JP4177070B2 - Document search device - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a document search apparatus that searches a document requested by a user efficiently and without omission.
[0002]
[Prior art]
A conventional example will be described below.
§1: Conventional example 1
(1): With respect to document search devices, a method in which a user specifies a keyword as a search condition and presents a document including the specified keyword is widely used. However, it is rare that the user's true search purpose is to search for the “document including the input keyword” as described above, and the user assumes a phrase included in the “document that the user wants to search”. Generally, a target document is searched indirectly by inputting a keyword as a keyword. Focusing on this point, the conventional search method has the following two problems.
[0003]
(2): The first problem is that when a user's target document does not include a keyword assumed by the user, the document cannot be searched. In order to solve this problem, a method is widely used in which synonyms and thesaurus information are used to search for synonyms, broader terms, or narrower terms of user input keywords.
[0004]
However, when this method is used, it is necessary to prepare synonym information and thesaurus information related to a keyword that is supposed to be input, which causes another problem that it is very difficult. As an invention for solving this problem, the invention of Patent Document 1 is raised.
[0005]
This invention prepares a co-occurrence case database that describes a dependency relationship including a word and its semantic concept dictionary and case information between concepts, and if an unregistered word that does not exist in the concept dictionary appears, Search the co-occurrence example database with the word concept and its case information included in the clauses that have a dependency relationship with the clause containing the registered word, and find the semantic concept of the word that is at the position of the unregistered word in the co-occurrence case that is the search result The gist is that the user is extinguished as a candidate for a semantic concept of an unregistered word, the user selects a correct semantic concept, and the result is newly registered in the concept dictionary.
[0006]
However, the word may have ambiguity, and the semantic concept obtained only in one co-occurrence case is used for the analysis of all subsequent sentences even if the user's confirmation phase is added. The ambiguity may not be noticed, and there is a risk of incorrect registration.
[0007]
(3) The second problem is that unnecessary documents other than the user's target document are also retrieved. A keyword search is an operation in which a word that is likely to be included in a document whose contents are to be searched by a user is input as a keyword. However, even if the document includes an input keyword, the actual description content is innumerable. This is because documents other than the description contents assumed by the user are hit.
[0008]
For example, if a user enters “Free & Piano Class” as a keyword for the purpose of knowing information related to “Free Piano Class”, “The result of the lesson in the piano class was performed at a free public concert The sentence "" hits the search condition, but does not match the user's search intention.
[0009]
On the other hand, before presenting the search document itself, the part before and after the input keyword in the sentence including the input keyword in the document is presented, and the user suppresses unnecessary access to the original document by viewing it. Ingenuity has been made. However, in this method, the user needs to look through all of the presented information after all, and cannot narrow down using the presented information.
[0010]
In addition, as a method of searching for a document including a sentence having a description content assumed by a user, a method in which the user inputs a sentence as a search condition and searches a stored document for a sentence having the same meaning as the sentence. Is widely used. However, in this case, searching for a document containing exactly the same description as the input sentence can be easily realized by using a full-text search technique. However, it is not possible to search for a document containing a sentence having the same sentence meaning but a different surface expression.
[0011]
In other words, in order to realize a search with less leakage, processing that absorbs the difference in surface expression is required. As a means for such an absorption process, a method is widely used in which a sentence is decomposed into a dependency group to absorb excess or deficiency of sentence meaning information and variations in word order. As an example of the invention using this means, Patent Document 2 can be cited. The present invention will be described below.
[0012]
(4): Description of the invention described in Patent Document 2
In the invention described in Patent Document 2, when “search text” is specified as an input, the dependency relationship between two clauses “text” and “search” is analyzed, and the synonyms of words constituting the clause The search is performed by absorbing fluctuations in expressions such as word information and passive information, and sentences such as “search for documents” and “search for text” are searched.
[0013]
However, in the present invention, there is a problem that there is a document that cannot be searched only by conversion at the word level or phrase level used. For example, in this method, a sentence having the same content as an input sentence such as “search text” cannot be searched. This problem has the same contents as the input sentence, but the essence is that there is a sentence with a different dependency structure.
[0014]
As an invention for solving a part of this problem, the invention of Patent Document 3 is raised. The present invention will be described below.
[0015]
(5): Description of the invention of Patent Document 3
The present invention has the structure of embedded sentences (sentences whose predicates are jointly modified to the main case) and extended case structures (“(main case) is (predicate-derived noun) (verb clause representing execution)”. Sentence) is an invention that makes it possible to search. As a result, in response to the input sentence “Developed by Yamada Denki”, a sentence including the descriptions “Yamada Denki that developed ~” (embedded sentence) and “Yamada Denki completed development” (sentence of the extended case structure) The gist of this is to enable search. However, the variation of the dependency structure does not occur only by these, and cannot cope with a sentence such as “Development at Yamada Electric is going well”.
[0016]
On the other hand, as a document search technique, a technique called directory search is widely known in addition to the keyword search. This method is characterized by an interface in which a hierarchical classification category is assigned to a search target document and the user arrives at a target document by selecting the presented classification category.
[0017]
In this interface, for example, when searching for documents related to “barrier-free”, the classification category name that is presented first, “society”, lower-level classification category names such as “welfare” and “barrier-free” are selected. The operation of arriving at the document.
[0018]
In this method, since the search is performed by the contents of the document independently of the description sentence on the surface layer of the document, the above-described problems associated with the keyword search do not occur, but in which classification category the document requested by the user belongs. Another problem arises that it is necessary to infer whether it is classified, and if the classification category system is not grasped, search cannot be performed well.
[0019]
In the above example, if you select another category without selecting “Welfare”, which is the category name leading to the correct answer, among the classification categories presented when you select the classification category “Society” In some cases, “barrier free” does not exist in the lower category, and the search fails.
[0020]
[Patent Document 1]
JP-A-6-195371
[Patent Document 2]
JP-A-3-20866
[Patent Document 3]
Japanese Patent Laid-Open No. 7-244669
[0021]
[Problems to be solved by the invention]
In the prior art as described above, there are the following problems when the user performs a search correctly with the description content of the search target document.
