JP2002366577A - System, method, program for information retrieval, recording medium with information retrieval program recorded thereon, device, method, program for selecting output information and recording medium with output information selection program recorded thereon - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide technique by which browse of information to be required by a user is performed from a similarity retrieval result without waste to the utmost. SOLUTION: The information desired by the user is predicted and precedently presented by sequentially repeating procedures to extract a group of pieces of information with similarity of feature vectors higher than a threshold to a query, to present the contents of information having the feature vector with the highest similarity among them, when the user determines that the information as YES, to present information having high similarity with the feature vector of the information next, on the contrary, when the user determines that the information as NO, to use a virtual vector in the opposite direction to the feature vector afterward.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to an information retrieval system, an information retrieval method, an information retrieval program, a recording medium on which the information retrieval program is recorded, an output information selection device, an output information selection method, an output information selection program, and an output information selection. The present invention relates to a recording medium on which a program is recorded.

[0002]

2. Description of the Related Art When a user searches for a document on the Internet or accesses a document database installed in a specific place to search for a document, he or she inputs text data describing the contents of the document required as a creator. If you input text data of an existing document that is a model of the document you need, and give a search command, the search engine has text data similar to the text data of the document entered as a creator There is known an information search system that extracts a document and presents it to a user.

In such an information retrieval system, for example, a vector space model represented by tf * idf is used. This is to analyze text data of many documents, find feature vectors and register them in the database,
Finding feature vectors of text data entered by the user, extracting similar feature vectors from a group of feature vectors already registered in the database, and presenting a document having the extracted feature vectors to the user as a search result It is.

[0004] The above tf * idf is a feature amount obtained by adding a weight in consideration of the appearance frequency of the text data of another search target document to the appearance frequency of each word in the text data of the document. The weight w _t for a word is represented by the following Equation 1.

[0005]

(Equation 1) Here, N target document number, f _t is the number of documents containing the word t.

Each element w _{d, t} of the feature vector of the text data of the document is calculated as follows.

[0007]

(Equation 2) Here, f _{d, t} is the word t in the text data of document d.
Is the frequency of appearance.

[0008] As a measure for obtaining the similarity between documents in which the feature amount is represented by a vector, for example, there is a cosine coefficient. With this cosine coefficient, the similarity sim (x, y) between two vectors x and y
Is represented by the following equation (3).

[0009]

(Equation 3) Then, in the conventional information retrieval system, a document having a feature vector whose similarity with a feature vector of text data as a user query is higher than a predetermined threshold value is extracted, and as shown in FIG. Search results were presented in a format that sorts them in descending order.

FIG. 13 schematically shows a search result in order of similarity shown in FIG. 12 in a vector space. That is, the spatial positional relationship between the feature vector of the query Q and each of the feature vectors of the documents A to E having a high similarity within the threshold value is represented as shown in FIG. In FIG. 13, the order in which the similarity is high within the threshold value for the query Q is that the distance scale for the query Q is A <
Since D <B <C <E, they are A, D, B, C, and E.

As a model represented by a document feature vector, besides tf * idf, for example, “Automatic Te
xt Processing The Transformation, Analysis, and Re
trieval of Information by Computer Gerard Salton ''
As shown in the Term-discrimination Value and Proba
There are various models such as bilistic Term Weighting.

[0012] In addition to the above Cosine coefficient, as a measure for obtaining the similarity, as shown in the above-cited documents, Inner product, Dice coeffi
There are various distance measures such as cient and Jaccard coefficient.

[0013]

However, such a conventional information retrieval system has the following problems. FIG.
As shown in (1), even for documents A to E having a high similarity within the threshold value, the directions of their feature vectors with respect to the query Q are different. However, since the search results are usually displayed only in the order of the highest similarity as shown in FIG. 12, the user expands and reads the document A closest to the query Q and finds that he or she is seeking. Even if it is found that the content is the content, next, the document D having the second highest similarity, which is the opposite direction to the query Q from the document A, is opened. Next, the document B having the third similarity with the document A, which is almost opposite in direction to the document D with respect to the query Q but has a similar feature vector to the document A, is opened. Then, the document C whose direction is completely different from these documents is opened.

However, in reality, for the user, for example, when a document A having the highest similarity is opened, if the content is highly relevant to the document desired by the user, the next step is to open the feature vector. A document having a feature vector which has the second highest similarity but has a feature vector directed in the same direction as document A rather than opening document D whose feature vector is opposite to document A for query Q. It is desirable to open B.

[0015] However, in the past, it was merely developed in the order of similarity without considering the direction of the feature vector.
Such a user's request cannot be met, and there is a problem that it takes time and effort for the user to find a document having necessary contents.

The present invention has been made in view of such conventional problems, and has as its object to provide a technique which allows a user to browse required information from a similarity search result as efficiently as possible. I do.

[0017]

According to the present invention, there is provided an information retrieval system comprising: a query input unit for receiving a query input by a user; and a feature vector generating unit for generating a feature vector from the query received by the query input unit. And an information database in which indexes of a large number of search target information, data representing contents and feature vector data are registered, a feature vector of a query created by the feature vector creation unit, and a search registered in the information database. Calculate the similarity with the feature vector for each target information, specify the search target information having the feature vector indicating the similarity higher than the predetermined threshold value, and extract the index, the data representing the content and the feature vector data. A similarity calculation unit to be extracted, and a search result for storing data extracted by the similarity calculation unit. A holding unit, a search result output unit that outputs an index and data indicating the contents of one of the search target information extracted by the similarity calculation unit, and a user input to the search result output by the search result output unit. (1) In the case of a suitable input, the search result holding unit has a feature vector having the highest similarity to the feature vector of the search target information currently being output. Is searched, and if there is corresponding search target information, the content is output as the next candidate by the search result output unit. (2) In the case of inappropriate input, if it is the first stage, it is added to the feature vector of the query. On the other hand, 2
If it is at or after the stage, a virtual vector that has the opposite direction to the feature vector of the search target information currently being output is calculated with respect to the feature vector output at the previous stage, and the feature vector with the highest similarity to this virtual vector is calculated. A feedback process for searching the search result information held in the search result holding unit for other search object information having the search result information and outputting the content of the corresponding search object information as a next candidate by the search result output unit. And a part.

According to a second aspect of the present invention, there is provided an information retrieval system, comprising: a query input unit for receiving a query input by a user; a feature vector generating unit for generating a feature vector from the query received by the query input unit; An information database in which data representing the index of information, contents, and feature vector data are registered; a feature vector of a query created by the feature vector creation unit; and a feature vector for each search target information registered in the information database. A similarity calculating unit for calculating the similarity with the search target information having the feature vector indicating the similarity higher than the predetermined threshold, extracting the index, the data representing the content and the feature vector data. A search result storage unit for storing data extracted by the similarity calculation unit, For the feature vector group with a high similarity calculated by the similarity calculation unit, start with the feature vector with the highest similarity, and if the content is acceptable to the user, inappropriate if the content is not acceptable Branching, when appropriately branching, other search target information having a feature vector having the highest similarity to the feature vector as the next candidate search target information, and when branching inappropriately from the feature vector, If it is the first stage, the feature vector of the query is calculated. If it is the second stage or later, the virtual vector that is in the opposite direction to the feature vector of the inappropriate search target information is calculated for the feature vector that was the output candidate of the previous stage. Then, another search target information having a feature vector having the highest similarity to the virtual vector is searched, and if there is the corresponding search target information, its content is set as the next candidate search target information. The that process, the search result and the decision tree creation unit repeatedly to create a decision tree for all search target information held in the holding portion to be exhaustive, 1 of the extracted search target information of the similarity calculation unit
A search result output unit for outputting data indicating the index and the contents of the search tree, and accepting a user's suitability / inappropriate judgment input for the search result output from the search result output unit, and the decision tree creation unit creates A feedback processing unit that specifies the next search target information to be output based on the decision tree, and causes the search result output unit to output the contents of the corresponding search target information.

According to a third aspect of the present invention, there is provided an information retrieval system comprising: a query input unit for receiving a query input by a user; a feature vector generating unit for generating a feature vector from the query received by the query input unit; An information database in which data representing the index of information, contents, and feature vector data are registered; a feature vector of a query created by the feature vector creation unit; and a feature vector for each search target information registered in the information database. A similarity calculating unit for calculating the similarity with the search target information having the feature vector indicating the similarity higher than the predetermined threshold, extracting the index, the data representing the content and the feature vector data. A search result storage unit for storing data extracted by the similarity calculation unit, For the feature vector group with a high similarity calculated by the similarity calculation unit, start with the feature vector with the highest similarity, and if the content is acceptable to the user, inappropriate if the content is not acceptable Branching, when appropriately branching, other search target information having a feature vector having the highest similarity to the feature vector as the next candidate search target information, and when branching inappropriately from the feature vector, If it is the first stage, the feature vector of the query is calculated. If it is the second stage or later, the virtual vector that is in the opposite direction to the feature vector of the inappropriate search target information is calculated for the feature vector that was the output candidate of the previous stage. Then, another search target information having a feature vector having the highest similarity to the virtual vector is searched, and if there is the corresponding search target information, its content is set as the next candidate search target information. The that process, the search for results held search target information group held in the unit and the decision tree creation unit for creating a decision tree is repeated up to a predetermined stage, 1 of the extracted search target information of the similarity calculation unit
A search result output unit for outputting data indicating the index and the contents of the search tree, and accepting a user's suitability / inappropriate judgment input for the search result output from the search result output unit, and the decision tree creation unit creates The next search target information to be output is specified based on the decision tree, and the content of the relevant search target information is output by the search result output unit,
And a feedback processing unit for instructing the decision tree creating unit to grow the decision tree by a predetermined stage when a condition for growing an existing decision tree is reached. .

According to a fourth aspect of the present invention, in the information search system of the first to third aspects, the query and the search target information are text data.

According to a fifth aspect of the present invention, there is provided an information retrieval method, wherein a query input by a user is received, and a feature vector is created from the received query.
Calculating the similarity between the feature vector of the created query and the feature vector for each piece of search target information registered in the information database, and searching for a feature vector having a similarity higher than a predetermined threshold. Step 3 for specifying the target information and extracting the index, data representing the content and feature vector data, and outputting the index and data indicating the index and content of one of the search target information extracted in Step 3; In step 4, the user's input for determining whether or not the search result is output is accepted. (1) In the case of an appropriate input, the feature vector having the highest similarity to the feature vector of the search target information currently being output The search target information retrieved in step 3 is searched for other search target information having Then, the contents are output as the next candidate. (2) In the case of inappropriate input, the feature vector of the query is output in the first stage, and the feature vector output in the previous stage is output in the second and subsequent stages. A virtual vector that is in the opposite direction to the feature vector of the search target information that is currently being output is calculated. Step 5 of searching the target information and outputting the content of the relevant search target information as a next candidate if any.

