JP2003223465A - Patent document retrieval method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a patent document retrieval system or a patent document retrieval method classifying words into a synonym dictionary and a homonym dictionary essential in a retrieval using a natural language as a key and reducing leak and noise. <P>SOLUTION: As for index terms for every documents accumulated along with field information in a text index file 10, the synonym is standardized and classified by a synonym dictionary 11 by field, to which the document is pertained, and the emergence frequencies of the index terms for every fields and documents are accumulated in an index term emergence frequency file 12. A clustering/ranking arithmetic part 14 receives a retrieval response set of the classified retrieval stored in a retrieval set storage part 8, calls emergence frequency data of the index terms for every hit document by the index term emergence frequency file 12, and performs clustering with a source document or raking using the source document as an index according to the command of a user. <P>COPYRIGHT: (C)2003,JPO

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a patent document search system capable of displaying a search set which is a search result in a ranking display or a clustering display in a patent document search.

[0002]

2. Description of the Related Art In a full-scale search of patent documents, a classification search key is used as an indispensable item. Since the classification search key searches patent documents by concept matching without depending on individual languages, language-specific problems in information search, i.e., fluctuations in writing, synonyms and synonyms, noise and more. Since it is possible to avoid, it is said that a highly accurate search with less noise and leakage is possible.

On the other hand, however, in the search by classification, the tendency of patent applications changes with the change of technology, so that it is necessary to develop the patent classification up to that point and reform the classification system. The burden of having to re-assign a new classification to the documents that have been classified into. For this reason, it is said that there is a limit economically and physically in maintaining an appropriate amount of narrowing down the search by classification.

In consideration of such a situation, Japanese Patent Application No. 2001
As disclosed in Japanese Patent Application No. 357469 (hereinafter referred to as a prior application), a hybrid-type search method has been proposed which utilizes the advantages of a classification search and combines the high economic efficiency of a search key in natural language.

In the above-mentioned prior application, as shown in FIG. 1, the answer set created by the F-term search (patent document search system adopted by the Patent Office) is similar to each document and the seed document by using the seed document to be searched as an index. The degrees are compared, and ranking or clustering display is performed according to the degree of similarity. Note that the seed document (theme sentence) is a free expression of the item to be searched for in ordinary sentences using natural language.
In the case of examination of a patent application or prior art search of an objection, the explanation part extracted from the examination case, the whole sentence, or the sentence made from this can be used as the seed document.

[0006] By the way, the search key consisting of natural language corresponds to a free expression of the creator of the document, instead of the economical burden of creating a classification table in advance and assigning a classification like classification search. There is a problem that it has to be, and this can be summarized as a problem of how to arrange synonyms and synonyms.

[0007]

A simple example of synonyms will be given. In one patent specification, "golf club head, shaft" is used in another patent specification, "golf club head, handle" or "game equipment head, handle".
And another form of notation (natural language, word).

Similarly, in the patent specification, words having the same notation but different meanings, that is, synonyms are often used. "Golf club head, shaft", "engine (cylinder) head, (crank) shaft", "hard disk (seek) head, shaft for sliding seek head" and the like are examples thereof.
In other words, even if the same notations such as “shaft” and “head” are used, what is meant (meaning) is different beyond the technical field. There are so many examples of such synonyms, especially in terminology.

In a search using a natural language as a key, the above-mentioned synonyms and synonyms are included in the documents to be searched as they are, so that the synonyms cause more and the synonyms cause noise. As described above, in a search using a natural language as a key, it is indispensable to organize a synonym dictionary and a synonym dictionary in order to prevent leakage and noise. However, even if the head and the shaft are synonymous with the head and the handle, respectively, they are synonyms in the field of golf clubs and cannot be applied to the fields of engines and hard disks. This means that synonyms must be organized by field. In other words, it means that unless a synonym dictionary is created for each field and the synonyms of documents belonging to the field are unified using the dictionary, the leak of search by the synonym cannot be eliminated.

Regarding the synonyms, as can be seen from the above-mentioned examples of synonyms, most of the synonyms of technical terms exist between different technical fields. The main groups and subgroups of the International Patent Classification (IPC) currently in use (both are collectively referred to as groups).
Looking at one theme field of F term and F term, one technical field in which synonyms are hardly mixed is formed therein, and it is not substantially necessary to consider the measures against synonyms. By creating a synonym dictionary for each field with the IPC group or the F-term theme field as a unit and unifying and organizing the words that appear therein for the patent documents belonging to that field, the ranking is more accurate than in the prior application technology. Or clustering is possible.

