JP5564898B2 - Information search program and information search apparatus - Google Patents

Description

  The present invention relates to an information search program and an information search apparatus.

  There has been proposed a technique for retrieving a document according to a question sentence input in a natural language.

  As a technology related to this, in Patent Document 1, when a document search is requested by a question sentence composed of a natural language, an independent word and an attached word are extracted from the question sentence by language analysis, and the attached word Is converted into one of a plurality of preset operators and combined with a corresponding independent word to generate a search condition.

JP-A-8-339383

  An object of the present invention is to provide an information search program and an information search apparatus that make an appropriate answer when a sentence included in a document that is an appropriate answer to the question sentence does not contain an independent word extracted from the question sentence. Is to provide.

[1]
An independent word extracting means for extracting the first independent word from the accepted question sentence;
A sentence search means for searching for a sentence contained in a document stored in a storage means, including the first independent word extracted from the question sentence by the independent word extraction means;
Identifying means for extracting a second independent word contained in the sentence searched by the sentence searching means and identifying a third independent word used characteristically from the second independent word;
From a first content words the independent word extracting means has extracted, question type specifying means for specifying a content words independent words that match the self Tatsugo that is included in the question type information indicating the type of question the question When,
If the type of question the question is predetermined type, the corresponding document by searching the document including a Related Terms that are related to the independent words indicating the type of the question and the third independent words, An information retrieval program for extracting as an answer to the question sentence and functioning as an answer extracting means for extracting the sentence retrieved by the sentence retrieving means as an answer when the question type of the question sentence is not a predetermined type.

[2] Of the second independent words included in the sentence, the specifying unit is an independent word that is frequently used in the sentence and is less frequently used in the sentence included in the entire stored document. The information search program according to [1], in which is specified as the third independent word.

[3] the answer extraction means, when said identifying means is unable to identify the third content words from said second independent word, claim for extracting the sentences the sentence search means searches a reply [1] Or the information search program as described in [ 2 ].

[4] Independent word extracting means for extracting the first independent word from the accepted question sentence;
A sentence search means for searching for a sentence contained in a document stored in a storage means, including the first independent word extracted from the question sentence by the independent word extraction means;
Identifying means for extracting a second independent word contained in the sentence searched by the sentence searching means and identifying a third independent word used characteristically from the second independent word;
From a first content words the independent word extracting means has extracted, question type specifying means for specifying a content words independent words that match the self Tatsugo that is included in the question type information indicating the type of question the question When,
If the type of question the question is predetermined type, the corresponding document by searching the document including a Related Terms that are related to the independent words indicating the type of the question and the third independent words, An information search apparatus comprising: an answer extraction unit that extracts the sentence searched by the sentence search unit as an answer when the question is extracted as an answer to the question sentence and a question type of the question sentence is not a predetermined type.

According to the inventions according to claims 1 and 4, in the case where an independent word extracted from the question sentence is not included in the sentence included in the document that is an appropriate answer to the question sentence, the question type is determined in advance. In some cases, an independent word indicating the type of question and related words of the independent word can be reflected in the answer, and an appropriate answer can be made, and if it is not a predetermined question type, it is extracted from the question text Sentences that contain independent words can be extracted as answers .

  According to the invention which concerns on Claim 2, an answer can be extracted using the independent word specified based on appearance frequency.

According to the invention which concerns on Claim 3, when there is no feature word in a related sentence , the sentence containing the independent word extracted from a question sentence can be extracted as an answer .

It is the schematic which shows the structural example of the information search system of this invention. It is the schematic which shows the structural example of an information search device. It is the schematic which shows the structural example of question classification information. It is the schematic which shows the structural example of product specific information. (A) And (b) is the schematic which shows the structural example of a question sentence. (A)-(c) is the schematic which shows the structural example of the related text relevant to a question sentence. (A)-(c) is the schematic which shows the structural example of the reply extracted from the feature word extracted from a related sentence, and a feature word. It is a flowchart which shows the operation example of an information search system.

(Configuration of information retrieval system)
FIG. 1 is a schematic diagram showing a configuration example of an information search system of the present invention.

  The information search system 5 is configured by connecting the information search device 1, the document management server device 2, and the terminal device 3 so as to communicate with each other via a network 4.

  The information search apparatus 1 is an information processing apparatus having an electronic component such as a CPU (Central Processing Unit) and a storage unit having a function for searching for information according to a question sentence.