According to a sixth aspect of the present invention, in the information search method, a step 1 for receiving a query input by a user and a step 2 for creating a feature vector from the received query.
Calculating the similarity between the feature vector of the created query and the feature vector for each piece of search target information registered in the information database, and searching for a feature vector having a similarity higher than a predetermined threshold. Step 3 of identifying the target information and extracting the index, data representing the contents and feature vector data, and starting with the feature vector having the highest similarity for the feature vector group having the highest similarity obtained in Step 3; If the content is accepted by the user, it branches appropriately if it is not accepted, and if it is not, then another branch that has the feature vector with the highest similarity to the feature vector If the information is the search target information of the next candidate and the branch is inappropriately performed from the feature vector, if it is the first stage, the query If it is the second or subsequent stage, a virtual vector having the opposite direction to the feature vector of the inappropriate search target information is calculated with respect to the feature vector as the output candidate of the previous stage, and the similarity to this virtual vector is the highest. The process of searching for other search target information having a feature vector and, if there is such search target information, setting the content thereof as the next candidate search target information is repeated until all the search target information extracted in step 3 is covered. Step 4 of creating a decision tree
And step 5 for outputting data indicating the index and contents of one of the search target information extracted in step 3, and accepting a user's suitability / inappropriate determination input for the search result output in step 5; Step 6 of specifying the search target information to be output next based on the decision tree created in Step 4, and outputting the contents of the relevant search target information.

According to a seventh aspect of the present invention, in the information search method, a step 1 for receiving a query input by a user and a step 2 for creating a feature vector from the received query.
Calculating the similarity between the feature vector of the created query and the feature vector for each piece of search target information registered in the information database, and searching for a feature vector having a similarity higher than a predetermined threshold. Step 3 of identifying the target information and extracting the index, data representing the contents and feature vector data, and starting with the feature vector having the highest similarity for the feature vector group having the highest similarity obtained in Step 3; If the content is accepted by the user, it branches appropriately if it is not accepted, and if it is not, then another branch that has the feature vector with the highest similarity to the feature vector If the information is the search target information of the next candidate and the branch is inappropriately performed from the feature vector, if it is the first stage, the query If it is the second or subsequent stage, a virtual vector having the opposite direction to the feature vector of the inappropriate search target information is calculated with respect to the feature vector as the output candidate of the previous stage, and the similarity to this virtual vector is the highest. A process of searching for other search target information having a feature vector and, if there is such search target information, setting the contents thereof as the next candidate search target information is repeatedly determined for the search target information extracted in step 3 up to a predetermined stage. Step 4 of creating a tree
And step 5 for outputting data indicating the index and contents of one of the search target information extracted in step 3, and accepting a user's suitability / inappropriate determination input for the search result output in step 5; Based on the decision tree created in step 4, the next search target information to be output is specified, the contents of the corresponding search target information are output, and if the condition for growing the decision tree created in step 4 is reached, Step 6 of growing the decision tree by a predetermined stage
And

According to an eighth aspect of the present invention, in the information search method of the fifth to seventh aspects, the query and the search target information are text data.

According to a ninth aspect of the present invention, there is provided an information retrieval program comprising:
Processing 1 for receiving a query input by the user, processing 2 for creating a feature vector from the received query,
Calculate the similarity between the feature vector of the created query and the feature vector for each search target information registered in the information database, and search target information having a feature vector indicating a similarity higher than a predetermined threshold value And a process 4 for extracting data indicating the index and the contents and the feature vector data, a process 4 for outputting the index and the content of one of the search target information extracted in the process 3, and a process 4 Accepts the user's suitability / inappropriate judgment input for the search result output in
(1) In the case of a suitable input, whether there is any other search target information having a feature vector having the highest similarity to the feature vector of the search target information currently being output, Is searched, and if there is corresponding search target information, the content is output as the next candidate. With respect to the feature vector output in the previous stage, a virtual vector that is in the opposite direction to the feature vector of the currently output search target information is calculated, and other search target information having a feature vector with the highest similarity to this virtual vector is calculated. The computer performs a process 5 of searching the search target information extracted in the process 3 for the presence of the search target information and outputting the content of the relevant search target information as a next candidate, if any.

An information retrieval program according to a tenth aspect of the present invention comprises a processing 1 for receiving a query input by a user;
Process 2 for creating a feature vector from the received query
Calculating the similarity between the feature vector of the created query and the feature vector for each piece of search target information registered in the information database, and searching for a feature vector having a similarity higher than a predetermined threshold. A process 3 for specifying the target information and extracting the data representing the index, the content and the feature vector data, and a feature vector group having a high similarity obtained in the process 3 starting from a feature vector having the highest similarity, If the content is accepted by the user, it branches appropriately if it is not accepted, and if it is not, then another branch that has the feature vector with the highest similarity to the feature vector If the information is set as the search target information of the next candidate, and the branch is inappropriately performed from the feature vector, if the branch is in the first stage, two stages are required for the feature vector of the query. The feature vector obtained by the preceding stage output candidate if later, to calculate the virtual vector of the opposite direction, wherein vectors having the search target information unsuitable,
A process of searching for other search target information having a feature vector having the highest similarity to the virtual vector, and setting the content of the corresponding search target information, if any, as the next candidate search target information,
A process 4 for repeatedly creating a decision tree until all the search target information extracted in the process 3 is covered, and an index of one of the search target information extracted in the process 3,
Processing 5 for outputting data indicating the contents, and accepting the user's appropriateness / inappropriate determination input for the search result output in processing 5 and searching target information to be output next based on the decision tree created in processing 4 And causing the computer to execute processing 6 for outputting the contents of the corresponding search target information.

[0027] An information retrieval program according to an eleventh aspect of the present invention includes a processing 1 for receiving a query input by a user;
Process 2 for creating a feature vector from the received query
Calculating the similarity between the feature vector of the created query and the feature vector for each piece of search target information registered in the information database, and searching for a feature vector having a similarity higher than a predetermined threshold. A process 3 for specifying the target information and extracting the data representing the index, the content and the feature vector data, and a feature vector group having a high similarity obtained in the process 3 starting from a feature vector having the highest similarity, If the content is accepted by the user, it branches appropriately if it is not accepted, and if it is not, then another branch that has the feature vector with the highest similarity to the feature vector If the information is set as the search target information of the next candidate, and the branch is inappropriately performed from the feature vector, if the branch is in the first stage, two stages are required for the feature vector of the query. The feature vector obtained by the preceding stage output candidate if later, to calculate the virtual vector of the opposite direction, wherein vectors having the search target information unsuitable,
A process of searching for other search target information having a feature vector having the highest similarity to the virtual vector, and setting the content of the corresponding search target information, if any, as the next candidate search target information,
A process 4 for repeatedly creating a decision tree for the search target information extracted in the process 3 until a predetermined stage, a process 5 for outputting an index and data indicating the content of one of the search target information extracted in the process 3, 5 accepts the user's input of suitability / inappropriateness for the search result output, specifies the search target information to be output next based on the decision tree created in process 4, and determines the contents of the relevant search target information. When the condition for growing the decision tree generated in the process 4 is reached, the process 6 for growing the decision tree by a predetermined stage is executed by the computer.

According to a twelfth aspect of the present invention, in the information search program of the ninth to eleventh aspects, the query and the search target information are text data.

A recording medium according to a thirteenth aspect of the present invention provides a process 1 for receiving a query input by a user, a process 2 for creating a feature vector from the received query, a feature vector of the created query, and registration in an information database. Calculate the similarity with the feature vector for each piece of search target information that has been searched, specify search target information having a feature vector indicating a similarity higher than a predetermined threshold, their indexes, data representing the contents and Processing 3 for extracting the feature vector data, and 1 of the search target information extracted in the processing 3
Processing 4 for outputting the data indicating the index and the contents of each of them, and accepting the user's appropriateness / inappropriate determination input with respect to the search result output in the processing 4 are received. Searches the search target information extracted in the process 3 for other search target information having the feature vector having the highest similarity to the feature vector of the search target information. The content is output as the next candidate. (2) In the case of inappropriate input, the feature vector of the query is output in the first stage, and the feature vector output in the previous stage is output in the second and subsequent stages. A virtual vector that is in the opposite direction to the feature vector of the search target information is calculated. And in searching for the search target information is a record the appropriate information retrieval program for executing the content, if the search target information and process 5 for outputting a next candidate in the computer.

According to a fourteenth aspect of the present invention, there is provided a recording medium for receiving a query input by a user, a process for creating a feature vector from the received query, a feature vector for the created query, and registering the feature vector in an information database. Calculate the similarity with the feature vector for each piece of search target information that has been searched, specify search target information having a feature vector indicating a similarity higher than a predetermined threshold, their indexes, data representing the contents and Process 3 for extracting the feature vector data, and for the feature vector group having a high similarity obtained in Process 3, starting from the feature vector having the highest similarity, if the content is acceptable to the user, it is appropriately accepted. If it does not exist, it branches inappropriately, and if it branches properly, it has the feature vector with the highest similarity to the feature vector. The other search target information is set as search target information of the next candidate, and if the branch is inappropriately performed from the feature vector, the query vector is the first stage and the output candidate of the previous stage is the second or subsequent stage. With respect to the obtained feature vector, a virtual vector that is in the opposite direction to the feature vector of the inappropriate search target information is calculated, and another search target information having a feature vector with the highest similarity to this virtual vector is searched, If there is applicable search target information, the process of setting the content as the next candidate search target information is repeated until all search target information extracted in process 3 is covered, and a process 4 for creating a decision tree; Processing 5 for outputting data indicating the index and contents of one of the retrieved information to be searched, and determination of suitability of the user for the search result output in processing 5 An information search program is recorded that causes the computer to execute the process 6 of receiving the force, specifying the search target information to be output next based on the decision tree created in the process 4, and outputting the contents of the relevant search target information. Things.