Therefore, the present invention aims to provide a patent document search system or a patent document search method that enables highly accurate ranking / clustering in a patent document search for displaying a ranking / clustering display of answer sets of a search by classification. It is what

[0012]

The above-mentioned problems can be solved by the following means. That is, the solution means of the first invention compares the similarity of the answer set created by the search by classification with the seed document that is the survey target as an index, and displays the ranking or clustering according to the similarity. In the hybrid search, after synonyms are unified and organized by referring to the synonym dictionary dedicated to the field for all the words of the documents forming the answer set, similarity comparison is performed, and ranking or ranking is performed according to the similarity. A patent document search method characterized by clustering.

According to a second aspect of the present invention, the answer sets created by the search by classification are compared for similarity by using the seed document that is the object of survey as an index, and ranked or clustered according to the similarity. In the hybrid search, the synonyms are unified and organized by referring to the synonym dictionary dedicated to the relevant field for all the words of the document forming the answer set, and the group of the unified and organized words is used in the technical field. Is a patent document search method characterized by performing similarity comparison after removing words that no longer show characteristics, and ranking or clustering according to this similarity.

The solution means of the third invention compares the similarity of the answer sets created by the search by classification with the seed document to be surveyed as an index, and displays the ranking or clustering according to the similarity. In the hybrid search, the synonyms are unified and organized by referring to the synonym dictionary dedicated to the relevant field with respect to the index words that characterize each document forming the answer set, and then the similarity comparison is performed to determine the similarity. According to the patent document search method, ranking or clustering is performed according to the ranking.

According to a fourth aspect of the invention, the answer sets created by the search by classification are compared for similarity by using the seed document which is the survey target as an index, and ranked or clustered according to the similarity. In the hybrid search, the synonyms are unified and organized by referring to the synonym dictionary dedicated to the relevant field with respect to the index terms that characterize each document that forms the answer set, and the synonyms are grouped into This is a patent document search method characterized by removing index words that no longer show features in the technical field, performing similarity comparison, and ranking or clustering according to this similarity.

The solution means of the fifth invention refers to a means for searching patent documents by classification and a synonym dictionary dedicated to the relevant field for all the words of the patent documents forming the answer set of the above search. Means for unifying and organizing synonyms,
A means for comparing the similarities by using the seed document that is the object of the survey as an index, a means for ranking or clustering according to the similarity, and a patent document or a patent document and a patent document that are ranked or clustered are displayed. It is a patent document search system characterized by comprising:

The solution means of the sixth invention refers to a means for searching patent documents by classification and a synonym dictionary dedicated to the relevant field for all the words of the patent documents forming the answer set of the search. Means for unifying and organizing synonyms,
In order to compare the degree of similarity with the means for removing the words that have no characteristics in the technical field from the group of words that have been unified and arranged, and the seed document that is the object of the survey as an index. And a means for ranking or clustering according to the degree of similarity, and a patent literature or patent literature that has been ranked or clustered and means for displaying a seed document. System.

The solution means of the seventh invention is a means for searching a patent document by classification and an index word characterizing the patent document among all the words of each patent document forming the answer set of the search. On the other hand, a means for unifying and organizing synonyms by referring to a synonym dictionary dedicated to the field concerned, and a means for comparing the similarities using the seed document as the survey target as an index,
A patent document search system comprising means for ranking or clustering according to the degree of similarity and means for displaying ranked or clustered patent documents or patent documents and seed documents.

The solution means of the eighth invention is a means for searching a patent document by classification and an index word characterizing the patent document among all the words of each patent document forming the answer set of the search. On the other hand, the means for unifying and organizing synonyms by referring to the synonym dictionary dedicated to the relevant field, and the group of index terms that have been unified and unorganized, no longer show the characteristics within the technical field. Means for removing the index words, means for comparing the similarities using the seed document that is the survey target as an index, means for ranking or clustering according to the similarities, and ranking or clustering. A patent document or a patent document retrieval system comprising a patent document and means for displaying a seed document.