  The document management server device 2 is connected to a document database 2A for storing documents, and includes a CPU having a function for performing management operations such as saving, deleting, writing, moving, copying, and the like of documents stored in the document database 2A. An information processing apparatus having electronic components such as a storage unit.

  The terminal device 3 is a device for accessing the information search device 1 to input a question sentence and displaying a search result for the question sentence, and includes an operation unit for operation input and a display unit such as a liquid crystal display And a control unit provided with electronic components such as a CPU and a storage unit. The terminal device 3 may be, for example, a personal computer, a PDA (Personal Digital Assistant), a mobile phone, or the like. Further, although one terminal device 3 is illustrated in the figure, a plurality of terminal devices 3 may be provided.

  The network 4 may be a LAN (Local Area Network), the Internet, etc., and may be wired or wireless.

  Here, the “question sentence” is a sentence composed of a natural language, and the length and the number of sentences do not matter. The “search result” is a word or sentence composed of a natural language, a document itself containing a sentence, a link to the document, or the like. In the present embodiment, for example, a case will be described in which a specific product or product name is searched as a search result using a set of independent words indicating the nature of the product or product as a question sentence.

  FIG. 2 is a schematic diagram illustrating a configuration example of the information search apparatus 1.

  The information retrieval apparatus 1 includes a CPU and the like, and controls each unit and executes various programs. A storage unit 11 includes a storage medium such as an HDD (Hard Disk Drive) or a flash memory and stores information. And a communication unit 12 that communicates with the outside via the network 4.

  The control unit 10 executes an information search program 11A shown in FIG. 2 to be described later, whereby an independent word extraction unit 10A, a question type specification unit 10B, a related sentence search unit 10C, a feature word specification unit 10D, an answer extraction unit 10E, It functions as the output means 10F and the like.

  The independent word extracting unit 10A accepts the question sentence input in the terminal device 3 and extracts an independent word.

  The question type specifying means 10B specifies the type of question (hereinafter also referred to as “question type”) from the independent words included in the question sentence. Here, “question type” is synonymous with the type of answer requested by the question text. For example, when a product name of a video camera, the name of a politician, the name of a capital, etc. are requested as an answer. In the question sentence, independent words such as “What is“ video camera ”?”, “What is“ politician ”?” And “What is“ capital ”?” Are used as question types. Specifically, it is specified from the contents described in advance in question type information 11B described later.

  The related sentence search unit 10C searches for sentences having high relevance according to the contents of the question sentence from the sentences constituting the document stored in the document database 2A.

  The feature word specifying unit 10D specifies a feature word used characteristically from the independent words constituting the sentence from the sentences searched by the related sentence search unit 10C, and the storage unit 11 as feature word information 11D described later. To remember.

  The answer extracting unit 10E extracts the answer associated with the feature word specified by the feature word specifying unit 10D and stores it in the storage unit 11 as answer information 11E described later.

  The output means 10F outputs the answer information 11E in information in a format that can be displayed by the terminal device 3.

  The storage unit 11 includes an information search program 11A that causes the control unit 10 to operate as the above-described units, question type information 11B that is information associating a specific question type with its category, a specific question type, and a specific product. Product specific information 11C for associating a name, feature word information 11D specified by the feature word specifying means 10D, and answer information 11E extracted by the answer extracting means 10E are stored.

  FIG. 3 is a schematic diagram illustrating a configuration example of the question type information 11B.

  The question type information 11B includes a question type column 110B indicating the question type included in the question sentence, and an affiliation column 111B indicating the affiliation category corresponding to each question type.

  FIG. 4 is a schematic diagram illustrating a configuration example of the product specific information 11C.

  The product specific information 11C includes a question type column 110C indicating a question type included in the question sentence, and a product name column 111C indicating a specific product name corresponding to each question type.

(Operation of information retrieval system)
Hereinafter, the operation of the information search system 5 will be described with reference to the drawings.

  FIG. 8 is a flowchart showing an operation example of the information search system 5.

  First, when a user operates the operation unit of the terminal device 3 to make a search request for a product or product to the information search device 1, the control unit 10 of the information search device 1 is based on the information search program 11A. A display screen for accepting questions is displayed on the display unit of the terminal device 3.

