JP3373086B2 - Information retrieval device - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an information retrieval device, and in an information device equipped with an electronic information database, a document preparation device such as a word processor, or the like, target data is obtained by using the sensitivity information input by a user. The present invention relates to an information search device capable of searching.

[0002]

2. Description of the Related Art Conventionally, a large-scale database system has been used as a means for searching various data. This system focuses on how quickly and accurately searches a large amount of data and presents it to the user.Therefore, the search is performed by inputting keywords and based on the keywords, individual data Search for,
Alternatively, the data is searched using the index file.

Further, a system using a fuzzy search, in which an ambiguous word can be input as a keyword, or a character string of a keyword is judged to be almost coincident with a search target, a keyword thesaurus table is used. Thus, a search system that estimates the meaning concept and a system that searches for data by inputting a natural language and extracting an appropriate keyword from it have been devised.

[0004] For example, "Multimedia Search Using Knowledge of Encyclopedia", Musta, et al. Sharp Technical Report No. 60/1
In December 994 (reference d), a method of expanding a word into a meaning vector and performing image search by the word is proposed.

Further, a character processing device with a postcard printing function which the applicant of the present application has already applied (Japanese Patent Application No. 3-265097).
No.) (Document a), a document creation device for creating a New Year's card is proposed.

This document creating apparatus has a postcard printing function in which a New Year's card design is automatically laid out by simply inputting a new word, an annotation, and an illustration candidate to be displayed on the screen. The information search function of this device presents the input items of "handwriting feeling", "delicate", "POP feeling", "powerful", and "standard" using the "completion image selection screen", and an appropriate one is selected from them. When the user selects, an image file group suitable for the “sensitivity” selected by the user is selected. Also,
Japanese Unexamined Patent Publication No. 61-3289 (Reference b) discloses an interior evaluation scale for deriving an interior most suitable for a user's intended use intention and atmosphere based on user's evaluation. This evaluation scale is calculated by creating a multiple regression equation from the results of a questionnaire conducted on users (the user's intention and image evaluation words based on a number of adjective pairs), and arranging interior image photographs on predetermined coordinates. This is a structure that can evaluate the interior image of a thing.

As a search using the factor space, Japanese Patent Application No. 6-223298 (Document c) has already disclosed an optimal document file extraction mechanism for sensitivity input proposed by the present inventor. In this file extraction mechanism, the user's sensitivity is input using adjective information, and a document file or the like is searched based on the input adjective information (adjective input value).

This file extraction mechanism is (1) database search by the adjective input value, (2) database search by factor analysis of adjectives in factor space, and (3) user preference by learning factor search results. It is characterized by the adaptation of.

[0009]

However, each of the above-mentioned prior arts has the following problems.

In the above-mentioned document (a), the object that can be extracted (searched) is the file group level, and the files in the file group cannot be prioritized. Further, the sensitivity input is limited to five types, and even with the same sensitivity input, the degree of sensitivity is not taken into consideration, and detailed proposal cannot be made.

Further, in the reference (b), it is very useful for the designer of the interior to evaluate and classify in advance through questionnaires, but on the contrary, suggest the interior requested by a user through this evaluation scale. Is not mentioned.

Further, in the reference document (c), in order to solve these problems, the sensitivity database using adjectives and factor scores obtained by factor analysis of the adjectives are used in an adjective space and a factor space in order to solve these problems. Then, the data close to the user's emotional input is prioritized and presented by calculating the distance.

However, according to this method, as the number of databases increases, the ranks of data that are very close to each other in the space become almost meaningless, and the order of data becomes smaller. No means for narrowing down the data was presented so that the difference in sensitivity appears.

According to the present invention, by gradually narrowing down the data once searched in the factor space, the difference in the sensitivity information of the data which did not appear in the space targeted for all the searched data is taken in, and the sensitivity of the user is improved. Information that is close to can be more accurately ranked and can be presented accurately, and as a result, based on the user's emotional input information, data that can present data as close to the user's emotion as possible from the database with emotional information The purpose is to obtain a search device.