[0022]
(1): It is not possible to simply search for a sentence including a keyword input by the user, but to present information for searching for a sentence that matches the user's search intention as information for the next narrowing operation. . Further, in the conventional directory type search method, information for the next narrowing-down operation is presented, but the information presented here is not dynamically generated from a user's search viewpoint.
[0023]
{Circle around (2)} A sentence described using a synonym of an input keyword cannot be made searchable without preparing a synonym dictionary in advance.
[0024]
The present invention solves such a conventional problem,
(1): Instead of simply searching for a sentence including a keyword input by a user, information for searching for a sentence that matches the user's search intention can be presented as information for the next narrowing operation. Like that. The information presented at that time is not set in advance as in the conventional directory type search method, but is dynamically generated from the user's point of view.
[0025]
{Circle around (2)} An object is to realize at least the above two points that a sentence described using a synonym of an input keyword can be searched without preparing a synonym dictionary in advance.
[0026]
[Means for Solving the Problems]
In order to achieve the above object, the present invention is configured as follows.
[0027]
(1): In a document search device having a user interface and a keyword search engine, a phrase including an input keyword; The keyword obtained by dependency analysis A refined information creating unit for creating and presenting refined information based on a clause that is a source or an affiliated person, and a dependency information search engine that retrieves a document based on information selected by the user from the presented refined information It is characterized by having.
[0028]
(2): In the document search apparatus according to (1), the narrowed-down information creating unit has a function of creating narrowed-down information based on the type of phrase presented together with the phrase including the input keyword. It is characterized by.
[0029]
(3): In the document search device according to (1) or (2), the narrowing-down information creating unit creates a hierarchical structure by collecting common phrases among the phrases presented as narrowing information, and The narrowing information to be presented has a function of presenting the hierarchical structure.
[0030]
(4): In the document search device according to any one of (1) to (3), the narrowed-down information creating unit is based on a search result in the selected narrowed-down information based on another dependency relationship designated by the user. It has a function of creating narrowed-down information and presenting it as new narrowed-down information.
[0031]
(5): In the document search device according to any one of (1) to (3), according to one or more search viewpoints prepared in advance, a sentence characterizing the content of a search target document that matches the search viewpoint or A subject extraction unit that extracts a part of a sentence together with a search viewpoint, and a user interface for a user to input the search viewpoint at the time of search, and a narrowing information to be created from a subject description that matches the input search viewpoint It is characterized by having a refined information creating unit that creates from what is acquired.
[0032]
(Function)
The operation of the present invention based on the above configuration will be described with reference to FIG.
[0033]
(a): In the above (1), the refinement information creating unit 5 includes a phrase including the input keyword, The keyword obtained by dependency analysis Narrow down information is created with the phrase that becomes the source or destination, and is presented to the user. The dependency information search engine 6 searches the document based on the information selected by the user from the presented narrowing information.
[0034]
In this way, it is possible to present information for searching for a sentence that matches the user's search intention as information for the next narrowing operation, instead of simply searching for a sentence including the keyword entered by the user. become. The information presented at that time is not preset as in the conventional directory-type search method, but can be dynamically generated from the user's search viewpoint. In addition, since a word that is a synonym is also presented, a sentence described using the synonym of the input keyword can be searched without preparing a synonym dictionary in advance.
[0035]
(b): In the above (2), the narrowing information creating unit 5 creates narrowing information based on the type of phrase to be presented together with the phrase including the input keyword.
[0036]
In this way, instead of simply searching for a sentence containing the keyword entered by the user, information for searching for a sentence that matches the user's search intention is presented to the user as information for the next refinement operation. become able to. The information presented at that time is not preset as in the conventional directory-type search method, but can be dynamically generated from the user's search viewpoint. In addition, since a word that is a synonym is also presented, a sentence described using the synonym of the input keyword can be searched without preparing a synonym dictionary in advance.
[0037]
(c): In the above (3), the narrowed-down information creating unit 5 creates a hierarchical structure by collecting common phrases among the clauses presented as the narrowed-down information, and the hierarchical structure as the narrowed-down information presented to the user Present.
[0038]
In this way, it is possible to present information for searching for a sentence that matches the user's search intention as information for the next narrowing operation, instead of simply searching for a sentence including the keyword entered by the user. become. The information presented at that time is not preset as in the conventional directory-type search method, and can be dynamically generated from the user's search viewpoint. In addition, since a word that is a synonym is also presented, a sentence described using the synonym of the input keyword can be searched without preparing a synonym dictionary in advance.
[0039]
Furthermore, since the information is presented in a unified format as the narrowed-down information, there is too much information presented, so that the selection operation of the next user is not hindered.
[0040]
(d): In the above (4), the refinement information creation unit 5 creates refinement information from the search result in the selected refinement information based on other dependency relationships designated by the user, and newly refines it. Present as information.
[0041]
In this way, it is possible to present information for searching for a sentence that matches the user's search intention as information for the next narrowing operation, instead of simply searching for a sentence including the keyword entered by the user. become. The information presented at that time is not preset as in the conventional directory-type search method, but can be dynamically generated from the user's search viewpoint.
[0042]
In addition, since a word that is a synonym is also presented, a sentence described using the synonym of the input keyword can be searched without preparing a synonym dictionary in advance. Furthermore, since the process is repeated until the user obtains a sufficient narrowing result, a search with sufficient narrowing can be performed.
[0043]
(e): In the above (5), the subject extraction unit 7 determines a sentence or a part of the sentence that characterizes the content of the search target document that matches the search viewpoint according to one or more search viewpoints prepared in advance. Extract with search perspective. Further, the user interface 3 inputs the search viewpoint when the user searches. The refinement information creation unit 5 creates refinement information to be created from information obtained from the subject description that matches the input search viewpoint.
[0044]
In this way, it is possible to present information for searching for a sentence that matches the user's search intention as information for the next narrowing operation, instead of simply searching for a sentence including the keyword entered by the user. become. The information presented at that time is not preset as in the conventional directory-type search method, but can be dynamically generated from the user's search viewpoint. In addition, since a word that is a synonym is also presented, a sentence described using the synonym of the input keyword can be searched without preparing a synonym dictionary in advance.