[0020]

DETAILED DESCRIPTION OF THE INVENTION According to one invention disclosed in this specification, an answer set created by a search by classification is used to search a seed document (examination of a patent application, prior art search of an objection) to be searched. In the hybrid search that compares the degree of similarity between each document and the seed document by using (in case of examination) as an index, and displays ranking or clustering according to the degree of similarity, a synonym dictionary dedicated to the field that formed the answer set The accuracy of ranking or clustering can be improved by unifying and organizing the index terms of documents belonging to the relevant field.

Depending on the index words in the answer document set, as a result of being unified and organized by the synonym dictionary, some words may appear in almost all documents forming the answer set. In other words, when looking at all technical fields, there are words that, even if they are index words, do not show any characteristics within the technical field. Such words are
It is hard to say that the document is characterized in the answer set, and the value as a word is considered to be extremely low. It is preferable to perform similarity comparison for words that appear in substantially all documents (all documents in the field) except for these. As a result, the differences between the documents can be highlighted and more effective distinction can be performed.

According to another invention disclosed in this specification,
The answer sets created by the classification search are compared for similarity using the subject sentence (examination of patent application, examination case in the case of prior art search of opposition) as an index,
In a hybrid search that displays a ranking or clustering, the synonym dictionary of the field that formed the answer set unifies and sorts the index terms of documents that belong to the field, and then extracts only the words that characterize the document again, In other words, words that do not show features in the technical field are removed from the group of index words that are unified and organized,
By using the removed words as a group of new index words (field index words), the accuracy of ranking or clustering can be improved.

FIG. 2 is a schematic diagram of the system of the first embodiment. It will be described with reference to FIG. Patent publication, patent publication 1
Are input to the patent document accumulation file 3 from the input unit 2,
Accumulated. The classification search index creation unit 4 extracts articles related to patent classification for each document number (application number, publication number, etc.) from the patent document storage file 3 and creates a classification index for each document number. The created classification index for each document number is accumulated in the classification index file 5 in the inverted file format. In the search system, the input / output unit 6
This is equivalent to an input / output terminal of a normal computer system for inputting search queries and seed documents, and displaying answers of search results.

The arithmetic unit 7 receives the query expression input from the input / output unit 6, calculates the document number including the classification corresponding to the question from the inverted file of the classification index file 5, and stores it in the search set storage unit 8. Output. The search set storage unit 8 stores the search answer set, and outputs and displays the patent documents accumulated in the patent document accumulation file 3 in response to the user's document output request from the input / output unit 6.

On the other hand, the text search index creating section 9
Extracts an article part relating to the invention for each document number from patent publications, patent publications, etc., and cuts out words using morphological analysis software. TF-I out of the extracted words
The DF method extracts the index word for each document and calculates the appearance frequency of the index word to create the index word index together with the field information. The index word index is stored in the text index file 10 in the inverted file format.

Although not shown here in order to avoid complication, when entering a text,
In response to the search query by the combination of the index words from the output unit 6, the calculation unit 7 selects the document number including the combination of the index words corresponding to the query from the text index file 10. The document number which is the selected search result is stored in the search set storage unit 8 and is also input / output unit 6.
In response to the user's request for document output from, the patent documents stored in the patent document storage file 3 are output and displayed.

The index words for each document stored together with the field information in the text index file 10 are synonymously organized by the field-specific synonym dictionary 11 to which the document belongs, and the frequency of appearance of the index word for each field / document. Is stored in the index word appearance frequency file 12 as index word appearance frequency information.

The index word appearance frequency calculation unit 13 cuts out words from the seed document input by the user from the input / output unit 6 by using the morphological analysis software used in the text search index creation unit 9 and the index word appearance frequency file 12 The index word is selected and the appearance frequency of the index word is calculated with reference to, and sent to the clustering / ranking calculation unit 14.

On the other hand, the clustering / ranking calculation unit 14 receives the search answer set (hit documents) of the classification search stored in the search set storage unit 8 and the index word appearance for each hit document from the index word appearance frequency file 12. The frequency data is called, and clustering is performed together with the seed document or ranking is performed using the seed document as an index according to a user's instruction.

The clustering or ranking result is displayed on the input / output unit 6 and is used by the user. If a user finds a patent document (seed patent document) that is closer to the information requested from the seed document in the hit documents as a result of the user's scrutiny, if the seed patent document is designated by the input / output unit 6, the clustering / ranking calculation unit In 14, the similarity calculation is performed again using the patent document of that kind, and the ranking is displayed. If the user desires clustering, clustering is performed again on the set excluding the seed document, and the result is displayed.