[0015]

An information retrieval apparatus according to the present invention retrieves multimedia data including character information, graphic information, image information, image information and sound information.

This information retrieval apparatus includes sensitivity input means for inputting information corresponding to a person's sensitivity as position information in a multidimensional adjective coordinate space having dimensions corresponding to a plurality of adjectives, respectively. The position information in the adjective coordinate space attached to the multimedia data is factor-analyzed to set a factor coordinate space having a predetermined correspondence relationship with the adjective coordinate space, and each multimedia data is stored in the factor coordinate space. Factor analysis means corresponding to coordinates, conversion means for mapping the input affective information on the factor coordinate space, position in the factor coordinate space corresponding to the input affective information, and the multimedia data. And a distance calculation means for calculating the distance to the position in the factor coordinate space.

Then, this information retrieval apparatus is arranged such that the multimedia data located within a predetermined distance from the position of the inputted feeling information on the main factor coordinate space obtained by the factor analysis of all the multimedia data. To obtain a sub-data set consisting of the obtained multimedia data, perform a factor analysis of the position information in the adjective coordinate space of the multimedia data included in the sub-data set,
A sub-factor space whose correspondence with the adjective coordinate space is different from the main factor coordinate space is set, and in the sub-factor space,
It is configured to search the multimedia data based on the inputted sensitivity information. Thereby, the above object is achieved.

The present invention provides the above information retrieval device,
The process of converting the position information in the adjective coordinate space into the position information in the factor coordinate space is performed based on the factor load amount that defines the relationship between the position in the adjective coordinate space and the position in the factor coordinate space. It is a thing.

The operation will be described below.

In the present invention, as described above, since data having sensitivity information closer to the user's sensitivity information is presented, when searching in the factor coordinate space (hereinafter also referred to as factor space), the target (target Rather than narrowing down the search data) and prioritizing it, a subset is constructed by data that falls within a certain range, that is, within a certain distance from the user sensitivity value, and this subset is factor-analyzed The new factor space in is reconstructed, and the ranked data are presented by the distance to the user's sensitivity value again in this space.

Thus, by reconstructing the factor space within the sub-set, the differences between the data, which are hardly clarified due to the correlation with the entire data in the factor space with respect to the entire data, are clarified. Can be

As a result, when the user searches for data to be used for creating a document, etc., satisfying data can be retrieved from a database system having a large amount of data.
Data that is close to the user's sensibilities can be presented in order simply by performing sensibility input rather than complicated keyword or character input.

Further, when creating a document such as a multimedia document in which images and the like are abundantly included, it is easy for an amateur even to design image data that cannot be searched for by a professional with a good design sense. It can be quickly searched and used to create documents.

[0024]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described below.

FIG. 1 is a diagram conceptually showing the system configuration of an information retrieval apparatus according to an embodiment of the present invention.

In the figure, reference numeral 1 is an operation panel for inputting a user's sensitivity as position information in a multidimensional adjective coordinate space having dimensions corresponding to each of a plurality of adjectives. Further, the panel 1 can add emotional information to the searched multimedia data in advance, and the emotional information thus obtained is a three-dimensional common factor (α, β) that defines each adjective derived by factor analysis. , Γ) as an axis and are associated with positions on the factor space 2.

Reference numeral 2a is a sub-set of data that is within a certain distance from the user sensitivity value on the factor space 2. A new factor space can be reconstructed by factoring the data in this subset.

In short, this system is composed of a system for constructing a sensitivity database for arranging each data of the database in the factor space and a search system for the sensitivity database.

Further, FIG. 3 is a block diagram showing a configuration for data retrieval in the present information retrieval apparatus. In addition,
The solid line in FIG. 3 indicates the flow of data (sensitivity input value).