[0045]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.
[0046]
§1: Outline of document retrieval device
The outline of the document search apparatus will be described below with reference to FIG.
[0047]
(1): In the document search apparatus according to the present invention, a conventional keyword type comprising a user interface 3 for inputting a keyword of a user and presenting a search result and a keyword search engine 4 for searching a sentence including the input keyword. In the document search apparatus, further, a refined information creating unit 5 that organizes and presents dependency information related to a phrase including an input keyword, and dependency information that searches a document based on a result selected by the user from the stopped refined information A search engine 6 and a subject extraction unit 7 are provided.
[0048]
(2): An example of each processing unit and an outline of the processing will be described by taking a search task as an example. All processes other than the user interface 3 can be realized as a computer program.
[0049]
In order to solve the problem that a temporary file that should be deleted after the end of processing is left in a software product, an example search task is a past failure case (in this case, A database (DB) in which “documents described” are stored is searched.
[0050]
(3): User interface 3
A means for inputting a search condition for the user to perform a search and a means for presenting the processing result to the user are provided, which can be realized by an input device such as a keyboard and an output device such as a monitor. In the following search task, “temporary file” is input as a keyword.
[0051]
(4): Keyword search engine 4
A sentence including a keyword input by the user interface 3 is searched, and a conventional full-text search engine or the like can be used as it is.
[0052]
§2: Description and configuration of document search device
(1): Explanation of the configuration of the document search device
FIG. 2 is a block diagram of the document search apparatus. The document search device 2 includes a user interface 3, a keyword search engine 4, a refinement information creation unit 5, a dependency information search engine 6, a subject extraction unit 7, a refinement information creation rule storage DB 8, and a search target document DB 9. And a subject extraction rule storage DB 10 and the like. The DBs are all databases.
[0053]
In the above configuration, the refinement information creation rule storage DB 8 is a database in which the refinement information creation unit 5 creates refinement information with reference to the information. The search target document DB 9 is a DB used (referenced / stored) by the keyword search engine 4 and the dependency information search engine 6. The theme extraction rule storage DB 10 is a DB that the theme extraction unit 7 refers to when extracting a theme from a search target document.
[0054]
(2): Explanation of processing example
FIG. 3 is an explanatory diagram of processing of the document search apparatus. Hereinafter, processing of the entire document search apparatus will be described with reference to FIGS. 1 and 3.
[0055]
{Circle around (1)} The narrowed-down information creating unit 5 creates narrowed-down information based on the phrase including the input keyword and the phrase that is the source or destination, and presents it to the user. The dependency information search engine 6 searches the document based on the information selected by the user from the presented narrowing information.
[0056]
{Circle around (2)} The narrowed-down information creating unit 5 creates narrowed-down information based on the type of phrase presented together with the phrase including the input keyword.
[0057]
{Circle around (3)} The narrowed-down information creation unit 5 creates a hierarchical structure by collecting common phrases presented as narrowed-down information, and presents the hierarchical structure as narrowed-down information presented to the user.
[0058]
{Circle around (4)} The narrowed-down information creating unit 5 creates narrowed-down information based on other dependency relationships designated by the user from the search result in the selected narrowed-down information, and presents it as new narrowed-down information.
[0059]
{Circle around (5)} The subject extraction unit 7 extracts a sentence or a part of the sentence characterizing the search target document that matches the search viewpoint together with the search viewpoint in accordance with one or a plurality of search viewpoints prepared in advance. Further, the user interface 3 inputs the search viewpoint when the user searches. The refinement information creation unit 5 creates the refinement information to be generated from information obtained from the subject description that matches the input search viewpoint. The contents of the arrows shown in FIG. 1 and (a) to (Z) in the figure are as follows. The arrows in FIG. 1 indicate the flow of information, (a) is a search condition such as a keyword input by the user, (g) is a search area of (a), (b) is a search result for (g), (c) is information for creating refinement information from search results, (d) is refinement information created from (b) and (c), and (e) is refinement information in presented (d). (I) is the search area of (e), and (f) is the search result for (i). As for the search result (f), new refinement information (d) is generated and presented to the user using (c) as in (b).
[0060]
In the devices of Examples 1 to 3 (corresponding to claims 1 to 3), reference information to the search target document is presented to the user from the newly generated (d), but Example 4 (corresponding to claim 4) In the case of the device of), when the narrowing down information (d) has achieved sufficient narrowing down, the reference information to the search target document in (d) is also presented to the user. Further, new selection information (e) is given to the user, and (d) is generated again based on the selection information (e).
[0061]
Further, regarding the invention of Example 5 (corresponding to claim 5) described in the other examples, this means that when the search target document DB is constructed, the search target is determined in the search processing from the keyword or dependency information. Instead of targeting all sentences included in the document, the database is constructed by limiting only the subject extraction results.
[0062]
This process is performed at the time of DB construction prior to the user's search operation. The flow of information in the process is also shown in FIG. 1, and (X) is the text information in the search target document and the search target. Reference information to the document, (Y) is a rule for subject extraction, and (z) is reference information to the subject extracted by applying (Y) to (X) and the search target document.
[0063]
(3): Explanation of processing example 1 by flowchart
FIG. 4 is a processing flowchart 1 at the time of retrieval. Hereinafter, based on FIG. 4, the processing (processing example 1) of the processing flowchart 1 at the time of search will be described. This processing example is a processing example corresponding to claims 1 to 3 and will be described with reference to FIG. S1 to S6 indicate each processing step.
[0064]
First, when a search condition (a) is input by a user operation (S1), the keyword search engine 4 searches document information in the search target document DB 9 to obtain a search result (b) (S2). Next, the refinement information creation unit 5 receives the search result (b) from the keyword search engine 4, creates refinement information with reference to the refinement information creation rule storage DB 8 (S 3), and notifies the user via the user interface 3. Present (S4).
[0065]
Next, when the presented refinement information is selected by user operation and user selection information (e) is input (S5), the dependency information search engine 6 searches the search target document DB 9 as a target. To obtain the search result (f). Thereafter, the refined information creation unit 5 receives the search result (f) from the dependency information search engine 6 and presents it to the user via the user interface 3 (S6).