A recording medium according to a fifteenth aspect of the present invention provides a process 1 for receiving a query input by a user, a process 2 for creating a feature vector from the received query, a feature vector of the created query, and registration in an information database. Calculate the similarity with the feature vector for each piece of search target information that has been searched, specify search target information having a feature vector indicating a similarity higher than a predetermined threshold, their indexes, data representing the contents and Process 3 for extracting the feature vector data, and for the feature vector group having a high degree of similarity obtained in Process 3, starting from the feature vector with the highest similarity, if the contents are acceptable to the user, the process is appropriately accepted. If it does not exist, it branches inappropriately, and if it branches properly, it has the feature vector with the highest similarity to the feature vector. The other search target information is set as search target information of the next candidate, and if the branch is inappropriately performed from the feature vector, the query vector is the first stage and the output candidate of the previous stage is the second or subsequent stage. With respect to the obtained feature vector, a virtual vector that is in the opposite direction to the feature vector of the inappropriate search target information is calculated, and another search target information having a feature vector with the highest similarity to this virtual vector is searched, If there is applicable search target information, the process of setting the contents as search target information of the next candidate is repeated up to a predetermined stage for the search target information extracted in process 3 to create a decision tree, and the process extracted in process 3 Processing 5 for outputting data indicating the index and content of one of the search target information, and determining whether the user is appropriate or inappropriate for the search result output in processing 5 Acceptance,
If the search target information to be output next is specified based on the decision tree created in the process 4, the content of the corresponding search target information is output, and the condition for growing the decision tree created in the process 4 is reached Processing 6 for growing the decision tree by a predetermined stage
And an information search program for causing a computer to execute the above.

According to a sixteenth aspect of the present invention, in the recording medium according to the thirteenth to fifteenth aspects, the query and the search target information are text data.

[0033] The output information selection device according to the seventeenth aspect of the present invention provides
A feature data holding unit that holds feature vector data together with information expressing the content of each of the group of search target information;
A display information output unit that outputs information expressing the content of the search target information specified from the group of search target information,
A feedback receiving unit that receives a user's suitability / inappropriate determination input with respect to the output search target information;
(1) If the user's judgment input is appropriate, there is no other search target information having a feature vector having the highest similarity to the feature vector of the search target information currently output by the display information output unit. Or search the group of search target information,
If there is applicable search target information, the information representing its contents is output to the display information output unit. (2) If the user's judgment input is inappropriate, if it is the first stage, it is given in advance. With respect to the reference vector, if it is the second or subsequent stage, the display information output unit is oriented in the opposite direction to the feature vector of the search target information currently output, with respect to the feature vector of the search target information output in the previous stage. A virtual vector is calculated, and the group of search target information is searched for other search target information having a feature vector having the highest similarity to the virtual vector. An output information selection unit for outputting information to the display information output unit.

The output information selection device according to the eighteenth aspect of the present invention
A feature data holding unit that holds feature vector data together with information expressing the content of each of the group of search target information;
A display information output unit that outputs information expressing the content of the search target information specified from the group of search target information,
For a group of search target information held in the feature data holding unit, starting from search target information having a feature vector having the highest similarity with a reference vector given in advance, when the content is accepted by the user. If it is not suitable, it branches inappropriately if it is not accepted, and if it branches properly, other search target information having the feature vector with the highest similarity to the feature vector is searched for the next candidate. In the case of inappropriate branching from the feature vector, if it is the first stage, with respect to the reference vector,
If it is the second stage or later, the virtual vector that is in the opposite direction to the feature vector of the inappropriate search target information is calculated with respect to the feature vector of the search target information as the output candidate of the previous stage, and similar to this virtual vector. The other search target information having the highest feature vector is searched, and if there is the corresponding search target information, the content of the search is set as the next candidate search target information. A decision tree creating unit that repeatedly creates a decision tree until the search target information is covered, and accepts a user's suitability / inadequacy determination input for the search target information output by the display information output unit, and generates the decision tree. The search information to be output next is specified based on the decision tree created by the unit, and the information representing the contents of the relevant search target information is output by the display information output unit. It is obtained by a readback processor.

An output information selection device according to a nineteenth aspect of the present invention
A feature data holding unit that holds feature vector data together with information expressing the content of each of the group of search target information;
A display information output unit that outputs information expressing the content of the search target information specified from the group of search target information,
For a group of search target information held in the feature data holding unit, starting from search target information having a feature vector having the highest similarity with a reference vector given in advance, when the content is accepted by the user. If it is not suitable, it branches inappropriately if it is not accepted, and if it branches properly, other search target information having the feature vector with the highest similarity to the feature vector is searched for the next candidate. In the case of inappropriate branching from the feature vector, if it is the first stage, with respect to the reference vector,
If it is the second or subsequent stage, a virtual vector that is in the opposite direction to the feature vector of the inappropriate search target information is calculated with respect to the feature vector of the search target information as the output candidate of the previous stage, and similar to this virtual vector The other search target information having the highest feature vector is searched, and if there is such search target information, the content of the search information is used as the next candidate search target information. A decision tree creating unit that repeatedly creates a decision tree for a target information group up to a predetermined stage; and accepts a user's suitability / inadequacy determination input with respect to the search target information output by the display information output unit. Specifying the next search target information to be output based on the created decision tree, and causing the display information output unit to output information representing the contents of the corresponding search target information A feedback processing unit for giving an instruction to the decision tree creating unit to grow the decision tree by a predetermined stage when a condition for growing an existing decision tree is reached. is there.

According to a twentieth aspect of the present invention, there is provided an output information selecting method comprising:
Step 1 of holding feature vector data together with information representing the contents of each of the group of search target information, and Step 2 of outputting information representing the contents of the search target information specified from the group of search target information. Receiving the user's appropriateness / inappropriate determination input with respect to the currently output search target information, and (1) if the user's determination input is appropriate, the feature vector of the currently output search target information Searching the group of search target information for other search target information having a feature vector with the highest similarity to the search target information, and outputting information representing the contents of the corresponding search target information, if any, If the user's judgment input is inappropriate, if the input is the first stage, the reference vector given in advance is used. , A virtual vector that is in the opposite direction to the feature vector of the search target information currently being output is calculated, and if there is no other search target information having a feature vector with the highest similarity to the virtual vector, Step 3 of searching for the search target information, and outputting information representing the contents of the corresponding search target information, if any.

According to a twenty-first aspect of the present invention, there is provided an output information selecting method comprising:
A step 1 of holding feature vector data together with information representing the contents of each of the group of search target information, and a step 2 of outputting information representing the contents of the search target information specified from the group of search target information. , For the group of search target information, starting with search target information having a feature vector having the highest similarity to a given reference vector, if the content is acceptable to the user, it is not accepted. If it branches inappropriately, and if it branches appropriately, other search target information having a feature vector having the highest similarity to the feature vector is set as search target information of the next candidate, and inappropriately determined from the feature vector. In the case of branching, if it is the first stage, the reference vector is used. With respect to the vector, a virtual vector that is in the opposite direction to the feature vector of the inappropriate search target information is calculated, and other search target information having a feature vector with the highest similarity to this virtual vector is searched, and the corresponding virtual vector is searched. Step 3 of creating a decision tree by repeating the process of setting the content of the search target information as the next candidate search target information, if any, until the entire set of search target information is covered; Receiving the user's appropriate / inappropriate judgment input to the server, specifying the next search target information to be output based on the decision tree, and outputting information expressing the contents of the corresponding search target information. It is.

The output information selecting method according to the twenty-second aspect is
A step 1 of holding feature vector data together with information representing the contents of each of the group of search target information, and a step 2 of outputting information representing the contents of the search target information specified from the group of search target information. , For the group of search target information, starting with search target information having a feature vector having the highest similarity to a given reference vector, if the content is acceptable to the user, it is not accepted. If it branches inappropriately, and if it branches appropriately, other search target information having a feature vector having the highest similarity to the feature vector is set as search target information of the next candidate, and inappropriately determined from the feature vector. In the case of branching, if it is the first stage, the reference vector is used. For the vector, calculate a virtual vector that is in the opposite direction to the feature vector of the inappropriate search target information, search for other search target information that has the feature vector with the highest similarity to this virtual vector, and Step 3 of creating a decision tree by repeating the process of setting the content of the search target information, if any, as the next candidate search target information up to a predetermined stage for the group of search target information; Is received, the search target information to be output next is specified based on the decision tree, the information expressing the contents of the relevant search target information is output, and the existing decision tree is grown. And a step 4 of growing the decision tree by a predetermined stage when the conditions for the decision tree are reached.

According to a twenty-third aspect of the present invention, there is provided an output information selection program for processing 1 for retaining feature vector data together with information representing the contents of a group of search target information, and a process designated from the group of search target information. Processing 2 for outputting information expressing the contents of the search target information, and accepting the user's appropriateness / inappropriate determination input for the currently output search target information, and (1) the user's determination input is appropriate In this case, the group of search target information is searched for other search target information having a feature vector having the highest similarity to the feature vector of the search target information that is currently being output. If it outputs information expressing its contents,
(2) If the user's judgment input is inappropriate, the reference vector is given in advance if it is the first stage, and the feature vector of the search target information output in the previous stage if it is the second or subsequent stage. For this, a virtual vector that is in the opposite direction to the feature vector of the currently output search target information is calculated, and there is no other search target information having a feature vector having the highest similarity in the virtual vector. And a process 3 for searching for the search target information and outputting information representing the contents of the search target information, if any.

An output information selection program according to a twenty-fourth aspect of the present invention provides a process 1 for storing feature vector data together with information representing the contents of a group of search target information, and a process designated from the group of search target information. A process 2 for outputting information representing the contents of the search target information, and starting with the search target information having the feature vector having the highest similarity to the reference vector given in advance for the group of search target information, If the content is accepted by the user, it branches inappropriately if it is not accepted, and if it branches off appropriately, if it branches properly, other search target information having a feature vector with the highest similarity to the feature vector Is the next candidate search target information, and if branching inappropriately from the feature vector, if the first stage, the second and subsequent stages with respect to the reference vector If there is any, the feature vector of the search target information as the output candidate in the previous stage is calculated, and a virtual vector that is in the opposite direction to the feature vector of the inappropriate search target information is calculated, and the feature having the highest similarity to this virtual vector is calculated. The process of searching for other search target information having a vector and, if there is the corresponding search target information, setting the content thereof as the next candidate search target information is repeated until the entire set of search target information is covered. And a user input for judging the suitability / inappropriateness of the search target information currently being output is specified, the next search target information to be output is specified based on the decision tree, and the corresponding search target information And a process 4 for outputting information expressing the contents.