As described above, in the seed document, the matters to be searched are described in sentences, and in addition to the seed document, the subject of examination such as ordinary patent examination, opposition, invalidation trial, etc. If the application description as is determined and is published as a publication, it can be used as a seed document by specifying the document number after the classification search.

FIG. 3 is a schematic diagram of the system of the second embodiment. The same blocks as those in the embodiment of FIG. 2 are denoted by the same reference numerals, and the description of the overlapping portions will be omitted. Second
The field index word appearance frequency calculation unit 15 added to the embodiment (FIG. 2) stores the index word appearance frequency data of each document of the document set including the search answer set and the seed document stored in the search set storage unit 8. It is called from the index word appearance frequency file 12 and the index word appearance frequency calculation unit 13, and the TF-IDF is called again.
Using the method, it selects the field index word, calculates the frequency of the field index word, and sends the field index word appearance frequency data to the clustering / ranking calculation unit 14. The clustering / ranking calculation unit 14 uses the field index word appearance frequency data to perform clustering / ranking.

In the above description, synonyms are unified and organized for the index words, but the synonyms are unified and organized by the synonym dictionary for all the words of the document forming the answer set. It is also possible to do so.

FIG. 4 shows an image of the procedure for creating a synonym organization patent dictionary by field. The patent document-word matrix is created as shown in the table above with the subgroup or main group (IPC group) of the international patent classification as a unit. This table shows the frequency of appearance of words that appear in each document in a virtual document of patent document-word matrix in units of subgroup A63B53 / 02. Among these words, there are synonym candidates that can be treated as synonyms in the field of A63B53 / 02 (a golf club head-shaft coupling structure). The synonym patent dictionary by field below is a unified and organized list of synonym candidates by adding the frequency (number) of occurrences of synonyms in this field.

In the unified organization, words with a high frequency of appearance that appear in a biased manner in a specific small number of documents in the A63B53 / 02 field (these words are index words that characterize the documents in the A63B53 / 02 field, that is, field index words). )
To target. The extraction of specific field index words is A63B
TF-I targeting the literature set belonging to 53/02 field
The DF method is used.

A synonym expansion is performed on the extracted field index word using a synonym dictionary or a thesaurus, and A63B53 /
The corresponding synonym groups are extracted from all the words that appear in the 02 field, and these are unified and organized. However, there are patent-specific synonyms that do not exist in the synonym dictionary or thesaurus, so a visual check process is added to complete the field-specific synonym organization dictionary.

As shown in the synonym organization dictionary by field, as a result of unifying and synonymous synonyms, as a result of unifying the play equipment with the golf club, the head with the head, and the handle with the shaft, A63B
It may be a word that appears in all documents in the 53/02 field. Such a word has extremely low value as an index word in the field of A63B53 / 02. Similarly, the words appearing in the titles of A63B53 / 02 and their upper classifications are also low in value as words that characterize documents in the field.

Although the IPC subgroup is used as a unit here, synonyms can be unified by using one theme of the main group or the F term as one field unit. Even in this case, there is no problem because one theme of the main group or F term constitutes a uniform technical field.

FIG. 5 shows the subgroup A63B53 / 02.
As a result of the patent dictionary in which the words that characterize the literature in the relevant field, that is, the field index word, and the above-mentioned visual process are subordinated to,
It shows an image of a confirmed synonym dictionary for each field. In FIG. 5, the word indicated by * is the above A63B53.
/ 02 indicates that the word appears in almost all documents in the field.

[0040]

In the patent document search method or the patent document search system according to the present invention, the classification search is first performed by the patent classification, and the documents belonging to the relevant field are referred to by referring to the synonym dictionary dedicated to the field in which the answer set is formed. After the index terms of are unified and organized, the similarity is compared and ranked or clustered, so that patent document search with less occurrence of leakage and noise can be performed. Further, since the synonym dictionary is organized for the technical field only, it is possible to perform a patent document search with less leakage and noise.

[Brief description of drawings]

FIG. 1 is a simple explanatory diagram for explaining a prior application.

FIG. 2 is a system schematic diagram of the first embodiment.

FIG. 3 is a system schematic diagram of a second embodiment.