[0066]
(4): Explanation of processing example 2 by flowchart
FIG. 5 is a processing flowchart 2 at the time of retrieval. Hereinafter, based on FIG. 5, processing (processing example 2) of the processing flowchart 2 during search will be described. In addition, this process example is a process example corresponding to Claim 4, and S11-S17 show each process step.
[0067]
First, when a search condition (a) is input by a user operation (S11), the keyword search engine 4 searches document information in the search target document DB 9 to obtain a search result (b) (S12). Next, the refinement information creation unit 5 receives the search result (b) from the keyword search engine 4, creates refinement information with reference to the refinement information creation rule storage DB 8 (S 13), and notifies the user via the user interface 3. Present (S14).
[0068]
Next, it is determined whether or not the narrowing is sufficient by the user (S15). If not enough, narrowing information selection is performed by the user operation (S16), and then the process proceeds to S12, and the processes from S12 are performed. Is repeated. Further, if the narrowing down by the user is sufficient, the search result is presented to the user via the user interface 3 (S17).
[0069]
§3: Detailed explanation of the refinement information creation unit
(1): Explanation of Example 1
(1): Overview
The narrowed-down information creation unit 5 uses a known dependency analysis means from the retrieved sentence and expresses it as a dependency group (a combination of a source phrase notation and a destination phrase notation. Information for the next refinement operation is created based on the dependency group.
[0070]
In addition, the dependency group used to generate the refinement information may be obtained by performing dependency analysis processing on the sentence searched as described above. However, dependency analysis is performed on all search target documents in advance. Further, it is possible to obtain the search target of the keyword search engine not by using the sentence in the document but by using the dependency group.
[0071]
When generating refined information from the acquired dependency group, it is necessary to unify the surface representation of the acquired dependency group first. This is a process for unifying dependency groups that mean the same content into dependency groups having the same surface representation. By performing this processing, it is possible to present the narrowing information that matches the purpose of the present invention to search by the description content, and to reduce the narrowing information to be presented for a certain input, thereby selecting the user The burden of processing can also be reduced. For example, the following processing can be given as the surface layer expression unification processing.
[0072]
(2): Unified processing of phrases
It is a process that absorbs the shaking of the phrase expression. In the known dependency analysis method, sentences are first decomposed into morphemes by a morphological analysis method, then they are grouped into clauses based on grammatical rules, and finally the dependency relationship is estimated based on attribute information of the clauses. Process. In this case, since the morpheme that constitutes the phrase is known in each phrase, a process of deleting a morpheme, for example, grammatically called an auxiliary particle, is performed based on the morpheme information. As an adjunct particle, for example, “mo” in the phrase “to omo” is mentioned.
[0073]
In addition to the dependency analysis processing, processing such as complementing case particles with respect to parallel clauses can also be performed by using a known parallel structure analysis means. For example, a dependency such as “delete file or −delete” or “delete −directory” is extracted from a sentence such as “delete file or directory”, but the clause “file or” and clause “ Since it is known that "directory" has a parallel structure, delete the above auxiliary particle with respect to the former clause, and complement the particle of the phrase with the parallel structure. Is a process for generating a phrase such as
[0074]
This process is also applied to the next sentence “deleting a file or directory” that expresses a parallel structure with punctuation without adverbial particles, and similarly generates a dependency group “deleting a file”. It will be a thing. Furthermore, when a clause serving as a predicate of a sentence includes a past auxiliary verb or the like, a process of deleting the auxiliary verb and converting it to the present form may be considered. As a result, the phrase “remaining” is converted to “remaining”.
[0075]
Further, in the morpheme analysis dictionary used in the morpheme analysis method, information for absorbing the fluctuation of the notation is stored, for example, even when the notation in the search target document is “interface”, Processing such as converting the notation of the morpheme constituting the clause, such as converting to “interface”, is also possible.
[0076]
(3): Explanation of unified processing of dependency groups
This process is a process for standardizing the notation as a dependency group. For example, if a clause that is a predicate has a passive auxiliary verb, a dependency group consisting of a noun equivalent word and a phrase composed of a ga case particle is used as a noun equivalent word and a wo case particle. This is a process of converting into a dependency group consisting of a clause composed of the above and an active predicate clause.
[0077]
By this processing, the dependency group “the file is deleted” is converted to “the file is deleted”. Similar processing can be performed between intransitive verbs and transitive verbs. In this case, information on verbs in the relationship between intransitive verbs and transitive verbs is necessary, but this information can be obtained from a general electronic dictionary. By this process, for example, “leave the file” is converted to “leave the file”.
[0078]
(4): Explanation of clause combination processing
This is a process of combining the dependency pairs acquired from the retrieved sentences and converting them into dependency combinations having a form useful for narrowing down. For example, processing in a sentence “cell becomes blank” is raised. This sentence has two dependency pairs, “cell becomes-” and “becomes blank”, in the dependency processing based on the normal grammar, but the verb as “being” has little information as a search key. The description content cannot be guessed only by the dependency group “cell becomes-”.
[0079]
Therefore, by combining these two and configuring a dependency group in the form of “cell becomes -blank”, the content can be grasped only by the dependency group.
[0080]
In addition, in the sentence “Information may not be saved.”, The dependency analysis result based on normal grammar is “Information is not saved”, “Not saved” is “Things are” Become. However, in this description, the main expression content is that “information is not saved”, and “something” can be regarded as adding a nuance of “not necessarily happening” to this content.
[0081]
Since such a dependency group representing auxiliary contents is meaningless as a search key, these are combined to make “information may not be saved” as dependency information. The dependency clause of this dependency group is further converted to “information is not saved” by the above processing. Also, when the original sentence is “information is never saved”, a conversion rule is set so that the information is similarly converted to “information is not saved”.
[0082]
The clauses that should be subjected to such join processing should be limited to a small number of nouns and predicates that are too wide as a narrow-down key, such as “things”, “things”, “yes”, “yes”, “no”. Well, it is easy to set rules about them.
[0083]
As a result of the above processing, the keywords entered in this search task are “temporary file-deleted”, “temporary file-remains”, “temporary file-does not delete”, “temporary file-destroys”, “directory” Dependent groups such as “temporary file” and “generate-temporary file” are presented as narrowing information.