According to a twenty-fifth aspect of the present invention, there is provided an output information selection program for processing 1 for holding feature vector data together with information representing the contents of a group of search target information, and a process designated from the group of search target information. A process 2 for outputting information representing the contents of the search target information, and starting with the search target information having the feature vector having the highest similarity to the reference vector given in advance for the group of search target information, If the content is accepted by the user, it branches inappropriately if it is not accepted, and if it branches off appropriately, if it branches properly, other search target information having a feature vector with the highest similarity to the feature vector Is the search target information of the next candidate, and if the branch is inappropriate from the feature vector, if the first stage, the second and subsequent stages with respect to the reference vector If there is any, the feature vector of the search target information as the output candidate in the previous stage is calculated, and a virtual vector that is in the opposite direction to the feature vector of the inappropriate search target information is calculated, and the feature having the highest similarity to this virtual vector is calculated. A process of searching for other search target information having a vector and, if there is such search target information, making the content thereof as the next candidate search target information is repeated up to a predetermined stage for the group of search target information to create a decision tree. Processing 3 and accepting the user's appropriateness / inappropriate determination input for the currently output search target information, specifying the next search target information to be output based on the decision tree, and rewriting the contents of the relevant search target information. And outputting the information to be expressed, and when the condition for growing the existing decision tree is reached, causes the computer to execute a process 4 for growing the decision tree by a predetermined stage. It is intended.

According to a twenty-sixth aspect of the present invention, there is provided a recording medium, comprising: processing 1 for storing feature vector data together with information representing the contents of a group of search target information; Processing 2 for outputting information expressing the content of the information, and accepting a user's suitability / inappropriate determination input for the search target information currently being output,
(1) If the user's decision input is appropriate, the group of search targets is searched for other search target information having a feature vector with the highest similarity to the feature vector of the search target information currently being output. Information is searched, and if there is applicable search target information, information representing the contents thereof is output. (2) If the user's judgment input is inappropriate, if it is the first stage, a predetermined reference is given. For vectors
If it is the second or subsequent stage, a virtual vector that is in the opposite direction to the feature vector of the currently output search target information is calculated with respect to the feature vector of the search target information output in the previous stage, and the similarity to this virtual vector is calculated. And searching the group of search target information for other search target information having the highest feature vector, and outputting the information representing the content of the search target information, if any, to the computer. An output information selection program is recorded.

A recording medium according to a twenty-seventh aspect of the present invention provides processing 1 for storing feature vector data together with information representing the contents of a group of search target information, and a search target specified from the group of search target information. Processing 2 for outputting information representing the content of the information, and starting from the search target information having the feature vector having the highest similarity with the reference vector given in advance for the group of search target information,
If the content is accepted by the user, it branches appropriately if it is not accepted, and if it is not, then another branch that has the feature vector with the highest similarity to the feature vector If the information is the search target information of the next candidate, and the branch is inappropriately performed from the feature vector, the search target information is set to the reference vector in the first stage, and to the previous stage in the second and subsequent stages. Calculates a virtual vector that has the opposite direction to the feature vector of the inappropriate search target information for the feature vector of
A process of searching for other search target information having a feature vector having the highest similarity to the virtual vector, and setting the content of the corresponding search target information, if any, as the next candidate search target information,
Processing 3 for repeatedly creating a decision tree until the entire set of search target information is covered, and accepting a user's suitability / inappropriate determination input for the search target information currently being output;
An output information selection program for causing a computer to execute a process 4 of specifying search target information to be output next based on the decision tree and outputting information representing the contents of the relevant search target information is recorded. .

A recording medium according to a twenty-eighth aspect of the present invention provides a process 1 for storing feature vector data together with information representing the contents of a group of search target information, and a search target specified from the group of search target information. Processing 2 for outputting information representing the content of the information, and starting from the search target information having the feature vector having the highest similarity with the reference vector given in advance for the group of search target information,
If the content is accepted by the user, it branches appropriately if it is not accepted, and if it is not, then another branch that has the feature vector with the highest similarity to the feature vector If the information is the search target information of the next candidate, and the branch is inappropriately performed from the feature vector, the search target information is set to the reference vector in the first stage, and to the previous stage in the second and subsequent stages. Calculates a virtual vector that has the opposite direction to the feature vector of the inappropriate search target information for the feature vector of
A process of searching for other search target information having a feature vector having the highest similarity to the virtual vector, and setting the content of the corresponding search target information, if any, as the next candidate search target information,
Processing 3 for repeatedly creating a decision tree for the group of search target information up to a predetermined stage, and accepting a user's appropriateness / inappropriate determination input for the search target information currently being output, and outputting the next based on the decision tree A process 4 for identifying information to be searched and outputting information representing the contents of the corresponding information to be searched, and, when conditions for growing an existing decision tree have been reached, growing the decision tree by a predetermined stage. An output information selection program to be executed by a computer.

[0045]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Embodiments of the present invention will be described below in detail with reference to the drawings. The feature of the present invention is that a group of documents obtained by performing a similarity search based on a feature vector is classified in consideration of the direction of the feature vector possessed by the user, and the user can select one from the group of documents presented as a search result. Open a single document, and the user ’s satisfaction with the content (i.e.,
Whether or not the document has the content requested by the user)
By inputting S and NO, and automatically selecting and presenting the next document to be developed according to the user's satisfaction on the computer side, the direction of the feature vector is based on the degree of similarity. In consideration of the nature, it is desirable to be able to sequentially develop search results according to the information required by the user.

FIG. 1 shows a functional configuration of the information retrieval system of the present invention. This information retrieval system is configured by a computer system or a network system that connects a plurality of computers with cables. The system includes, as functional components, a query input unit 1 for inputting, as a query, text data expressing the content required by the user or text data of a document including the content required by the user as a query. A feature vector creating unit 2 for creating a feature vector from text data input from the input unit 1, an index of a large number of stored documents, text data, a document database 3 storing feature vector data for each document, The similarity between the feature vector of the query created by the feature vector creating unit 2 and the feature vector of each document registered in the document database 3 is calculated, and a feature vector indicating a similarity higher than a predetermined threshold is calculated. Classify documents that have documents and extract their indexes, text data and feature vectors Degree calculation unit 4, and a retrieval result storage 5 for storing the retrieved data in the similarity calculation unit 4.

The present system also includes a search result output unit 7 which displays the index and text data of a group of documents having a high similarity extracted by the similarity calculation unit 4 and prints out the data as necessary. The apparatus includes a feedback processing unit 8 that receives an operation and provides a corresponding process to the search result holding unit 5.

Next, an information retrieval method by the information retrieval system having the above configuration will be described with reference to FIG. In this information retrieval system, a user inputs text data expressing contents required by the query input unit 1,
Alternatively, when text data of a document including the contents required by the user is input as a query (step S1), the feature vector creation unit 2 creates a feature vector from the input text data of the creation (step S3). The creation of this feature vector is based on tf * idf using Equations 1 and 2 that have been widely used in the past. The method of creating the feature vector may be any of the other methods listed in the conventional example. The similarity to the feature vector of the query created by the creating unit 2 is calculated using the above-described equation (3),
A document having a feature vector indicating a similarity higher than a predetermined threshold is specified, their index, text data, and feature vector are extracted (step S5), and temporarily stored in the search result holding unit 5 as search result data. (Step S7). The calculation of the similarity may be performed by other methods listed in the conventional example.

With respect to the search result obtained in this way, the search result output unit 7 displays information representing the contents of the document having the feature vector having the highest similarity to the query (step S9). The information indicating the contents of the document includes the title of the document, the name of the author, the date of issue, and the outline. Also,
All text data of the document can be expanded and displayed by the user's specification.

Regarding the display of the content of this document, whether the user has requested the content using a UI device (for example, a keyboard or a pointing device such as a mouse) (YES) or not (NO) Is input, the feedback processing unit 8 accepts this input and performs one of the following YES / NO processes (step S11).

The user determines that the content of the document presented first is the content he / she has requested, and determines YES.
Is input, the feedback processing unit 8 searches the search result holding unit 5 for a document having a feature vector having the highest similarity to the feature vector of the currently output document. Is displayed by the search result output unit 7 (steps S13, S19, S21).

If the user determines that the content of the document presented first is not the content he / she has requested, and inputs NO in step S11, the feedback processing unit 8 sets the A virtual vector that is in the opposite direction to the feature vector of the currently output document is calculated (step S15). The document is extracted from the group and the document content is displayed on the search result output unit 7 (steps S17 to S2).
1).

As described above, when the user inputs a YES / NO determination using the UI device with respect to the content of the document displayed second in step S21, the process proceeds to step S11.
To S21 are repeated. The same applies to the third and subsequent documents. Then, when there is no document having the corresponding feature vector in the search result holding unit 5, the output of the search result is ended (NO in step S19).

The output form of the search result by the above information search method will be described with reference to FIGS. Similarity calculator 4
It is assumed that the document group obtained as a result of the similarity calculation according to is similar to that shown in FIG. That is, for a query Q, documents having feature vectors whose similarities exceed the threshold value are sorted in order of A, D, B, C, E
Assume that It is assumed that the feature vectors of each document are oriented in various directions as illustrated.

In the first search result display, as shown in FIG. 3, the contents of the document A having the highest similarity to the query Q are displayed. Then, if the user selects YES / N
"Relevant (suitable)"
The selection buttons 11 and 12 of “not relevant (unsuitable)” are displayed.

Then, the user checks the contents of document A,
If the user operates the “Related” button 11 because the content is what he or she has sought, subsequently, in FIG. Search among the resulting documents,
Extract document B. Then, as shown in FIG.
The contents of the extracted document B are displayed.

On the contrary, if the user confirms the contents of the document A and operates the “no relation” button 12, as described above, the query Q, which is the document based on the similarity, is A virtual vector A 'in the direction opposite to the feature vector of the document A is obtained by the following equation (4).

[0058]

(Equation 4) Here, α is a constant given in advance.

The virtual vector A 'is in the opposite direction to the query Q as shown in FIG.

Using this virtual vector A ′, a document having a feature vector closest to the virtual vector A ′ is searched, and a document D
Is extracted, and the contents of the document D are displayed as shown in FIG.