  Next, on the display screen for accepting a question displayed on the terminal device 3, when the user operates the operation unit and inputs a question sentence, the independent word extracting means 10 </ b> A is connected via the network 4 and the communication unit 12. The question text is received and accepted (S1).

  FIGS. 5A and 5B are schematic diagrams illustrating an example of the structure of a question sentence.

  The question sentence 100 shown in FIG. 5A is composed of, for example, a sentence “What is a video camera that can be taken beautifully in a dark place?”

  Next, the independent word extracting unit 10A extracts the independent words 100a to 100d shown in FIG. 5B from the question 100 shown in FIG. 5A (S2). The question type identification unit 10B checks whether there are any words in the question type column 110B of the question type information 11B shown in FIG. 3 among the independent words 100a to 100d. In the present embodiment, since “video camera” of the independent word 100d exists in the question type information column 110B, the question type specifying means 10B specifies “video camera” of the independent word 100d as the question type (S3). .

Further, the question type specifying unit 10B determines which category the “video camera” belongs to from the belonging column 111B of the question type information 11B (S4), and when the belonging corresponds to “product / product” (S 4 Yes), the related sentence search means 10C searches for a related sentence (hereinafter referred to as “related sentence”) from the documents stored in the document database 2A based on the independent words 100a to 100d (S5). At this time, the related text search unit 10C searches for related text that includes as many extracted independent words 100a to 100d as possible, but the search condition is not limited to this.

  FIGS. 6A to 6C are schematic diagrams illustrating configuration examples of related sentences related to the question sentence.

  The related sentences 102a to 102c shown in FIGS. 6A to 6C are sentences searched using the independent words 100a to 100d shown in FIG. All the related sentences 102a to 102c include independent words 100a to 100c of the question sentence 100 such as “dark”, “beautiful”, and “photographing”.

  Next, the feature word specifying unit 10D extracts an independent word from the retrieved related sentences 102a to 102c (S6), and specifies the feature word from the independent word (S7). The feature word is specified using, for example, the TF / IDF method described below.

は、頻出する自立語について値が大きくなる。また、自立語ｔ_ｉを含む文章の数をＭ（ｔ_ｉ）とし、全文章の数をＭとすると、

は、めずらしい自立語について値が大きくなる。以上より、

は、特徴的に頻出する自立語について値が大きくなるため、ＴＦＩＤＦ（ｔ_ｉ）が大きくなる単語ｔ_ｉを特徴語とする。一例として、ＴＦＩＤＦ（ｔ_ｉ）が予め定められた閾値を超える自立語を特徴語として特定する。 First, let F (t _i ) be the frequency of occurrence of an independent word t _i in any related sentence.

Increases for frequent independent words. Further, if the number of sentences including the independent word t _i is M (t _i ) and the number of all sentences is M,

Increases the value for rare independent words. From the above,

Since the value of an independent word that frequently appears is large, the word t _i that increases TFIDF (t _i ) is used as a feature word. As an example, an independent word whose TFIDF (t _i ) exceeds a predetermined threshold is specified as a feature word.

  FIGS. 7A to 7C are schematic diagrams illustrating configuration examples of feature words extracted from related sentences and answers extracted from the feature words.

  A feature word 103a shown in FIG. 7A is a feature word specified from the related sentence 102a shown in FIG. Moreover, the feature word 103b shown in FIG.7 (b) is a feature word specified from the related sentence 102b shown in FIG.6 (b), and the feature word 103c shown in FIG.7 (c) is shown in FIG.6 (c). It is the feature word specified from the related sentence 102c shown in FIG.

  When a feature word is specified as shown in FIGS. 7A to 7C (S7; Yes), the answer extraction unit 10E determines the product / product name included in the document including the feature words 103a to 103c at the same time. Among them, what is included in the product name column 111C of the product specific information 11C shown in FIG. 4 is extracted as an answer (S8).

  For example, as shown in FIG. 7A, an answer 104a “hand cam AB15, hand cam AB25, hand cam AB35” is extracted as an answer based on the feature word 103a “Seymour Grace Sensor”. The answers 104b and 104c are extracted in the same manner. The extracted feature words 103a to 103c and answers 104a to 104c are stored in the storage unit 11 as feature word information 11D and answer information 11E, respectively, by the answer extracting means 10E.