FIG. 4 is a diagram showing an image of a procedure for creating a synonym organization patent dictionary by field.

FIG. 5 shows an image of a determined synonym dictionary for each field.

[Explanation of symbols]

1 Patent publication 2 Input section 3 Patent document accumulation file 4 Classification search index creation section 5 Classification index file 6 Input / output section 7 operation part 8 Search set storage 9 Text search index creation section 10 Text index file 11 Synonym dictionaries by field 12 Index word frequency file 13 Index word frequency calculator 14 Clustering / Ranking calculator 15 Field index word frequency calculator

   ─────────────────────────────────────────────────── ─── Continued front page    (72) Inventor Michihiro Isoda             2-2-12 Akasaka, Minato-ku, Tokyo NT             IT Data Technology Co., Ltd. F term (reference) 5B075 ND20 ND34 NK35 QM08 UU06                       UU40

Claims (8)

[Claims] 1. A hybrid search in which an answer set created by a classification search is compared in similarity by using a seed document that is a survey target as an index, and ranking or clustering is displayed according to the similarity. The feature is that after all synonyms are unified and organized by referring to the synonym dictionary dedicated to the relevant field for all the words of the documents forming the set, similarity comparison is performed, and ranking or clustering is performed according to this similarity. Patent literature search method. 2. A hybrid search in which the similarity of an answer set created by a classification search is compared using a seed document that is a survey target as an index, and ranking or clustering is displayed according to the similarity. The synonyms are unified and organized by referring to the synonym dictionary dedicated to the relevant field for all the words of the documents that form the set, and the group of the unified and organized words no longer show the characteristics in the technical field. A patent document search method characterized by performing similarity comparison after removing unacceptable words, and ranking or clustering according to the similarity. 3. A hybrid search in which an answer set created by a classification search is compared for similarity using a seed document that is a survey target as an index, and ranking or clustering is displayed according to the similarity. For the index words that characterize each document forming a set, refer to the synonym dictionary dedicated to the relevant field to unify and organize synonyms, then compare the similarities, and rank or cluster according to the similarities. Patent document search method characterized by the following. 4. A hybrid search in which an answer set created by a search by classification is compared for similarity using a seed document that is a survey target as an index, and ranking or clustering is displayed according to the similarity. For index words that characterize each document that forms a set, synonyms are unified and organized by referring to a synonym dictionary dedicated to the relevant field, and from this group of index words that have been unified and organized, it is no longer characteristic within the technical field. A patent document search method characterized by performing a similarity comparison after removing index words that do not indicate "," and ranking or clustering according to the similarity. 5. A means for searching for patent documents by classification and a synonym dictionary dedicated to the relevant field for all words of the patent documents forming the answer set of the search is unified and organized. Means, means for comparing the similarities using the seed document that is the object of the survey as an index, means for ranking or clustering according to the similarity, and patent documents or patent documents ranked or clustered And a means for displaying a seed document. 6. A means for searching for patent documents by classification and a synonym dictionary dedicated to the field for all the words of the patent documents forming the answer set of the search is unified and organized. Means for removing words that are no longer characteristic in the technical field from the group of words that have been unified and organized as described above, and are similar using the seed document that is the survey target as an index. A means for comparing degrees, a means for ranking or clustering according to the degree of similarity, and a means for displaying patent documents or patent documents and seed documents that have been ranked or clustered, and means for displaying seed documents Patent literature search system. 7. A means for searching for patent documents by classification, and a reference to a synonym dictionary dedicated to the relevant field for index words that characterize each patent document forming the answer set of the search, to unify the synonyms. A means for organizing and organizing, a means for comparing the similarities using the seed document that is the object of the survey as an index, a means for ranking or clustering according to the similarities, a patent document ranked or clustered, or A patent document search system comprising a patent document and means for displaying a seed document. 8. A means for searching for patent documents by classification, and a synonym dictionary for each index document that forms an answer set of the above search is referred to a synonym dictionary dedicated to the relevant field to unify the synonyms. A means for organizing and organizing, and a means for removing the index terms that are no longer characteristic in the technical field from the group of index terms that have been unified and organized, and the species to be investigated. A means for comparing the similarities using the documents as indexes, a means for ranking or clustering according to the above-mentioned similarities, and a means for displaying patent documents or patent documents and patent documents and seed documents that have been ranked or clustered A patent document retrieval system characterized by being provided.