Similarly, for example, the user operates the “Related” button while the content of the document B in FIG. 4A is displayed, or the content of the document D in FIG. When the user operates the button "Related" while displaying the "?", The documents having the feature vectors having the highest similarity to the feature vectors of the documents B and D are extracted from the document group of the search result. .

On the other hand, for example, when the user operates the “no relation” button while displaying the contents of the document B in FIG. With respect to the feature vector of the document A, a virtual vector B 'in the opposite direction to the feature vector of the document B is obtained by Expression 4, and a document having a feature vector having the highest similarity to the virtual vector B' is searched. . As a result, the document D is extracted and its contents are displayed.

In the following, in a similar manner, a document group having a content close to the content required by the user can be sequentially presented with priority, reflecting the user's judgment input.

In the system shown in FIG. 1, depending on the performance of the computer used, the query input unit 1 to the similarity calculation unit 4 have functions on the server side, and can be connected to the client side connected to a LAN, the Internet or other networks. A system configuration including the search result holding unit 5, the search result output unit 7, and the feedback processing unit 8 can be adopted. Alternatively, a query input unit 1 to a search result holding unit 5
May be a server-side function, and a system in which a search result output unit 7 and a feedback processing unit 8 are provided on the client side connected to the LAN or the Internet or another network may be used.

Next, an information retrieval system according to a second embodiment of the present invention will be described with reference to FIG. The information search system according to the second embodiment has, as functional components, a query input unit 1, a feature vector creation unit 2, a document database 3, and a query unit similar to those of the system according to the first embodiment shown in FIG. The apparatus includes a similarity calculation unit 4, a search result holding unit 5, a search result output unit 7, and a feedback processing unit 8. The system according to the present embodiment further creates a decision tree by a logical operation described later based on the feature vector for each document stored in the search result holding unit 5, and stores the created decision tree data in the search result holding unit. A decision tree creating unit 6 to be held by the unit 5 is provided.

Next, an information retrieval method by the information retrieval system according to the second embodiment having the above-described configuration will be described with reference to the flowchart of FIG. In this information retrieval system as well, similar to the system of the first embodiment shown in FIG. 2, text data expressing the content required by the user or text data of a document including the content required by the user. Is input as a query (step S
1) A feature vector is created from the input text data of create (step S3). Then, the similarity between the feature vector of each of the large number of documents stored in the document database 3 and the feature vector of the query is calculated, and a document having a feature vector indicating a similarity higher than a predetermined threshold is specified. , Their indexes, text data, and feature vectors (step S
5) The data is temporarily stored in the search result holding unit 5 as search result data (step S7).

For the search results obtained in this way, the decision tree creating unit 6 which is a feature of the present embodiment creates a decision tree as shown in FIG. It is held (step S8). Then, the search result output unit 7 displays information representing the content of the document having the feature vector having the highest similarity to the query (step S9).

In response to the display of the contents of the document, when the user inputs a determination as to whether or not the document has the content desired by the user via the UI device (YES) or not (NO), the feedback processing unit 8 makes this determination. Accepts the input and the next Y
Perform either ES / NO processing (step S1)
1).

The user determines that the content of the document presented first is the content requested by the user, and determines YES.
Is input, the feedback processing unit 8 refers to the decision tree held in the search result holding unit 5, specifies the next document to be shifted to the case of YES for the currently output document, and It is displayed by the search result output unit 7 (steps S12, S16, S18).

If the user determines that the content of the document presented first is not the content he / she has requested and inputs NO in step S11, the feedback processing unit 8 is stored in the search result storage unit 5. With reference to the decision tree that is present, the next document to be shifted to in the case of NO for the currently output document is specified, and its contents are displayed by the search result output unit 7 (steps S14, S16, S18).

As described above, when the user inputs a YES / NO determination through the UI device with respect to the content of the document displayed second in step S18, the process proceeds to step S11.
To S18 are repeated. The same applies to the third and subsequent documents. Then, when there is no document corresponding to the search result holding unit 5, the output of the search result is ended (step S1).
6 branches to NO).

Next, the decision tree creation processing by the decision tree creation section 6 will be described with reference to the example of the decision tree in FIG. It is assumed that feature vector groups A to E having similarities within the threshold value have been searched for the feature vector Q of the query. The similarity to the query Q is assumed to be higher as the distance from the query Q is shorter, and the vector direction is shown as a relative position to the query Q.

The decision tree creator 6 starts with the feature vector A having the highest similarity for the feature vectors A to E having the higher similarity obtained by the similarity calculator 4, and when the content is accepted by the user. If the answer is YES, the process branches to NO.

If the determination is YES, the document having the feature vector B having the highest similarity to the feature vector A is set as the document to be output next to the document having the feature vector A. On the other hand, when branching from the document of the feature vector A to NO, a virtual vector A ′ in the opposite direction to the feature vector A is assumed based on the query Q as described above,
The document of the feature vector D closest to the virtual vector A 'is determined as the document to be displayed next. And
Similarly, the process branches to YES and NO for each feature vector, and sequentially extracts vectors to be output from the vector group of the search result, thereby creating a decision tree as shown in FIG.

In this way, when the user sequentially checks the contents from the group of documents searched for the query Q, the feedback processing unit 8 creates the decision tree creating unit 6 and creates the search result holding unit 5 With reference to the information of the decision tree shown in FIG.

First, A is output as a document having a feature vector closest to the query Q. On the other hand, if the user performs a determination operation of NO, the feedback processing unit 8 causes the search result output unit 7 to output the document of the feature vector D next. If the user performs a determination operation of NO for the document of the feature vector D, the document of the feature vector C is output next. If YES for the document of the feature vector C, the document of the feature vector B is output next. Finally, the document of the feature vector E is output.

FIG. 9 is a flowchart generally showing the decision tree creation processing by the decision tree creation unit 6. A document group having a feature vector of similarity within the threshold value is searched for the query Q input first, and the processing of the flowchart in FIG. 9 is repeated in the document group to create a decision tree. Keep it.

According to the second embodiment, particularly in a system having a server capable of high-speed processing, a user who accesses the server as a client and receives a search service has his or her own client machine. Even if the search result does not have high-speed processing capability, the search results can be browsed at high speed in order from the one with the highest priority.

Next, an information retrieval system according to a third embodiment of the present invention will be described with reference to the flowchart of FIG. 10 and the operation explanatory diagram of FIG. The functional configuration of the information search system according to the third embodiment is common to that of the second embodiment shown in FIG. However, the decision tree creator 6 applies to all the documents extracted by the similarity calculator 4 as shown in FIG.
Instead of creating a decision tree as shown in (1), the decision tree is grown in predetermined steps in accordance with the user's judgment input.

That is, if the document group having the feature vector within the threshold value obtained for the query Q by the similarity calculation unit 4 is the one shown in FIG. 13, the decision tree creation unit 6 first As shown in FIG. 11A, for example, up to three levels (the number of levels is not specified and may be up to two levels, or may be four or more levels if the number of documents in the search result is large) A decision tree is created (step S31).

If the user determines that the contents of the first stage document A are relevant (YES) (step S
33-S37) Since the direction of document development is almost determined,
A decision tree is grown as shown in FIG.
39, S41, S43). Then, a document B having the shortest distance feature vector with respect to the feature vector of the document A is extracted and output (steps S45 and S35).

In step S41, the condition for growing the decision tree is that, for example, unpresented documents remain in the document group as a search result, but the decision tree up to the current stage reaches the terminal document. In this case, the decision tree is increased by one every time one stage or a predetermined number of candidate documents are determined.
It is grown step by step or an appropriate number of steps. The advance condition can be determined in advance, for example, when a document of one stage or an appropriate number of stages is presented.

According to the third embodiment, the decision of the user is fed back, and a decision tree is created only for a system which is likely to be expanded, so that unnecessary calculation processing can be reduced. It is possible to reduce the calculation load of the server.

In each of the above embodiments, the search service and the arithmetic processing are all executed on the server machine. All or a part of the processing functions described above can be provided on the client machine side and connected to the server machine side via a network.

Also, in the system shown in FIG. 6, depending on the performance of the computer used, the query input unit 1 to the similarity calculation unit 4 function as a server, and a client connected to a LAN, the Internet, or another network. A system configuration including the search result holding unit 5, the decision tree creating unit 6, the search result output unit 7, and the feedback processing unit 8 can be adopted. Alternatively, query input units 1 to
The decision tree creating unit 5 may be a server-side function, and a system configuration in which a search result output unit 7 and a feedback processing unit 8 are provided on the client side connected to the LAN or the Internet or another network may be used.

The present invention provides a program for realizing the processing functions of the information retrieval system according to each of the above embodiments, a computer-readable recording medium on which the program is recorded, and a computer readable program by incorporating the program into a computer. The information retrieval method executed by the system is also within the technical scope.

Further, in each of the embodiments described above, the document information retrieval based on the text data has been described as an example. However, the present invention is not limited to this. Can be applied. Also, regarding music information and image information, the data is converted into a feature vector and registered in a database, and the text data of the index, title, author, publisher, sales company, etc. are also registered together, and the music information is used as a query. , Data representing image information is directly input and converted into a feature vector, music information and image information having a high similarity between the feature vectors are extracted, and the music information or image information data, feature vector, index, title Is retrieved from the database, it is possible to perform a search using the same decision tree as in the above embodiments.

[0088]

As described above, according to the present invention, a group of information in which the similarity of a feature vector is higher than a threshold value for a query input by a user is extracted, and the feature vector having the highest similarity among them is extracted. Is presented first, and if the user determines that the information is appropriate (YES), information having a feature vector with a high similarity to the feature vector of the information is presented next, and conversely, the information is inappropriate. If determined to be (NO), the procedure of using the virtual vector in the opposite direction to the feature vector and presenting the information having the feature vector with a high similarity to the next is sequentially repeated thereafter, and the information requested by the user is repeated. Is predicted and presented preferentially. Therefore, even when there is a large number of pieces of information having a high similarity to the information required by the user, whether the user is appropriate / unsuitable for one piece of presented information is determined. Correspondingly, the user can successively presented automatically selects the information of the contents in need.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a functional configuration of an information search system according to a first embodiment of the present invention.

FIG. 2 is a flowchart of a search process and a search result output process according to the embodiment.

FIG. 3 is an explanatory diagram of a display example of a search processing result according to the embodiment.