In the figure, 100 is this information retrieval apparatus,
An adjective input means for inputting information corresponding to a plurality of adjectives as position information (sensibility input) in a multidimensional adjective coordinate space having dimensions corresponding to each of the plurality of adjectives 21 and a factor load amount storage unit 23 that stores a factor load amount that is a parameter when converting the position information of the adjective coordinate space into the position information of the factor coordinate space having a dimension smaller than the dimension of the space.
And input adjective conversion means 22 for converting position information in the adjective coordinate space for the sensitivity input to position information (factor vector) in the factor coordinate space based on the factor load amount. The corresponding factor vector is stored in the input value storage means 24.

The information retrieval apparatus 100 further includes a database 26 including a plurality of document files (data to be retrieved) to which sensitivity information about the data is added as position information (factor vector) in the factor coordinate space, and the conversion means. Based on the factor vector converted from the sensitivity input by 22 and the factor vector added to the document file in the database 26, a distance calculating means 25 for calculating the distance between the two and a calculation of the distance calculating means 25. Based on the result, a document file extraction unit 27 that performs a process of extracting a candidate of a document file suitable for the sensitivity input or a process of optimally ranking the document files with respect to the sensitivity input, and the extracted document file candidate Alternatively, it has a document file presenting means 28 for presenting the priority of the document file candidates.

The information retrieving apparatus 100 also performs factor analysis based on the calculation result of the distance calculating means 25,
It has a factor analysis means 29 for changing the factor load amount stored in the factor load amount storage unit 23. Further, the factor analysis means 29 converts the sensitivity input value in the factor coordinate space used for the distance calculation into a value in the adjective coordinate space and returns it to the input adjective conversion means 22. That is, the factor analysis unit 29 sets a new factor space for the document files in the sub-set, and the document files are narrowed down in the factor space corresponding to the sub-set.

Needless to say, in this system, the sensitivity parameter (factor vector) must be added to the data used for the search, that is, the searched data in the database in advance.

Next, the operation will be described.

As described above, this information retrieval apparatus has a phase of a function of constructing a sensitivity database and a function of performing a search from the database (sensitivity engine). First, the sensitivity database construction will be described below.

(A) KANSEI DATABASE CONSTRUCTION The data to be searched is added to the sensitivity information in advance and placed in the sensitivity space (factor space). An SD adjective method based on adjective pairs is used as a means for giving sensitivity information to each target data. That is, by performing factor analysis on the evaluation parameters of the nine adjective pairs collected in the questionnaire format, a three-dimensional common factor that defines each adjective is derived to form a sensitivity factor space. Each data is arranged in this space using the following equation using the factor score (factor vector in the factor coordinate space).

[0037]

[Equation 1]

Here, the factor loading amount is a weight for each common factor of each of the above adjective pairs.

As an example of data collection, when the target data is an image, each image is presented one by one by a slide, and the impression of each image is entered on the questionnaire form with an adjective pair on an 11-point scale. Questionnaire forms are collected and shown in Figure 4.
Input to the database is performed using the input panel 1 shown in. In the input panel 1, the numerical value for each adjective pair can be adjusted using the scroll bar 1a. Moreover, it is also possible to collect data online using the input panel 1. In this case, the input value is immediately written in the database.

The adjective parameters collected in this way are transformed into a factor space. This state is shown in the flow chart 6.

That is, individual target data is presented (step S1), and the corresponding sensitivity is input using the adjective parameter (step S2). Then, it is determined whether or not all the data has been presented (step S3). If there is data that has not been presented, the process returns to step S1. If all the data has been presented, each image is displayed. Factor analysis of the adjective parameter of (step S
4).

That is, the adjective parameter of each image is converted into a position on the affective factor space using the factor loading amount.

Then, the coordinates of the position are calculated as a factor score and added to the image data (step S5).

(B) Kansei Engine The Kansei engine 3 (see FIG. 2) receives the parameter input by the adjective pair as the Kansei expressed by the user, converts this into the Kansei factor space, and the database in this factor space. Search for. The search takes the following two steps.