[0084]
(2): Explanation of Example 2 (corresponding to claim 2)
In Example 2 (corresponding to claim 2), the narrowed-down information creating unit 5 has a function of creating narrowed-down information based on the type of phrase to be presented together with the phrase containing the input keyword. In this regard, it is possible to add to the user interface a function that allows the user to input a usage in the sentence of the keyword together with the keyword, and to show only the dependency pairs that match the usage as the narrowing information.
[0085]
The keyword usage is, for example, “the input keyword is related to a predicate with a particle” or “modified expression with the input keyword as a destination”. When the user inputs the former usage, the above example In FIG. 2, since the two dependency groups of “temporary file in directory” and “generated temporary file” do not match the usage specified by the user, they are not presented as narrowing information. By this process, information on the dependency relationship that is not assumed by the user is not presented, so that it is possible to suppress presentation of unnecessary narrowing information.
[0086]
Next, the generated dependency groups are collected into a format that can be easily used by the user as narrowing-down information. The simplest example of this process is to present the modification number together with the number of documents including the sentence to be acquired. In this search task, when the usage of a keyword “input keyword is related to a predicate with a particle” is specified, the keyword is formatted into a format such as “presentation result 1” in FIG. Will be presented.
[0087]
Based on the presented information, the user can select “Do not delete temporary files” and “Remain temporary files” in the current search task, and search for documents including the selected dependency group. .
[0088]
At this time, not only the expression “temporary file remains” assumed by the user in this search task but also a sentence including an expression of consent “temporary file is not deleted” is presented as narrowing information. So you can search.
[0089]
In general, “do not delete” and “remain” are not synonyms, but synonyms that are synonymous as phrases expressed in a dependency group cannot be processed with synonyms in word units. This is achieved by leaving the decision to the user.
[0090]
It is also possible to have a search mode that does not search for a document from this presentation result, but performs the same processing as when the presented phrase is input as a keyword, and again presents it as refinement information. It is. For example, when the phrase “remaining” in the refined information is selected and “noun phrase with a particle related to the predicate phrase” is specified as the usage, information such as “presentation result 2” in FIG. 7 is presented to the user. .
[0091]
In this presentation result, it can be estimated that “data.tmp” is a specific file name of “temporary file” that the user assumed as a keyword, so the dependency group “data.tmp remains” is used as a search key. By doing so, it is possible to search for documents that do not use the keyword that was initially input, but that contain sentences with the same content.
[0092]
Although the above example is the simplest presentation format, when there are many variations of clauses for the input key, there is a possibility that the process of selecting from there will be inefficient.
[0093]
(3): Explanation of Example 3 (corresponding to claim 3)
In Example 3 (corresponding to claim 3), the refinement information creation unit 5 creates a hierarchical structure by collecting common phrases among the phrases presented as refinement information, and uses the hierarchy as refinement information to be presented to the user. It has a function to present the structure.
[0094]
As described above, when there are many clause variations for the input key, there is a possibility that the process of selecting from there will be inefficient. In order to solve this, in Example 3, hierarchical summarization processing is performed on the presentation result, and the amount of information presented at one time is suppressed by presenting the category name instead of presenting each entry at once. And a function of presenting in a format that supports efficient selection by the user.
[0095]
For example, in “presentation result 2” in FIG. 7, attention is paid to the particles possessed by the phrase to be presented, the phrases with the same particle are listed together, and the category name of the particle type is presented. The presentation result in this case is as “presentation result 3” in FIG.
[0096]
Here, when the user selects the category name “˜ga”, for example, a menu “expand category” is added to the user interface screen, and when the user selects it, the specific clause “temporary file” Actual phrases such as "", "Process is", etc. are presented, and the user can further perform a selection operation from the presentation result.
[0097]
It is also conceivable to select a category name, prepare a menu “search” as another menu, and perform a document search with the selected category name. In that case, the search is performed assuming that the phrase belonging to the classification category is selected.
[0098]
As such a summary enumeration process, not only the classification category focused on the particle as described above but also various other processes can be considered.
[0099]
As an example of such a classification criterion, a category based on the matching of partial morphemes of compound words constituting noun phrases can be created. For example, when there are many variations of the phrase “~ file is” such as “file is”, “temporary file”, “data file is”, “image file is”,... When the category “file” is automatically created and the narrowing information is presented, the narrowing information once presented can be reduced by first presenting the category name “˜file is”.
[0100]
In addition, by preparing a plurality of classification category names in this way, another category name may be output as a development result of a certain category. In the case of the above example, when the classification category “toga” is first expanded, a specific clause such as “directory is” and “process is” and a category name such as “to is file” are presented to the user.
[0101]
If the category “~ file is” specified in the presentation result and further expansion is performed, the specific clauses belonging to the category are “file is”, “temporary file”, “data file”, “image file” Will be presented.
[0102]
In the above example, specific clauses of clause A and clause B are input or presented in a dependency group that can be expressed as “phrase A (˜)”-clause B (˜). The user designates and searches by selection from the narrowing down information.
[0103]
(4): Explanation of Example 4 (corresponding to claim 4)
On the other hand, in Example 4 (corresponding to claim 4), the refinement information creation unit 5 creates refinement information from the search result in the selected refinement information based on other dependency relationships designated by the user, Is provided as new refinement information.
[0104]
That is, it is characterized by further narrowing down using other dependency groups relating to the phrase A or the phrase B. This provides, for example, a function for further narrowing down when there are a large number of documents having a dependency that “temporary files remain”.
[0105]
For example, when the original sentence is “temporary file remains in directory”, not only “temporary file remains” but also “remaining in directory” can be obtained. Further refined information is created using and presented to the user.
[0106]
The user can further narrow down by selecting a newly presented result. In the search device having such a function, for example, in the first selection process, “temporary file remains” is selected, and the dependency relationship used when creating the next narrowing information is expressed as “relationship clause”. When “other dependency for” is selected, a result such as “presentation result 4” in FIG. 9 is presented.