FIG. 4 is an explanatory diagram showing transition of display output of a search result according to the embodiment.

FIG. 5 is an explanatory diagram of search processing output processing according to the embodiment.

FIG. 6 is a block diagram showing a functional configuration of an information search system according to a second embodiment of the present invention.

FIG. 7 is a flowchart of a search process and a search result output process of the embodiment.

FIG. 8 is an explanatory diagram of a decision tree creation process for a search result according to the embodiment.

FIG. 9 is a flowchart of a decision tree creation process for a search result according to the embodiment.

FIG. 10 is a flowchart of a decision tree creation process for a search result by the information search system according to the third embodiment of the present invention.

FIG. 11 is an explanatory diagram of a decision tree creation process for a search result according to the embodiment.

FIG. 12 is an explanatory diagram illustrating an output example of a search processing result based on the similarity in the conventional example.

FIG. 13 is an explanatory diagram of a search process based on similarity in a conventional example.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Query input part 2 Feature vector preparation part 3 Document database 4 Similarity calculation part 5 Search result holding part 6 Decision tree preparation part 7 Search result output part 8 Feedback processing part

 ──────────────────────────────────────────────────の Continuation of the front page (54) [Title of the invention] Information retrieval system, information retrieval method, information retrieval program, recording medium recording information retrieval program, output information selection device, output information selection method, output information selection Recording medium on which program and output information selection program are recorded

Claims (28)