First, in the first step, the entire area of the factor space is targeted, and a certain metric space (subspace) whose center is a point (hereinafter referred to as the user's input point) corresponding to the sensitivity input value input by the user. ) Extract the data contained in. At this time, the Euclidean distance is used to calculate the distance, the size of the search target space depends on the number of data contained in the space, and the search is performed in order from the search target data that is closer to the user input point.

The sub-space containing the search target data extracted in the first step is composed of the search target data close to the user's designated sensitivity. Here, among the data contained in this space, The rank based on the calculated spatial distance is considered to include an error.

Therefore, the process of the second step is carried out in order to search the data contained in this space for data that is closer to the input sensitivity of the user.

When the user directly uses the sensitivity engine to search for data, it is possible to present a ranked candidate list as a search result and select an appropriate one from them. In such a case, it may be appropriate for the user to see all of the data and select the data without ranking the data as long as the data can be listed.

In the process of the second step, when the total number of data is so large that it cannot be listed even if the subspace is defined, or because the data is searched in batch processing, the data is ranked in advance. This is effective when it is necessary to do so.

In the second step, the searched data are ranked within the subspace. As the ranking method, the following three methods were used and changed according to the purpose.

(1) The first method is to rank the searched data using the Euclidean distance while keeping the values of all factor space coordinates.

(2) The second method is to obtain the variance of adjective input values by the person at the time of database construction for each data in the space, and set the one with the smaller variance as the upper candidate.

(3) The third method is to perform the factor analysis again only within the subspace.

(2) in the processing of the second step
The method is based on the fact that the sensitivity to certain data may differ considerably depending on the person. In the case of making a proposal as a search result, if the person presents data with a difference in the evaluation, the evaluation for the result also has a difference. Therefore, by presenting the data with the smallest variance, it is possible to expect the effect of extracting stable data.

The method (3) is intended to extract a stronger factor in the subspace, that is, a factor closer to the position of the adjective input value in the factor space.

(C) Kansei search Next, the procedure of the sensitivity search will be described.

FIG. 2 is a diagram conceptually showing the procedure of the sensitivity search.
FIG. 7 is a diagram showing the flow of this procedure.

(1) First, the user is asked to input the sensitivity by the adjective pair parameter using the adjective input panel 1 as shown in FIG. 4 (step S11).

(2) Next, the input parameters are converted into coordinates on the factor space, and images having coordinates close to the coordinates of the input parameters are sequentially extracted from the database to extract a candidate list. At this time, each method in the second step described above is performed in parallel to create each candidate list.

For example, when the factor analysis is performed again only within the subspace, the following processing is performed.

That is, the parameter input by the adjective pair is converted into the coordinates in the factor space by using the factor load amount (conversion matrix) calculated by the factor analysis in all factor spaces (step S12). Next, an image search is performed in all factor spaces to extract a sub-set consisting of a plurality of image candidates close to the sensitivity input value (step S13).

After that, a factor analysis is performed on the sub-set (step S14), and a factor load amount (conversion matrix) in the sub-set is obtained (step S15).

The inputted sensitivity information is converted into coordinates of a new sub-factor space (step S16), and in the sub-factor space,
An image search is performed (step S17).

(3) Then, the images in the candidate list having the highest score, that is, the position in the factor space of which the image is closer to the position of the inputted sensitivity in the factor space are displayed in order (step S18). .

FIG. 5 shows images to be searched, which are displayed in order from the highest score, and FIG. 8 shows a candidate list of images in order from the highest score.

(4) The user selects the one that he / she feels matches.

The above-mentioned data and rule formats,
The application examples are merely examples, and the present invention is not limited to these.

[0068]

The information retrieval device according to the present invention has the following advantages.