  Next, the output means 10F reads the response information 11E from the storage unit 11, transmits it to the terminal device 3, and displays the response on the display screen of the terminal device 3 (S9). As the display form, the answers 104a to 104c may be displayed as character strings, all the contents of the document including the answers 104a to 104c may be displayed, or a link to the document may be displayed.

  Further, when the question type is other than “product / product” in step S4 (S4; No), and when no feature word is extracted in step S7 (S7; No), the conventional general question The question is processed by the response system (S10). A general question answering system is, for example, a system that extracts independent words from a question sentence, searches for related sentences based on the extracted independent words, and displays the related sentences. Further, a person name equivalent to the product name, a place name, or the like may be returned from the related text.

[Other embodiments]
The present invention is not limited to the above embodiment, and various modifications can be made without departing from the gist of the present invention. For example, you may apply this invention with respect to things other than a product and goods. For example, if the slogan “Nippon to change, now is the time” is specified as a feature word for a question sentence “What is the most popular politician now?”, A search is performed using this feature word. Then, “Hanako Fuji” may be extracted as an answer.

  The independent word extraction means 10A, question type identification means 10B, related sentence search means 10C, feature word identification means 10D, answer extraction means 10E, and output means 10F used in the above embodiment are stored in a CD-ROM or the like. It may be read from a medium into a storage unit in the device, or may be downloaded from a server device connected to a network such as the Internet to a storage unit in the device. Moreover, you may implement | achieve part or all of the means used by the said embodiment by hardware, such as ASIC.

DESCRIPTION OF SYMBOLS 1 ... Information retrieval apparatus, 2 ... Document management server apparatus, 2A ... Document database, 3 ... Terminal device, 4 ... Network, 5 ... Information retrieval system, 10 ... Control part, 10A ... Independent word extraction means, 10B ... Question type specification Means, 10C: Related text search means, 10D: Feature word specifying means, 10E ... Answer extraction means, 10F ... Output means, 11 ... Storage section, 11A ... Information search program, 11B ... Question type information, 11C ... Product specific information, 11D ... feature word information, 11E ... answer information, 12 ... communication unit, 100 ... question sentence, 100 ... question, 100a-100d ... independent words, 102a-102c ... related sentences, 103a-103c ... feature words, 104a-104c ... Answer, 110B ... Question type column, 111B ... Affiliation column, 110C ... Question type column, 111C ... Product name column

Claims (4)

Computer
An independent word extracting means for extracting the first independent word from the accepted question sentence;
A sentence search means for searching for a sentence contained in a document stored in a storage means, including the first independent word extracted from the question sentence by the independent word extraction means;
Identifying means for extracting a second independent word contained in the sentence searched by the sentence searching means and identifying a third independent word used characteristically from the second independent word;
From the first independent word extracted by the independent word extracting means, a question type specifying means for specifying an independent word that matches the independent word included in the question type information as an independent word indicating the question type of the question sentence;
When the question type of the question sentence is a predetermined type, a document including the third independent word and a related word related to the independent word indicating the question type is searched, and the corresponding document is An information search program for extracting as a reply to a question sentence and functioning as an answer extraction means for extracting the sentence searched by the sentence search means as an answer when the question type of the question sentence is not a predetermined type.   The specifying means selects the independent words that are frequently used in the sentence among the second independent words included in the sentence and are low in the sentence included in the entire stored document. The information search program according to claim 1, which is specified as three independent words. The answer extraction means, when said identifying means is unable to identify the third content words from said second independent words, according to claim 1 or 2 for extracting the sentences the sentence search means searches a reply Information retrieval program. An independent word extracting means for extracting the first independent word from the accepted question sentence;
A sentence search means for searching for a sentence contained in a document stored in a storage means, including the first independent word extracted from the question sentence by the independent word extraction means;
Identifying means for extracting a second independent word contained in the sentence searched by the sentence searching means and identifying a third independent word used characteristically from the second independent word;
From the first independent word extracted by the independent word extracting means, a question type specifying means for specifying an independent word that matches the independent word included in the question type information as an independent word indicating the question type of the question sentence;
When the question type of the question sentence is a predetermined type, a document including the third independent word and a related word related to the independent word indicating the question type is searched, and the corresponding document is An information search apparatus comprising: an answer extraction unit that extracts the sentence searched by the sentence search unit as an answer when the question is extracted as an answer to the question sentence and the question type of the question sentence is not a predetermined type.

Images