[0107]
In this example, it is used as information for further narrowing down the dependency groups belonging to the dependency clause, but the dependency group whose temporary clause is “temporary file is”, for example, “update-temporary” It is also possible to use dependency groups such as “file” and “current-temporary file” as narrowing down information.
[0108]
Also, if it is a dependency clause of another dependency group, such as the phrase “update” presented here, further refinement information can be created with that dependency source clause. . For example, the selection information of the user at this time is “temporary file—remaining” “update-temporary file”, and the user selects to further subdivide into “update” with a clause related to “update”. In this case, narrowing-down information such as “Presentation result 5” in FIG. 10 is presented.
[0109]
In this way, with regard to a clause, the internal processing of the device that presents narrowing information with the dependency group related to the clause, and the series of processing of selecting the presentation result by the user and narrowing down the document from the selection result are sufficiently narrowed down. By repeating until the result is obtained, the target document can be efficiently searched. The presentation format to the user is not limited to the above example, and can be changed without changing the gist of the present invention.
[0110]
§4: Detailed explanation of dependency information search engine
The narrowed-down information created by the narrowed-down information creating unit 5 is presented to the user from the document set that is the search result. If the number of documents is sufficiently small, the document is searched from there, and the user interface 3 To the user. On the other hand, when the narrowing information is not sufficient and the user determines that it is necessary to further narrow down the document, the user performs an operation of selecting the presented narrowing information.
[0111]
The dependency information search engine 6 performs processing for searching for a sentence including the sentence or a document including the sentence, using the selected narrowing information. In this process, for example, a sentence including clauses constituting narrowing information is searched by full-text search, and a dependency analysis process is performed on the result to extract a dependency pair, and matching with the dependency group of the narrowing information is performed. It is possible to achieve this by presenting the matching result as a search result.
[0112]
It is also possible to extract all the dependency groups from the search target document in advance, store the dependency groups in the relational database, and present the stored relational database as a search result that matches the refinement information. Is possible.
[0113]
For example, when the top presentation result in “presentation result 5” in FIG. 10 is selected, “reservation data is updated—updated” “updated—temporary file” “temporary file remains” in one sentence. Dependency group information is created and presented to the user.
[0114]
§5: Detailed explanation of other examples (subject acquisition means, etc.)
The document search apparatus having the above configuration solves the problem, but as a document search apparatus that realizes more efficient search, “a search target that matches the search viewpoint according to one or more prepared search viewpoints” The subject acquisition means for extracting a sentence or a part of the sentence characterizing the document together with a search viewpoint, and a user interface for the user to input the search viewpoint at the time of search, and the refinement information to be generated are input There has been proposed a document retrieval apparatus having a configuration including a refinement information creation unit that is created from a subject description that matches a search viewpoint. The configuration of this apparatus will be described below. These means can be realized as a program on a computer.
[0115]
(1): Theme extraction unit
In the search task used in the explanation in the previous section, the search target document is a failure example describing the content of a defect related to a software product and the countermeasure method. Here, there are countless variations on how to describe each failure case, but there is always a specific description of the failure content, and this description is the part that is assumed as the search target when performing a user search. is there. Among various contents described in such a search target document, description contents assumed as a key when the user performs a search with the contents of the document will be referred to as a subject.
[0116]
For example, in the article sentence “A temporary file remains after cancel processing”, the actual defect description is “Temporary file remains”, and the previous part “The place where the cancel process was performed” is the defect. This is the part that describes the conditions for this to occur. When a user searches for a failure case, it is natural to search using the specific contents of the failure, and in this example, the part of “temporary file remains” is the subject of the failure case.
[0117]
As described above, the document search apparatus of Example 5 (corresponding to claim 5) includes means for extracting the subject, and processing such as keyword search and creation of narrowing-down information is performed on the part describing the subject in the document. By limiting the search, it is possible to search correctly with the contents assumed by the user.
[0118]
For example, when a user wants to search for a failure case such as “a temporary file (which should be deleted) remains”, a failure case with the description “temporary file remains as specified but data is destroyed” is searched. Should not.
[0119]
The problem with this case is that the data is destroyed, not the temporary file. According to the document search apparatus of Example 5 (corresponding to claim 5), the processing is limited to the portion describing the content of the failure, so that the processing is performed only for “data is destroyed” in this description. Therefore, this failure case is successfully eliminated.
[0120]
The means for extracting the subject from such a search target document is realized by preparing an extraction rule based on the type and characteristics of the target document and extracting a description that matches the rule. For example, if the target document is a newspaper article, this rule can be as simple as extracting the first sentence in the article.
(1) The first sentence is “(noun clause A) − (predicate clause B)”. “(Predicate clause B) − (predicate clause D)”. If it includes three dependency pairs of “(noun phrase ancestor) − (predicate phrase D)”, and the last adjunct of (predicate phrase B) ends with the connected particle “ga”, then “(noun phrase C) − (Predicate clause D) ”is extracted as the subject.
[0121]
(2) If a procedure for causing a failure is described, the sentence of the last item is extracted.
[0122]
For example, an extraction rule based on a dependency pattern of a sentence or a document structure.
[0123]
In the former rule, the description given above as an example “temporary file remains as specified, but data is destroyed” (the included dependency groups remain “as specified” but “temporary file remains”). "Remaining but destroyed" and "Data destroyed") match, so "Data destroyed" is extracted as the subject.
[0124]
In the latter rule, for example, in the case where a procedure for reproducing a failure is itemized as follows:
1. Start the update process.
2. Cancel the process immediately before the end of the process to interrupt the process.
3. Temporary files remain in the directory.
[0125]
“A temporary file remains in the directory” is extracted as a theme. Of course, a dependency analysis is performed on the sentence which is the extraction result, and a rule for selecting a subject dependency group (for example, “a noun clause related to a predicate and a noun clause related to a dual case is included. It is also possible to extract only “temporary file-remaining” as a theme by applying “deleting noun clauses in case of existence”.
[0126]
As mentioned above, some of the subject extraction rules are listed. However, it is not essential to apply the extraction rule to the sentence itself in the stored document as an example. The result of the automatic summarization process can be set as a processing target.