Conventionally, in order to obtain satisfactory data as a feeling from a database system having a large amount of data for a user, a system such as a keyword search could not be used. It is very effective that the sensitivity information can be searched by inputting a simple parameter.

At this time, in addition to global factor analysis, by performing factor analysis on a local data set, it is possible to deal with a finer sensitivity, which is also outstanding in databases and designs. Even a person who does not have knowledge has a great effect that the content search for creating a multimedia document can be easily performed.

[Brief description of drawings]

FIG. 1 is a diagram conceptually showing a system configuration of an information search device according to an embodiment of the present invention.

FIG. 2 is an image diagram conceptually showing a search in the information search device.

FIG. 3 is a block diagram showing a search function in the information search device.

FIG. 4 is a diagram showing an adjective input panel for performing sensitivity input and the like in the information search device.

FIG. 5 is a diagram showing a display panel on which search image candidates are displayed.

FIG. 6 is a diagram showing a flow of data collection and analysis for constructing a database in the information search device.

FIG. 7 is a diagram showing a flow of data search in the information search device.

FIG. 8 is a diagram showing an example of a display screen on which a data candidate list result is displayed.

[Explanation of symbols]

1 Input panel 1a scroll bar Two factor space 2a Subset 3 Kansei engine 21 Adjective input means 22 Input adjective conversion means 23 factor load storage 24 Input value storage 25 Distance calculation means 26 Database 27 Document File Extraction Means 28 Extracted document file presentation means 29 factor analysis means 100 information retrieval device

Continuation of front page (56) Reference JP-A-1-209528 (JP, A) JP-A-3-194655 (JP, A) JP-A 1-239631 (JP, A) JP-A-6-168277 (JP , A) JP-A 64-73460 (JP, A) JP-A 63-157229 (JP, A) JP-A 5-73625 (JP, A) Furugan Kenji Fujita Shuji Hata, 6-axis vision Image Retrieval Method Using Image Space, Proceedings of 17th Sensory Test Symposium, Japan, Japan Science and Technology Federation, September 17, 1987, pp. 47-52, Kuroda Akihiro et al., Adjective Pair Scale Classification of design images, Technical Report of Television Society, Japan, Television Society, May 29, 1992, VOL16 No. 31, pp. 19 to 24 (58) Fields investigated (Int. Cl. ⁷ , DB name) G06F 17/30 JISST file (JOIS)

Claims (2)

(57) [Claims] 1. An information retrieval device for retrieving multimedia data including character information, graphic information, image information, image information, and sound information, wherein information corresponding to human sensibilities is provided for each of a plurality of adjectives. Sensitivity input means for inputting as position information in a multidimensional adjective coordinate space having a dimension corresponding to, and position analysis in the adjective coordinate space attached to each multimedia data by factor analysis, and the adjective coordinates A factor coordinate space having a predetermined correspondence relationship with a space is set, and factor analysis means for making the respective multimedia data correspond to the coordinates of the factor coordinate space, and the inputted affective information is mapped on the factor coordinate space. Converting means, a position on the factor coordinate space corresponding to the input sensitivity information, and a position on the factor coordinate space of the multimedia data. A distance calculation means for calculating a distance between the multi-media data and the multi-media data within a predetermined distance from the position of the input feeling information on the main factor coordinate space obtained by the factor analysis of all the multimedia data. Obtaining the media data, creating a sub-data set consisting of the obtained multimedia data, of the multimedia data included in the sub-data set,
By performing a factor analysis of position information in the adjective coordinate space, a sub-factor space having a correspondence relationship with the adjective coordinate space different from the main factor coordinate space is set, and the input sensitivity in the sub-factor space is set. An information retrieval device configured to retrieve multimedia data based on information. 2. The information search device according to claim 1, wherein the process of converting the position information in the adjective coordinate space into the position information in the factor coordinate space is performed based on a factor load amount. The information retrieval device, wherein the load amount defines the relationship between the position in the adjective coordinate space and the position in the factor coordinate space.