[Claims] A query input unit for receiving a query input by a user; a feature vector creating unit for creating a feature vector from the query received by the query input unit; An information database in which feature vector data is registered; a similarity between a feature vector of a query created by the feature vector creation unit and a feature vector for each search target information registered in the information database is calculated; A similarity calculation unit for specifying search target information having a feature vector indicating a similarity higher than the threshold value, and extracting the index, data representing the content, and feature vector data; A search result storage unit for storing data, and a search target retrieved by the similarity calculation unit A search result output unit for outputting data indicating the index and content of one of the reports, and accepting a user's suitability / inappropriate determination input with respect to the search result output by the search result output unit; In the case of, the search result holding unit searches for other search target information having a feature vector having the highest similarity to the feature vector of the search target information currently being output, and the corresponding search target information is If so, the content is output as the next candidate by the search result output unit. (2) In the case of inappropriate input, if it is the first stage, 2
If it is at or after the stage, a virtual vector that is in the opposite direction to the feature vector of the search target information currently being output is calculated with respect to the feature vector output at the previous stage, and the feature vector with the highest similarity to this virtual vector is calculated. A feedback process for searching the search target information held in the search result holding unit for other search target information having the search result information, and outputting the content of the corresponding search target information as a next candidate by the search result output unit, if any. Information retrieval system comprising: 2. A query input unit for receiving a query input by a user; a feature vector generating unit for generating a feature vector from the query received by the query input unit; An information database in which feature vector data is registered; a similarity between a feature vector of a query created by the feature vector creation unit and a feature vector for each search target information registered in the information database is calculated; A similarity calculation unit for specifying search target information having a feature vector indicating a similarity higher than the threshold value, and extracting the index, data representing the content, and feature vector data; A search result storage unit for storing data; and a high similarity calculated by the similarity calculation unit. For the feature vector group, starting from the feature vector with the highest similarity, it is suitable if its contents are accepted by the user, inappropriately branches if it is not accepted, and inappropriately branches if it is not accepted, If the other search target information having the feature vector having the highest similarity to the feature vector is set as the next candidate search target information, and the branch is inappropriately branched from the feature vector,
If it is the first stage, the feature vector of the query is calculated. If it is the second stage or later, the virtual vector that is in the opposite direction to the feature vector of the inappropriate search target information is calculated for the feature vector that was the output candidate of the previous stage. Then, another search target information having a feature vector having the highest similarity to the virtual vector is searched, and if there is such search target information, the content is set as the next candidate search target information. A decision tree creating unit that repeatedly creates a decision tree until all the search target information held in the unit is covered, and data indicating the index and contents of one of the search target information extracted by the similarity calculation unit A search result output unit for outputting a search result, and receiving a user's suitability / inappropriate determination input for the search result output from the search result output unit, and generating the decision tree creating unit Then identifies the search target information to be output, the corresponding search target information retrieval system comprising a feedback unit for outputting by said search result output unit contents based on decision tree was. 3. A query input unit for receiving a query input by a user, a feature vector generating unit for generating a feature vector from the query received by the query input unit, an index of a large number of search target information, data representing contents, and An information database in which feature vector data is registered; a similarity between a feature vector of a query created by the feature vector creation unit and a feature vector for each search target information registered in the information database is calculated; A similarity calculation unit for specifying search target information having a feature vector indicating a similarity higher than the threshold value, and extracting the index, data representing the content, and feature vector data; A search result storage unit for storing data; and a high similarity calculated by the similarity calculation unit. For the feature vector group, starting from the feature vector with the highest similarity, it is suitable if its contents are accepted by the user, inappropriately branches if it is not accepted, and inappropriately branches if it is not accepted, If the other search target information having the feature vector having the highest similarity to the feature vector is set as the next candidate search target information, and the branch is inappropriately branched from the feature vector,
If it is the first stage, the feature vector of the query is calculated. If it is the second stage or later, the virtual vector that is in the opposite direction to the feature vector of the inappropriate search target information is calculated for the feature vector that was the output candidate of the previous stage. Then, another search target information having a feature vector having the highest similarity to the virtual vector is searched, and if there is such search target information, the content is set as the next candidate search target information. A decision tree creating unit for repeatedly creating a decision tree for a group of search target information held in a unit up to a predetermined stage, and data indicating the index and contents of one of the search target information extracted by the similarity calculation unit. A search result output unit to be output, and accepting a user's suitability / inappropriate determination input with respect to the search result output by the search result output unit; The next search target information to be output is specified based on the determined decision tree, the content of the corresponding search target information is output by the search result output unit, and the existing decision tree is output to the decision tree creation unit. An information retrieval system comprising: a feedback processing unit that gives an instruction to grow the decision tree by a predetermined stage to the decision tree creating unit when a condition for growing is reached. 4. The information search system according to claim 1, wherein the query and the search target information are text data. 5. A step 1 for receiving a query input by a user, a step 2 for creating a feature vector from the received query, a feature vector for the created query, and a search for each piece of search target information registered in the information database. Calculating a similarity with the feature vector, identifying search target information having a feature vector indicating a similarity higher than a predetermined threshold, and extracting the index, data representing the content, and feature vector data; Step 4 for outputting data indicating the index and contents of one of the search target information extracted in Step 3, and accepting a user's suitability / inappropriate determination input for the search result output in Step 4; ) In the case of appropriate input, the similarity is the highest in the feature vector of the search target information currently being output. The search target information retrieved in step 3 is searched for other search target information having a new feature vector, and if there is such search target information, its contents are output as the next candidate, and (2) improper input In the case of, in the first stage, the feature vector of the query is output in the opposite direction to the feature vector of the query, and in the second and subsequent stages, with respect to the feature vector output in the previous stage, A virtual vector is calculated, and the virtual vector is searched for other search target information having a feature vector having the highest similarity in the search target information extracted in step 3. Outputting the content as a next candidate. 6. A step 1 for receiving a query input by a user, a step 2 for creating a feature vector from the received query, a feature vector of the created query, and a search for each search target information registered in the information database. Calculating a similarity with the feature vector, identifying search target information having a feature vector indicating a similarity higher than a predetermined threshold, and extracting the index, data representing the content, and feature vector data; From the feature vector group having a high similarity obtained in step 3, start with the feature vector having the highest similarity and branch appropriately if the content is acceptable to the user or inappropriate if the content is not acceptable. However, if branching is appropriate, another search pair having the feature vector with the highest similarity to the feature vector The elephant information is used as search target information of the next candidate,
In the case of inappropriate branching from the feature vector, if it is the first stage, the query vector is the query feature vector. Calculate the virtual vector in the opposite direction to the feature vector that has it, search for other search target information that has the feature vector with the highest similarity to this virtual vector, and if there is applicable search target information, replace its contents with the next candidate. Step 4 of creating a decision tree by repeating the process of setting the search target information until all the search target information extracted in Step 3 is covered, and converting the index and contents of one of the search target information extracted in Step 3 Step 5 of outputting the data shown in the step S5; In identifying the search target information to be next output based on the decision tree created,
Outputting the content of the relevant search target information. 7. A step 1 for receiving a query input by a user, a step 2 for creating a feature vector from the received query, a feature vector of the created query, and a search for each search target information registered in the information database. Calculating a similarity with the feature vector, identifying search target information having a feature vector indicating a similarity higher than a predetermined threshold, and extracting the index, data representing the content, and feature vector data; From the feature vector group having a high similarity obtained in step 3, start with the feature vector having the highest similarity and branch appropriately if the content is acceptable to the user or inappropriate if the content is not acceptable. However, if branching is appropriate, another search pair having the feature vector having the highest similarity to the feature vector The elephant information is used as search target information of the next candidate,
In the case of inappropriate branching from the feature vector, if it is the first stage, the query vector is the query feature vector. Calculate the virtual vector in the opposite direction to the feature vector that has it, search for other search target information that has the feature vector with the highest similarity to this virtual vector, and if there is applicable search target information, replace its contents with the next candidate. Step 4 of creating a decision tree by repeating the process of setting the search target information as the search target information at a predetermined stage with respect to the search target information extracted in Step 3, and data indicating the index and contents of one of the search target information extracted in Step 3 Step 5; and accepting the user's appropriate / unsuitable input for the search result output in step 5; Identify next search target information to be output based on the decision tree created,
Output the contents of the relevant search target information and step 4
And a step of growing the decision tree by a predetermined stage if the condition for growing the decision tree created in the step is reached. 8. The information search method according to claim 5, wherein the query and the search target information are text data. 9. A process 1 for receiving a query input by a user, and a process 2 for creating a feature vector from the received query.
And calculating a similarity between the feature vector of the created query and a feature vector for each search target information registered in the information database, and searching for a feature vector having a similarity higher than a predetermined threshold. A process 3 for specifying the target information and extracting the index and the data representing the content and the feature vector data; a process 4 for outputting the index and the content indicating the index and the content of one of the search target information extracted in the process 3; The user's input for determining whether or not the search result is output in the process 4 is accepted. (1) In the case of the appropriate input, the feature vector having the highest similarity to the feature vector of the search target information currently being output. Searches the search target information extracted in the above process 3 for other search target information having (2) In the case of inappropriate input, the search target currently being output is the feature vector of the query in the first stage, and the feature vector output in the previous stage in the second and subsequent stages. A virtual vector that is in the opposite direction to the feature vector of the information is calculated, and there is no other search target information having a feature vector having the highest similarity in the virtual vector, and the search is performed in the search target information extracted in the process 3. An information search program for causing a computer to execute a process 5 of outputting the corresponding search target information, if any, as the next candidate. 10. A process 1 for receiving a query input by a user, and a process 2 for creating a feature vector from the received query.
And calculating a similarity between the created feature vector of the query and a feature vector for each piece of search target information registered in the information database, and searching for a feature vector having a similarity higher than a predetermined threshold. A process 3 for identifying the target information and extracting the data representing the index and the contents thereof and the feature vector data; and a feature vector group having a high similarity obtained in the process 3;
Starting with the feature vector with the highest similarity, it is suitable if its contents are accepted by the user, inappropriately branches if it is not accepted, and inappropriately branches if it is not accepted,
Other search target information having a feature vector having the highest similarity to the feature vector is set as search target information of the next candidate. In the second and subsequent stages, a virtual vector having the opposite direction to the feature vector of the inappropriate search target information is calculated with respect to the feature vector set as the output candidate in the preceding stage, and the virtual vector has the highest similarity. Search for other search target information with feature vector,
If there is applicable search target information, the process of setting the content as the next candidate search target information is repeated until all search target information extracted in process 3 is covered, and a process 4 for creating a decision tree; Processing 5 for outputting data indicating the index and content of one of the retrieved information to be searched, and accepting the user's input for determining whether or not the search result output in processing 5 is appropriate, and the decision made in processing 4 An information search program for causing a computer to execute processing 6 for specifying the next search target information to be output based on the tree and outputting the contents of the relevant search target information. 11. A processing 1 for receiving a query input by a user, and a processing 2 for creating a feature vector from the received query.
And calculating a similarity between the created feature vector of the query and a feature vector for each piece of search target information registered in the information database, and searching for a feature vector having a similarity higher than a predetermined threshold. A process 3 for identifying the target information and extracting the data representing the index and the contents thereof and the feature vector data; and a feature vector group having a high similarity obtained in the process 3;
Starting with the feature vector with the highest similarity, it is suitable if its contents are accepted by the user, inappropriately branches if it is not accepted, and inappropriately branches if it is not accepted,
Other search target information having a feature vector having the highest similarity to the feature vector is set as search target information of the next candidate. In the second and subsequent stages, a virtual vector having the opposite direction to the feature vector of the inappropriate search target information is calculated with respect to the feature vector set as the output candidate in the preceding stage, and the virtual vector has the highest similarity. Search for other search target information with feature vector,
If there is applicable search target information, the process of making the contents the search target information of the next candidate is repeated to a predetermined stage with respect to the search target information extracted in process 3 to create a decision tree, and the process extracted in process 3 Processing 5 for outputting data indicating the index and content of one of the search target information; accepting the user's appropriateness / inappropriate determination input with respect to the search result output in processing 5; Based on the search target information to be output next, the contents of the corresponding search target information are output, and if the condition for growing the decision tree created in process 4 is reached, the decision tree is grown by a predetermined level. An information search program for causing a computer to execute the processing 6 to be executed. 12. The information search program according to claim 9, wherein the query and the search target information are text data. 13. A process 1 for receiving a query input by a user, and a process 2 for creating a feature vector from the received query.
And calculating a similarity between the feature vector of the created query and a feature vector for each search target information registered in the information database, and searching for a feature vector having a similarity higher than a predetermined threshold. A process 3 for specifying the target information and extracting the index and the data representing the content and the feature vector data; a process 4 for outputting the index and the content indicating the index and the content of one of the search target information extracted in the process 3; The user's input for determining whether or not the search result is output in the process 4 is accepted. (1) In the case of the appropriate input, the feature vector having the highest similarity to the feature vector of the search target information currently being output. Searches the search target information extracted in the above process 3 for other search target information having (2) In the case of unsuitable input, the search target currently being output is the feature vector of the query in the first stage, and the feature vector output in the previous stage in the second and subsequent stages. A virtual vector that is in the opposite direction to the feature vector of the information is calculated, and there is no other search target information having a feature vector with the highest similarity in the virtual vector, and the search is performed in the search target information extracted in the process 3. And a computer-readable recording medium on which an information search program for executing a process 5 for outputting the content of the corresponding search target information, if any, as the next candidate. 14. A process 1 for receiving a query input by a user, and a process 2 for creating a feature vector from the received query.
And calculating a similarity between the created feature vector of the query and a feature vector for each piece of search target information registered in the information database, and searching for a feature vector having a similarity higher than a predetermined threshold. A process 3 for identifying the target information and extracting the data representing the index and the contents thereof and the feature vector data; and a feature vector group having a high similarity obtained in the process 3;
Starting with the feature vector with the highest similarity, it is suitable if its contents are accepted by the user, inappropriately branches if it is not accepted, and inappropriately branches if it is not accepted,
Other search target information having a feature vector having the highest similarity to the feature vector is set as search target information of the next candidate. In the second and subsequent stages, a virtual vector having the opposite direction to the feature vector of the inappropriate search target information is calculated with respect to the feature vector set as the output candidate in the preceding stage, and the virtual vector has the highest similarity. Search for other search target information with feature vector,
If there is applicable search target information, the process of setting the content as the next candidate search target information is repeated until all the search target information extracted in the process 3 is covered, and a process 4 for creating a decision tree; Processing 5 for outputting data indicating the index and content of one of the retrieved information to be searched, and accepting the user's input for determining whether or not the search result output in processing 5 is appropriate, and the decision made in processing 4 A computer-readable recording medium which records an information search program for executing processing 6 for specifying the next search target information to be output based on the tree and outputting the contents of the relevant search target information. 15. A process 1 for receiving a query input by a user, and a process 2 for creating a feature vector from the received query.
And calculating a similarity between the created feature vector of the query and a feature vector for each piece of search target information registered in the information database, and searching for a feature vector having a similarity higher than a predetermined threshold. A process 3 for identifying the target information and extracting the data representing the index and the contents thereof and the feature vector data; and a feature vector group having a high similarity obtained in the process 3;
