means that patents with the same technical content in each country can be found quickly.

The similar document search system searches an input question document based on a vector space model. Currently, for example, Vext Search, Komatsu Software Corporation, a registered trademark, Concept Base Search, Justsystem
Products such as the Company, registered trademarks, etc. are on sale. The outline of such a similar document search system is as follows.

Each document in the database is subjected to morphological analysis to extract a content word (term), and an index based on the term is created. Each term is treated as a term vector weighted according to its importance in the document. Each document is also expressed as a document vector obtained by synthesizing a term vector weighted with a term (vector space model).

At the time of retrieval, a question is also expressed as a question vector composed of a combination of weighted term vectors, and the similarity between the retrieval query and the document is calculated by the cosine of the angle formed by the two vectors. This is calculated assuming that the inner product of the document vector and the document vector represents the similarity, and the search documents are sorted and presented in descending order of the similarity.

The term weight is calculated using the TF / IDF method. TF (Term Frequency) is the frequency of occurrence of a term in a request for a question or each document.
quency) is the logarithm of the ratio of the total number of documents to the number of documents in which the term appears. Here, a term having a large weight by TF / IDF, that is, a term in which the term appears in a small number of documents, but appears in a specific document in a large number can be regarded as a keyword that characterizes the document. Normally, in such a system, a Boolean search (search by a logical expression) can be performed by using the index by the term as it is.

For the vector space model, see Salton,
G. and McGill, MJ "Introduction to Modern Inform
ation Retrieval ”McGraw-Hill (1983) or“ Ra
nking Algorithms "(Donna Harman, Information R
etrieval, p. 363-392).

From the above, if a patent family-related application can be extracted from a search set of Japanese patent documents matching a search query from a Japanese patent document database and a corresponding US patent application can be obtained, Similar US patent documents in the United States can be obtained by searching for similar documents using the invention content of the US patent application as a seed document.

By the way, in the outline of the similar document search,
"Each term is treated as a term vector weighted according to the degree of importance in the document. Each document is also expressed as a document vector obtained by combining the term vector weighted with the term." This means that operating the weight of the term vector changes the document vector, that is, the seed document content can be changed by the weighting operation. In other words, by manipulating the weights of terms that have particularly significant meaning in the seed document, it means that the search query in the US Patent Literature Database can be changed.

The present invention is an invention based on the above facts. Hereinafter, specific examples of the present invention will be described. That is, FIG. 1 and FIG. 2 are a block diagram and a flowchart, respectively, for explaining the first embodiment of the present invention.

In FIG. 1, reference numeral 1 denotes a search query input device for inputting a search query as a search request, 2 denotes a Japanese patent database, 3 denotes a text search device, and 4 denotes a text display device for displaying search results.

The patent information search system composed of the search query input device 1, the Japanese patent database 2, the text search device 3, and the text display device 4 is a PATOLIS (registered trademark, foundation, currently provided as a commercial service). A Boolean search method such as the Japanese Patent Information Organization, or a full-text search system (n-gram) method of a patent CD-ROM publication may be used, but a search query using a document format and a logical expression of a keyword is possible. In addition, it is preferable that the similar document search system is based on a vector space model system in which search results can be ranked.

When the program of this embodiment starts (step S00), when a search query of the Japanese search system is input in step S01, the search query is sent to the text search device 3 and searched by the Japanese patent database 2. Is executed (step S02). Step S
At 03, the text display device 4 displays the result obtained by the search. Items displayed at this time are Japanese texts (abstracts or texts, bibliographic items, etc.). When the searcher determines that a satisfactory document has been obtained by viewing the display (step S04, YES), the process proceeds to the next step S05, but otherwise, the process returns to step S01.
The correction of the search question (step S01), the search (step S02), and the display (step S03) are repeated several times until a satisfactory answer is obtained.

In step S05, a so-called patent family search is executed. The corresponding patent relation database 5 used at this time is a database in which Japanese patent document numbers and foreign patent document numbers having a patent family relationship are registered. Related US patent document numbers can be searched.

Reference numeral 7 denotes a compatible patent confirmation apparatus for evaluating the compatibility of Japanese patent documents having a US patent family. Normally, the ratio of U.S. patent documents in Japanese patent documents is about 5% on average, but it is said that this ratio will be significantly higher in technical fields where global activities are being carried out. However, conversely, this indicates that there may be no U.S.-compliant patent documents of Japanese patent documents that best match the search request. That is, it indicates that more relevant literature may be found in US patents. Therefore, the compatible patent confirmation device 7 displays the terms attached to the Japanese patent documents searched in step S05, the weight of the terms, the abstract, etc., and among the Japanese patent documents in which U.S.-compatible patent documents exist, This is provided to confirm the most suitable patent document (step S06).

Utilizing the ranking presented by the text search device 3 described above, the highest ranking document among Japanese patent documents in which US-compatible patent documents exist is automatically (mechanically) used. However, in the present embodiment, step S06 is performed so as to ensure higher compatibility.
After the user selects and confirms in step S0,
Go to 7.