[0127]
Further, it is not necessary that the description extracted as the subject of the search target document is only one. For example, from the description that “the system ends abnormally and temporary files remain”, “the system ends abnormally”. ”And“ Temporary files remain ”are extracted as themes.
[0128]
Furthermore, although the above description has been given for the case where there is only one subject of the document, the description content that the user is supposed to perform a search may not be limited to one. First, a search is performed with the description content “temporary file remains”, and if the search result is large, the above description further searches in the situation where the failure that was discarded as a secondary description occurs. The operation is natural.
[0129]
The document search apparatus of Example 5 (corresponding to claim 5) enables such a search operation. In this document search device, a plurality of search viewpoints are prepared without limiting the subject to one search viewpoint (in the example, “failure content”), and the description portion is used as the subject based on each search viewpoint. This is realized by extracting the subject from the document and subjecting the subject matter that matches the search viewpoint specified at the time of the search by the user.
[0130]
The rule for performing such extraction is to extract a relevant part from a description in a document as a subject together with its search viewpoint. As an example of this rule, there is the following one that extends the rule for extracting only the subject based on the above-mentioned one search viewpoint.
[0131]
(1): The first sentence is “(noun clause A) − (predicate clause B)”, “(predicate clause B) − (predicate clause D)”, “(noun clause C) − (predicate clause D)”. If it contains three dependency pairs and the last adjunct of (predicate clause B) ends with the connecting particle “ga”, then “(noun clause A) − (predicate clause B)” is used as a search viewpoint “expression condition description”. And “(noun clause C) − (predicate clause D)” is extracted as the subject search viewpoint “failure content description”.
[0132]
(2): When the procedure for generating a fault is described in a bulleted list, the sentence of the last item is extracted as the search viewpoint “description of fault contents”, and the sentence of the previous item is extracted as the search viewpoint “expression condition description”. To do.
[0133]
As a user operation, the user first selects one search viewpoint from the prepared search viewpoints, performs a search operation, and further narrows down the search result from another search viewpoint. For example, the user selects the search viewpoint “failure content description”, and first performs a search operation using the function provided by the apparatus of the present invention to search for a failure case in which the desired failure content is described.
[0134]
When the number of cases is large, an efficient search can be realized such that a search is performed from the search viewpoint “expression condition description” with respect to the search results.
[0135]
§6: Description of specific device examples and recording media
FIG. 11 shows a specific device example. The document retrieval apparatus can be realized by an arbitrary computer such as a workstation or a personal computer. This apparatus includes a computer main body 21, a display device 22 connected to the computer main body 21, an input device (keyboard / mouse, etc.) 23, a removable disk drive (referred to as “RDD”) 24, and a hard disk device (referred to as “HDD”). 25 or the like.
[0136]
The computer main body 21 includes a CPU 26 that performs various internal controls and processes, a ROM 27 (nonvolatile memory) for storing programs and various data, a memory 28, and an interface control unit (“I / F”). 29) and a communication control unit 30 are provided. The RDD 24 includes a flexible disk drive, an optical disk drive, and the like.
[0137]
In the apparatus configured as described above, for example, a program for realizing the processing of the document search apparatus is stored in a magnetic disk (recording medium) of the HDD 25, and the CPU 26 reads out and executes the program, whereby the document The processing performed by the search device is executed.
[0138]
However, the present invention is not limited to such an example. For example, the program can be stored in the magnetic disk of the HDD 25 as follows, and the CPU 26 can execute the process by executing the program. .
[0139]
{Circle around (1)} A program (program data created by another device) stored in a removable disk created by another device is read by the RDD 24 and stored in a recording medium of the HDD 25.
[0140]
{Circle around (2)}: Data such as a program transmitted from another device via a communication line is received via the communication control unit 30 and the data is stored in a recording medium (magnetic disk) of the HDD 22.
[0141]
The following configuration is appended to the above description.
(Appendix 1)
In a document search apparatus provided with a user interface and a keyword search engine,
A refinement information creation unit that creates and presents refinement information based on the clause including the input keyword and the clause that is the source or destination,
A document search apparatus comprising a dependency information search engine for searching for a document based on information selected by a user from presented refined information.
[0142]
(Appendix 2)
The document search device according to (Appendix 1), wherein the narrowed-down information creating unit has a function of creating narrowed-down information based on the type of phrase to be presented together with the phrase including the input keyword.
[0143]
(Appendix 3)
The refinement information creation unit has a function of creating a hierarchical structure by gathering common phrases presented as refinement information and presenting the hierarchical structure as refinement information to be presented to the user. The document search device according to (Appendix 1) or (Appendix 2), which is characterized.
[0144]
(Appendix 4)
The narrowed-down information creation unit has a function of creating narrowed-down information based on other dependency relationships designated by the user from a search result in the selected narrowed-down information and presenting it as new narrowed-down information. The document search device according to any one of (Appendix 1) to (Appendix 3).
[0145]
(Appendix 5)
In accordance with one or more search viewpoints prepared in advance, a subject extraction unit that extracts a sentence or a part of the sentence that characterizes the search target document that matches the search viewpoint together with the search viewpoint, and the user performs the search when searching A user interface for inputting a viewpoint, and
Any one of (Appendix 1) to (Appendix 3) is provided with a refinement information creation unit that creates refinement information to be created from a subject description that matches an input search viewpoint. Document retrieval device.
[0146]
(Appendix 6)
On the computer,
A refinement information creation unit that creates and presents refinement information based on the clause including the input keyword and the clause that is the source or destination,
A program for realizing a function of a dependency information search engine for searching a document based on information selected by a user from presented narrowing information, or a computer-readable recording medium on which the program is recorded.
[0147]
【The invention's effect】
As described above, the present invention has the following effects.
(1): In claim 1, the narrowed-down information creating unit creates narrowed-down information based on the phrase including the input keyword and the phrase that is the source or destination of the phrase and presents it to the user. The dependency information search engine searches for documents based on information selected by the user from the presented narrowing information.
[0148]
In this way, it is possible to present information for searching for a sentence that matches the user's search intention as information for the next narrowing operation, instead of simply searching for a sentence including the keyword entered by the user. become. The information presented at that time is not preset as in the conventional directory-type search method, but can be dynamically generated from the user's search viewpoint. In addition, since a word that is a synonym is also presented, a sentence described using the synonym of the input keyword can be searched without preparing a synonym dictionary in advance.