Starting with the feature vector with the highest similarity, it is suitable if its contents are accepted by the user, inappropriately branches if it is not accepted, and inappropriately branches if it is not accepted,
Other search target information having a feature vector having the highest similarity to the feature vector is set as search target information of the next candidate. In the second and subsequent stages, a virtual vector having the opposite direction to the feature vector of the inappropriate search target information is calculated with respect to the feature vector set as the output candidate in the preceding stage, and the virtual vector has the highest similarity. Search for other search target information with feature vector,
If there is corresponding search target information, the process of making the contents the search target information of the next candidate is repeated to a predetermined stage for the search target information extracted in process 3 to create a decision tree, and the process extracted in process 3 Processing 5 for outputting data indicating the index and contents of one of the information to be searched; and accepting the user's appropriateness / inappropriate determination input with respect to the search result output in processing 5; Based on the search target information to be output next, the contents of the relevant search target information are output, and if the condition for growing the decision tree created in process 4 is reached, the decision tree is grown by a predetermined stage. And a computer-readable recording medium on which an information search program for executing the processing 6 is recorded. 16. The computer-readable recording medium according to claim 13, wherein the query and the search target information are text data. 17. A feature data holding unit for holding feature vector data together with information representing the contents of each of the group of search target information, and expressing the contents of the search target information specified from the group of search target information. A display information output unit that outputs information; a feedback receiving unit that receives a user's appropriateness / inappropriate determination input with respect to the output search target information; and (1) when the user's determination input is appropriate The display information output unit searches the set of search target information for other search target information having a feature vector having the highest similarity to the feature vector of the search target information currently being output,
If there is applicable search target information, the information representing its contents is output to the display information output unit. (2) If the user's judgment input is inappropriate, if it is the first stage, it is given in advance. With respect to the reference vector, if it is the second or subsequent stage, the display information output unit is oriented in the opposite direction to the feature vector of the search target information currently output, with respect to the feature vector of the search target information output in the previous stage. A virtual vector is calculated, and the group of search target information is searched for other search target information having a feature vector having the highest similarity to the virtual vector, and if there is such search target information, the content is expressed. An output information selection device comprising: an output information selection unit that outputs information to the display information output unit. 18. A feature data holding unit for holding feature vector data together with information representing the contents of each of the group of search target information, and expressing the contents of the search target information designated from the group of search target information. A display information output unit for outputting information; and a group of search target information held in the feature data holding unit, from a search target information having a feature vector having the highest similarity to a reference vector given in advance. Initially, if the content is accepted by the user, it is appropriate, if it is not accepted, it branches inappropriately, and if it branches appropriately, another feature vector having the highest similarity to the feature vector is used. The search target information is set as search target information of the next candidate, and when inappropriately branching from the feature vector, if it is the first stage, with respect to the reference vector,
If it is the second stage or later, the virtual vector that is in the opposite direction to the feature vector of the inappropriate search target information is calculated with respect to the feature vector of the search target information as the output candidate of the previous stage, and similar to this virtual vector. The other search target information having the highest feature vector is searched, and if there is the corresponding search target information, the process of setting the content as the next candidate search target information is performed on all of the features held in the feature data holding unit. A decision tree creating unit that repeatedly creates a decision tree until the search target information is covered, and accepts a user's suitability / inadequacy determination input for the search target information output by the display information output unit, and generates the decision tree. The search information to be output next is specified based on the decision tree created by the unit, and the information representing the contents of the relevant search target information is output by the display information output unit. An output information selection device comprising a feedback processing unit. 19. A feature data holding unit for holding feature vector data together with information expressing the contents of each of a group of search target information, and expressing the contents of the search target information specified from the group of search target information. A display information output unit for outputting information; and a group of search target information held in the feature data holding unit. Initially, if the content is accepted by the user, it is appropriate, if it is not accepted, it branches inappropriately, and if it branches appropriately, another feature vector having the highest similarity to the feature vector is used. The search target information is set as search target information of the next candidate, and when inappropriately branching from the feature vector, if it is the first stage, with respect to the reference vector,
If it is the second stage or later, the virtual vector that is in the opposite direction to the feature vector of the inappropriate search target information is calculated with respect to the feature vector of the search target information as the output candidate of the previous stage, and similar to this virtual vector. The other search target information having the highest feature vector is searched, and if there is the corresponding search target information, the content thereof is set as the next candidate search target information. A decision tree creating unit that repeatedly creates a decision tree for a target information group up to a predetermined stage; Specifying the next search target information to be output based on the created decision tree, and causing the display information output unit to output information representing the contents of the corresponding search target information A feedback processing unit for instructing the decision tree creating unit to grow the decision tree by a predetermined stage when a condition for growing an existing decision tree is reached. Information selection device. 20. A step 1 for holding feature vector data together with information representing the contents of each of the group of search target information, and information representing the contents of the search target information designated from the group of search target information. Outputting step 2; accepting a user's appropriateness / inappropriate determination input with respect to the currently output search target information; and (1) if the user's determination input is appropriate, the currently output search target The group of search target information is searched for other search target information having a feature vector having the highest similarity to the feature vector of the information, and if there is the corresponding search target information, information representing the content is output. (2) When the user's judgment input is inappropriate, the search target output in the first stage is the reference vector given in advance, and the search target output in the previous stage is the second or subsequent stage. With respect to the feature vector of the report, a virtual vector that is in the opposite direction to the feature vector of the currently output search target information is calculated, and other search target information having the feature vector with the highest similarity to this virtual vector is calculated A step of searching the group of search target information for the search target, and outputting information representing the contents of the corresponding search target information, if any. 21. A step 1 for holding feature vector data together with information representing the contents of each of the group of search target information, and information representing the contents of the search target information designated from the group of search target information. And outputting the search target information with the feature vector having the highest similarity to the reference vector given in advance for the group of search target information, and assuming that the content is acceptable to the user. Suitable, if not accepted and branches inappropriately, if appropriately branched, other search target information having a feature vector having the highest similarity to the feature vector as the next candidate search target information, In the case of inappropriate branching from the feature vector, if it is the first stage, the reference vector is used. For the feature vector of the elephant information, a virtual vector that is in the opposite direction to the feature vector of the inappropriate search target information is calculated, and other search target information having the feature vector with the highest similarity to this virtual vector is calculated. Step 3 of creating a decision tree by repeating the process of searching and, if there is applicable search target information, making the contents thereof the search target information of the next candidate until the entire set of search target information is covered; Accepting a user's input regarding suitability / inappropriateness of the search target information in the search target information, specifying the next search target information to be output based on the decision tree, and outputting information representing the contents of the relevant search target information 4. An output information selection method comprising: 22. A step 1 for holding feature vector data together with information representing the contents of each of the group of search target information, and information representing the contents of the search target information designated from the group of search target information. And outputting the search target information with the feature vector having the highest similarity to the reference vector given in advance for the group of search target information, and assuming that the content is acceptable to the user. Suitable, if not accepted and branches inappropriately, if appropriately branched, other search target information having a feature vector having the highest similarity to the feature vector as the next candidate search target information, In the case of inappropriate branching from the feature vector, if it is the first stage, the reference vector is used. For the feature vector of the elephant information, a virtual vector that is in the opposite direction to the feature vector of the inappropriate search target information is calculated. Step 3 of creating a decision tree by repeating the process of searching and, if there is applicable search target information, the contents thereof as the next candidate search target information up to a predetermined stage for the group of search target information; It accepts the user's input for determining whether or not the search target information is appropriate, specifies the next search target information to be output based on the decision tree, and outputs information expressing the contents of the relevant search target information. And a step of growing the decision tree by a predetermined stage when the condition for growing the decision tree is reached. 23. Processing 1 for retaining feature vector data together with information representing the contents of each of a group of search target information, and information representing the contents of the search target information designated from the group of search target information. Processing 2 to output, and accepting the user's appropriateness / inappropriate determination input with respect to the currently output search target information. (1) If the user's determination input is appropriate, the currently output search target The group of search target information is searched for other search target information having a feature vector having the highest similarity to the feature vector of the information, and if there is the corresponding search target information, information representing the content is output. (2) When the user's judgment input is inappropriate, if the first stage, the search target information output in the previous stage is stored in the second stage or later with respect to the given reference vector. With respect to the feature vector, a virtual vector that is in the opposite direction to the feature vector of the search target information currently being output is calculated, and if there is no other search target information having a feature vector with the highest similarity to the virtual vector, An output information selection program for causing a computer to execute a process 3 of searching for a group of search target information and outputting information representing the contents of the search target information, if any. 24. A process 1 for retaining feature vector data together with information representing the contents of each of a group of search target information, and information representing the contents of the search target information designated from the group of search target information. In the case where it is assumed that the process 2 to be output starts with the search target information having the feature vector having the highest similarity to the reference vector given in advance for the group of search target information, and the content thereof is accepted by the user. Suitable, if not accepted and branches inappropriately, if appropriately branched, other search target information having a feature vector having the highest similarity to the feature vector as the next candidate search target information, In the case of inappropriate branching from the feature vector, the search target information is regarded as the output candidate of the preceding stage with respect to the reference vector if it is the first stage, and the output candidate of the preceding stage if it is the second stage or later. With respect to the feature vector having, a virtual vector that is in the opposite direction to the feature vector of the inappropriate search target information is calculated, and other search target information having the feature vector with the highest similarity to this virtual vector is searched, A process 3 of creating a decision tree by repeating the process of setting the content of the corresponding search target information as the next candidate search target information, if any, until the entire set of search target information is covered; A process 4 of receiving a user's suitability / inappropriate determination input for the target information, specifying the next search target information to be output based on the decision tree, and outputting information representing the contents of the relevant search target information. Output information selection program to be executed by computer. 25. A process 1 for retaining feature vector data together with information representing the contents of each of a group of search target information, and information representing the contents of the search target information designated from the group of search target information. In the case where it is assumed that the process 2 to be output starts with the search target information having the feature vector having the highest similarity to the reference vector given in advance for the group of search target information, and the content thereof is accepted by the user. If it is not suitable, if it is not accepted, it branches improperly, and if it branches properly, other search target information having the feature vector having the highest similarity to the feature vector is set as the next candidate search target information, In the case of inappropriate branching from the feature vector, the search target information is regarded as the output candidate of the preceding stage with respect to the reference vector if it is the first stage, and the output candidate of the preceding stage if it is the second stage or later. With respect to the feature vector having, a virtual vector that is in the opposite direction to the feature vector of the inappropriate search target information is calculated, and other search target information having the feature vector with the highest similarity to this virtual vector is searched, A process 3 of creating a decision tree by repeating the process of setting the content of the corresponding search target information, if any, as the next candidate search target information to a predetermined stage for the group of search target information
Information that accepts a user's input for determining whether or not the search target information is currently being output, specifies the next search target information to be output based on the decision tree, and expresses the contents of the relevant search target information. And an output information selection program for causing a computer to execute a process 4 for growing the decision tree by a predetermined stage when conditions for growing an existing decision tree are reached. 26. A processing 1 for retaining feature vector data together with information representing the contents of each of a group of search target information, and information representing the contents of the search target information designated from the group of search target information. Processing 2 to output, and accepting the user's appropriateness / inappropriate determination input with respect to the currently output search target information. (1) If the user's determination input is appropriate, the currently output search target The group of search target information is searched for other search target information having a feature vector having the highest similarity to the feature vector of the information, and if there is the corresponding search target information, information representing the content is output. (2) When the user's judgment input is inappropriate, if the input is the first stage, the search target information output in the previous stage is output from the reference vector given in advance if the input is the second or subsequent stage. With respect to the feature vector, a virtual vector that is in the opposite direction to the feature vector of the search target information currently being output is calculated, and if there is no other search target information having a feature vector with the highest similarity to the virtual vector, A computer-readable recording medium for recording an output information selection program for executing a process 3 for searching a group of search target information and outputting information representing the content of the search target information, if any; 27. A process 1 for holding feature vector data together with information representing the contents of each of a group of search target information, and information representing the contents of the search target information designated from the group of search target information. In the case where it is assumed that the process 2 to be output starts with the search target information having the feature vector having the highest similarity to the reference vector given in advance for the group of search target information, and the content thereof is accepted by the user. Suitable, if not accepted and branches inappropriately, if appropriately branched, other search target information having a feature vector having the highest similarity to the feature vector as the next candidate search target information, In the case of inappropriate branching from the feature vector, the search target information is regarded as the output candidate of the preceding stage with respect to the reference vector if it is the first stage, and the output candidate of the preceding stage if it is the second stage or later. With respect to the feature vector having, a virtual vector that is in the opposite direction to the feature vector of the inappropriate search target information is calculated, and other search target information having the feature vector with the highest similarity to this virtual vector is searched, A process 3 of creating a decision tree by repeating the process of setting the content of the corresponding search target information as the next candidate search target information, if any, until the entire set of search target information is covered; A process 4 of receiving a user's suitability / inappropriate determination input for the target information, specifying the next search target information to be output based on the decision tree, and outputting information representing the contents of the relevant search target information. A computer-readable recording medium recording an output information selection program to be executed. 28. A process 1 for retaining feature vector data together with information representing the contents of each of a group of search target information, and information representing the contents of the search target information designated from the group of search target information. Processing 2 to output, for the group of search target information, starting from search target information having a feature vector having the highest similarity to a reference vector given in advance, and assuming that the content is acceptable to the user If it is not suitable, if it is not accepted, it branches improperly, and if it branches appropriately, other search target information having the feature vector having the highest similarity to the feature vector is set as the next candidate search target information, In the case of inappropriate branching from the feature vector, the search target information as the output candidate of the preceding stage if it is the first stage and the preceding stage if it is the second stage or later. With respect to the feature vector having, a virtual vector that is in the opposite direction to the feature vector of the inappropriate search target information is calculated, and other search target information having the feature vector with the highest similarity to this virtual vector is searched, A process 3 of creating a decision tree by repeating the process of setting the content of the corresponding search target information, if any, as the next candidate search target information to a predetermined stage for the group of search target information
Information that accepts the user's input of suitability / inadequacy for the currently output search target information, specifies the next search target information to be output based on the decision tree, and expresses the contents of the relevant search target information And a processing 4 for growing the decision tree by a predetermined stage when a condition for growing an existing decision tree is reached, and a computer-readable recording medium recording an output information selection program.