The compatible patent selection system composed of the corresponding patent relation database 5, the corresponding patent search device 6, and the compatible patent confirmation device 7 provides a search set consisting of Japanese patent documents output from the text display device 4, When it is determined that the set matches the user's search request, the search set is output from the text search device 3 to the compatible patent selection system.

Next, the user selects a Japanese patent document which has a US-compliant patent document and is most suitable for the search request.
When confirmed, the document number of the U.S.-compliant patent document is output to the similar document search device 8. The similar document search device 8 is a search device capable of at least a number search and a similar document search. First, a number search is performed, and U.S.-compatible patent documents are extracted from the U.S. patent database 9 (step S07).

Subsequently, in step S08, similar documents are searched using the extracted documents as search type documents (question sentences), and similar US patent documents are ranked and output from the US patent database 9. This output is sent to the text display 1
0 is displayed (step S09). The text display apparatus 10 weights, for example, an appropriate number of patent documents such as the top 10 cases in the original document (original text, English), its translation (translated by the English-Japanese automatic translation device 11), and the original document. Characteristic terms (English) can be displayed together. The number of feature terms is also in the top 10
You can specify an appropriate number, such as.

The searcher inputs whether or not the document displayed in step S09 is appropriate (step S10).
If appropriate (step S10, YES), step S
At 11, the output of the document ends (step S99).

On the other hand, if the determination in step S10 is NO
In the case of, the process proceeds to step S21, in which the searcher performs the seed document meaning adjustment, that is, the adjustment of the feature term weight.

The seed document semantic adjustment device 12 is adapted to execute the search if the relevance of the search seed document extracted from the US-compliant patent document is insufficient.
The purpose is to adjust and change the meaning of a search type document, that is, a search query, by adjusting the weight of the feature term.
The user (searcher) changes the weighting of the feature terms with reference to the weighted feature terms of the original document, the translated document, and the original document, and repeats the similar document search to search for a U.S. patent document with higher relevance. I can do it. Steps S08 to S10 are repeatedly executed until an appropriate document is matched.

FIGS. 3 and 4 are explanatory diagrams for explaining how to change the term weighting described above using an actual example. FIG. 3 shows a screen (part) in which the original document displayed on the text display device 10 and its translation are displayed side by side. In FIG. 4, the term number,
This is a screen showing the bilingual terms, the terms of the original document, the weights assigned to them, and the weights that have been changed.

The screen shown in FIG. 3 and the screen shown in FIG. 4 can be switched to each other by pressing a switch button (not shown).
In these figures, the multilingual origin of multi language,
Intermediate language, descriptor,
Also, the weighting of the term efficiency has been changed. This change is performed by operating a change button (right side in FIG. 4).

As described above, in the first embodiment,
From the Japanese patent documents obtained by the Japanese patent search system, the corresponding US patent documents of the patent family are searched, and the US patent database is searched using the US patent documents as seed documents (question sentences). The U.S. Patent Database contains a large number of Japanese patents that do not have families, and covers a large number of U.S. patent documents that are not covered by the Japanese patent search system. Can be found.

If the result of the search is not satisfactory, the searcher changes the weight of the term to adjust the meaning of the seed document, thereby creating a more appropriate question, and a more appropriate US patent. Access to literature.

FIGS. 5 and 6 are a block diagram and a flowchart for explaining a second embodiment of the present invention. This embodiment is an example in which a re-search instruction device 13 is used instead of the seed document meaning adjusting device 12 of the first embodiment.
Only the re-search instructing device 13 in FIG. 5 and step S31 in FIG. 6 are different from those in the first embodiment, and a duplicate description will be omitted as much as possible.

As already described in the first embodiment, in the second embodiment, the user selects and confirms a Japanese patent document in which a US-compatible patent document exists and is most suitable for a search request. Then, the document number of the US-compliant patent document is output to the similar document search device 8.

The similar document search device 8 is a search device capable of at least a number search and a similar document search. First, a number search is performed, and a US-compliant patent document is extracted from the US patent database 9. Subsequently, a similar document search is performed using the extracted document as a search type document, and similar US patent documents are ranked and output from the US patent database 9.

The text display device 10 is, for example,
About an appropriate number of patent documents are displayed together with the original document (English), its translation (Japanese), and the weighted feature terms (English) of the original document (step S0).
9) You can do it.

When the desired US patent document cannot be obtained by the first search (step S10, NO), the user
With reference to this translation (Japanese), select a different document that matches the search request of the user (step S3).
1) Using this as a seed document, a similar document search is again performed by the re-search instruction device 13 (step S08). This is repeated several times so that the most suitable documents are ranked higher.

As described above, in the second embodiment, similar to the first embodiment, the corresponding US patent document of the patent family is searched from the Japanese patent document obtained by the Japanese patent search system. The U.S. patent database is searched using this U.S. patent document as a seed document (question text). The U.S. Patent Database contains a large number of Japanese patents that do not have families, and covers a large number of U.S. patent documents that are not covered by the Japanese patent search system. Can be found.

In the second embodiment, if the result of the first search is not satisfactory, the searcher selects another US patent document as a seed document while checking the search result.
By performing a search again, more appropriate US patent documents can be accessed.