[0149]
(2): In claim 2, the narrowed-down information creating unit creates narrowed-down information based on the type of phrase presented together with the phrase including the input keyword.
[0150]
In this way, it is possible to present information for searching for a sentence that matches the user's search intention as information for the next narrowing operation, instead of simply searching for a sentence including the keyword entered by the user. become. The information presented at that time is not preset as in the conventional directory-type search method, but can be dynamically generated from the user's search viewpoint. In addition, since a word that is a synonym is also presented, a sentence described using the synonym of the input keyword can be searched without preparing a synonym dictionary in advance.
[0151]
(3): In claim 3, the narrowed-down information creating unit creates a hierarchical structure by collecting common phrases among the phrases presented as the narrowed-down information, and presents the hierarchical structure as the narrowed-down information presented to the user. To do.
[0152]
In this way, it is possible to present information for searching for a sentence that matches the user's search intention as information for the next narrowing operation, instead of simply searching for a sentence including the keyword entered by the user. become. The information presented at that time is not preset as in the conventional directory-type search method, but can be dynamically generated from the user's search viewpoint. In addition, since a word that is a synonym is also presented, a sentence described using the synonym of the input keyword can be searched without preparing a synonym dictionary in advance.
[0153]
Furthermore, since the information for the narrowing operation to be presented is grouped into categories, it is possible to perform an efficient operation in the next narrowing.
[0154]
(4): In claim 4, the narrowed-down information creating unit creates narrowed-down information from the search result in the selected narrowed-down information based on other dependency relationships designated by the user, and uses this as new narrowed-down information. Present.
[0155]
In this way, it is possible to present information for searching for a sentence that matches the user's search intention as information for the next narrowing operation, instead of simply searching for a sentence including the keyword entered by the user. become. The information presented at that time is not preset as in the conventional directory-type search method, but can be dynamically generated from the user's search viewpoint.
[0156]
In addition, since a word that is a synonym is also presented, a sentence described using the synonym of the input keyword can be searched without preparing a synonym dictionary in advance. Furthermore, since the process is repeated until the user obtains a sufficient narrowing result, a search with sufficient narrowing can be performed.
[0157]
(5): In claim 5, the subject extraction unit searches a sentence or a part of a sentence that characterizes the content of a search target document that matches the search viewpoint according to one or more search viewpoints prepared in advance. Extract with. The user interface also inputs the search viewpoint when the user searches. The refinement information creation unit creates the refinement information to be generated from information obtained from the subject description that matches the input search viewpoint.
[0158]
In this way, it is possible to present information for searching for a sentence that matches the user's search intention as information for the next narrowing operation, instead of simply searching for a sentence including the keyword entered by the user. become. The information presented at that time is not preset as in the conventional directory-type search method, but can be dynamically generated from the user's search viewpoint. In addition, since a word that is a synonym is also presented, a sentence described using the synonym of the input keyword can be searched without preparing a synonym dictionary in advance.
[Brief description of the drawings]
FIG. 1 is a diagram illustrating the principle of the present invention.
FIG. 2 is a configuration diagram of a document search apparatus according to an embodiment of the present invention.
FIG. 3 is a process explanatory diagram of the document search apparatus according to the embodiment of the present invention.
FIG. 4 is a processing flowchart 1 at the time of search in the embodiment of the present invention.
FIG. 5 is a processing flowchart 2 at the time of search in the embodiment of the present invention.
FIG. 6 is a diagram showing a presentation result 1 in the embodiment of the present invention.
FIG. 7 is a diagram showing a presentation result 2 in the embodiment of the present invention.
FIG. 8 is a diagram showing a presentation result 3 in the embodiment of the present invention.
FIG. 9 is a diagram showing a presentation result 4 in the embodiment of the present invention.
FIG. 10 is a diagram showing a presentation result 5 in the embodiment of the present invention.
FIG. 11 is a specific example of an apparatus according to an embodiment of the present invention.
[Explanation of symbols]
1 user
2 Document search device
3 User interface
4 Keyword search engine
5 Refinement information creation department
6 dependency information search engine
7 Theme extraction part
8 Refinement information creation rule storage database (Refinement information creation rule storage database)
9 Search target document database (Search target document DB)
10 Theme extraction rule storage database (Theme extraction rule storage DB)
21 Computer body
22 Display device
23 Input devices (keyboard / mouse, etc.)
24 Removable disk drive (RDD)
25 Hard Disk Drive (HDD)
26 CPU (Central Processing Unit)
27 ROM (Read Only Memory)
28 memory
29 Interface control unit (I / F control unit)
30 Communication control unit

Claims (5)

In a document search apparatus provided with a user interface and a keyword search engine,
A refinement information creation unit that creates refinement information and presents the clause including the input keyword and the clause that is obtained by dependency analysis and is the source or destination of the keyword ;
A document search apparatus comprising a dependency information search engine for searching for a document based on information selected by a user from presented refined information.   2. The document search apparatus according to claim 1, wherein the narrowed-down information creating unit has a function of creating narrowed-down information based on the type of phrase presented together with the phrase including the input keyword.   The refinement information creation unit has a function of creating a hierarchical structure by collecting common phrases presented as refinement information and presenting the hierarchical structure as refinement information to be presented to the user. The document search apparatus according to claim 1, wherein the document search apparatus is a document search apparatus.   The narrowed-down information creating unit has a function of creating narrowed-down information based on other dependency relationships designated by the user from a search result in the selected narrowed-down information and presenting it as new narrowed-down information. The document search apparatus according to claim 1, wherein In accordance with one or more search viewpoints prepared in advance, a subject extraction unit that extracts a sentence or a part of the sentence that characterizes the search target document that matches the search viewpoint together with the search viewpoint, and the user performs the search when searching A user interface for inputting a viewpoint, and
The document according to any one of claims 1 to 3, further comprising: a refinement information creation unit that creates refinement information to be created from information obtained from a subject description that matches an input search viewpoint. Search device.

Images