As is clear from the first and second embodiments, the present invention provides a method of Japanese-language knowledge based on word-level knowledge with reference to a dictionary, even if foreign language grammar and document creation knowledge are insufficient. Patent documents and patent documents in foreign languages can be searched simultaneously.

Further, in the description of the embodiment of the present invention, an example is shown in which a patent family is used as documents having the same article content scattered in a Japanese literature database and a foreign language literature database. Is not limited to, but widely available academic papers,
Newspaper articles or PAJ (Patent Abstract of Jap)
It is obvious that the present invention can be applied to a database including documents in which the same content sentence such as an) is translated into a language other than the native language.

[0047]

The cross-language concept search system of the present invention has an effect that documents in many languages can be searched at the same time simply by creating a question formula in one language.

Further, the cross-language concept search system of the present invention has an effect that if there is word-level knowledge of a foreign language other than the native language, documents in various languages can be obtained more precisely and simultaneously. .

Further, the cross-language concept retrieval system of the present invention retrieves corresponding US patent documents of the patent family from Japanese patent documents obtained by the Japanese patent retrieval system, and converts the US patent documents into seed documents. (Question text)
Is searched in the US patent database. The U.S. Patent Database contains a large number of Japanese patents that do not have families, that is, a large number of U.S. patent documents that are not covered by the Japanese patent search system. This has the effect that documents can be found.

Further, the cross-language concept search system of the present invention, when the result of the first search is not satisfactory,
By changing the term weight by the searcher, the meaning of the seed document is adjusted, a more appropriate question sentence is created, and a more appropriate access to US patent documents can be obtained.

Further, the cross-language concept search system of the present invention, when the result of the first search is not satisfactory,
By re-selecting another U.S. patent document as a seed document while looking at the search results, and re-executing the search,
It is possible to obtain an effect that a more appropriate US patent document can be accessed.

[Brief description of the drawings]

FIG. 1 is a block diagram for explaining a first embodiment of the present invention.

FIG. 2 is a flowchart for explaining a first embodiment of the present invention.

FIG. 3 is an explanatory diagram, together with FIG. 4, for explaining how to change the term weighting using an actual example.

FIG. 4 is an explanatory diagram together with FIG. 3 for explaining how to change the term weighting using an actual example.

FIG. 5 is a block diagram for explaining a second embodiment of the present invention.

FIG. 6 is a flowchart for explaining a second embodiment of the present invention.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Search query input device 2 Japanese patent database 3 Text search device 4, 10 Text display device 5 Corresponding patent related database 6 Corresponding patent search device 7 Conforming patent confirmation device 8 Similar document search device 9 US patent database 11 English-Japanese automatic translation device 12 Seed document semantic adjustment device 13 Re-search instructing device 21 Patent (Japanese) information search system 22 Conforming patent selection system 23 US similar document search system

 ──────────────────────────────────────────────────続 き Continued on the front page (72) Inventor Shigeto Higuchi 4-7-1, Toyo, Koto-ku, Tokyo Sato Daiya Building F-term in Japan Patent Information Organization (Reference) 5B075 ND03 PP23 PQ02 PQ36 PQ72 PR06 QM05 QM08

Claims (5)

[Claims] 1. A system in which a plurality of similar document search systems are constructed to search for different language documents, and each similar document search system includes languages in which documents of the same content described in each language are scattered. In the cross-sectional concept search system, a document having the same content described in each language corresponding to the document present in a search result for a search query in one language is extracted, and a document having the same content is searched for in each language. A cross-language concept search comprising: a corresponding document search device that outputs as a document; and a similar document search system that simultaneously searches for similar documents in a plurality of languages using the output search type document. system. 2. A system in which a plurality of similar document search systems are constructed so as to search for different language documents, respectively, and each similar document search system includes languages in which documents of the same content described in each language are scattered. In the cross-sectional concept search system, a document having the same content described in each language corresponding to the document present in a search result for a search query in one language is extracted, and a document having the same content is searched for in each language. A corresponding document search device that outputs as a document, a similar document search system that searches for similar documents in multiple languages at the same time using the output search type document, and a selection is made from search results from the similar document search system. A cross-language concept retrieval system characterized by comprising a seed document semantic adjustment device that adjusts the meaning of a document so that a similar document can be searched again. . 3. A similar document search system constructed to search for different language documents, each including a plurality of similar document search systems, wherein each of the similar document search systems is a language in which documents of the same content described in each language are scattered. In the cross-sectional concept search system, a document having the same content described in each language corresponding to the document present in a search result for a search query in one language is extracted, and a document having the same content is searched for in each language. A corresponding document search device that outputs as a document, a similar document search system configured to simultaneously search for similar documents in multiple languages using the output search type document, and a search result from the similar document search system. A cross-language concept search system comprising a re-search instruction device for selecting an optimum seed document and performing a similar document search again. 4. The cross-language concept search system according to any one of claims 1 to 3, further comprising a text display device for displaying a search result. A cross-language concept retrieval system characterized by: 5. A cross-language concept search system according to any one of claims 1 to 4, wherein the cross-language concept search system further comprises any one language from any other language. A cross-language concept search system comprising an automatic translation device for translating into a language.