JP2011048527A - Sensitivity information extraction device, sensitivity retrieval device and method, and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To perform retrieval wherein various aspects of a retrieval target such as quality, an external characteristic, and personality are considered when performing the information retrieval by use of sensitivity representation representing an image of the retrieval target. <P>SOLUTION: In this sensitivity information extraction device, the sensitivity representation is extracted from a set of inputted texts by a retrieval target information extraction part 31, the extracted sensitivity representation is connected to the retrieval target, and sensitivity information of each aspect to the retrieval target is generated by use of a sensitivity representation DB 1 storing at least the sensitivity information to the sensitivity representation and aspect information to which the sensitivity representation belongs with the connected retrieval target and sensitivity representation as input by a sensitivity information generation part 32 and is stored in a retrieval target DB2, so that sensitivity retrieval about the aspect to which the sensitivity representation inputted as a retrieval condition becomes possible, it is avoided that an image related to the aspect entirely different from the aspect becomes noise, and retrieval accuracy can be improved. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to a technique for extracting sensibility information from text to form a database and performing information retrieval using this database.

  At present, when searching for information such as a product on the Web, a search using a search query (search condition) based on the name, category, qualitative characteristics, and the like of the search target is the mainstream. Taking a car as an example, the search conditions are “sedan” of “Company T”, “4-door”, color “white”, etc. However, there is information that I want, but I can't describe the search conditions as above, but only with an image (impression) such as “cute” or “classy” or “feel” when using it. If the object cannot be characterized, it is convenient if such sensitivity can be directly searched as a search condition. However, in the Kansei search currently in practical use, it is necessary to manually add quantitative information (kansei information) representing the image of the target for each search target (for example, non-patent literature) 1).

“Amana 1.3 million photos can be searched with“ adjectives ”-Case database: ITpro”, [online], Nikkei Business Publications, Inc., [August 14, 2009 search], Internet <URL: http: // itpro. nikkeibp.co.jp/article/JIREI/20070827/280366/ ＞

  (1) Since it is troublesome to manually apply the sensitivity information for each search target in order to perform a sensitivity search, it is desirable to save the work as much as possible.

  (2) As a technique for realizing such a Kansei search at a low cost, a method of expressing an image to be searched by extracting Kansei expressions representing the image to be searched from text and adding them together can be considered. However, Kansei expression represents the image of various aspects (appearance features, quality, personality, etc.) that the search target has, and if all of them are added together, there is a possibility that the features will be dull. .

  For example, if there are `` gorgeous '' and `` bitter '' as sensibility expressions representing an image to be searched, the former expresses the image of the appearance and the latter expresses the aspect of taste, and by adding these, `` gorgeous '' The images such as “beautiful” and “flashy” are lost, and desirable search results cannot be obtained. Therefore, it is necessary to avoid the feature of different aspects from becoming noise when the sensitivity information to be searched is generated using the sensitivity expression.

  (3) When using emotional expressions as search conditions, the aspect that the user is paying attention depends on what kind of emotional expression is used. For example, when searching for “bitter”, attention is paid to the taste of the search object, so it is desirable to output a search object whose image is similar in terms of taste. Therefore, it is necessary to calculate the distance between the search condition and the search target in the aspect that the user pays attention to.

The present invention has been made in view of the above points.
(1) An image to be searched (= sensitivity information) is extracted from a large set of texts (corpus) in advance, and at the time of a sensitivity search, an image represented by a sensitivity expression input as a search condition and an image to be searched Providing an apparatus and a method for performing a sensitivity search with reduced operation costs by using the sensitivity distance of
(2) Considering what aspect of the search object (appearance features, quality, personality, etc.) represents the image of the search object when generating the sensitivity information of the search object using the sensitivity expression.
(3) In order to output a search target that is close to the image in the aspect that the user pays attention to, consider what aspect the emotional expression that is the search condition represents.
With the above three points, the object is to improve the search accuracy by avoiding the image of completely different aspects from becoming noise.

  Hereinafter, information obtained by quantifying an image included in a search target or Kansei expression is described as “Kansei information”.

  Regarding the above problem (1), in the present invention, when text related to a search target is input, text analysis is performed to extract a sensitivity expression, and the extracted sensitivity expression is used to generate sensitivity information for each search target. The present invention is characterized by presenting a search target that is sensibly close to the emotional expression input as a search condition.

  As for the problem (2), when the emotion information of the search target is generated using the emotion expression extracted from the text, the sensitivity information is generated for each aspect such as “appearance features” and “quality” of the search target. Thus, the feature of different aspects is to avoid noise.

  As for the problem (3), when a Kansei expression is input as a search condition, the distance between the Kansei expression that is the search condition and the search target is calculated in terms of “appearance features” and “quality” represented by the Kansei expression. Thus, it is characterized in that a search target similar in image is output in the aspect that the user pays attention to.

In detail,
A sensitivity information extraction device that extracts sensitivity information from a set of texts and creates a search target database,
A sensitivity expression database storing at least the sensitivity information for the sensitivity expression and the side information to which the sensitivity expression belongs;
A search object information extraction unit that extracts a Kansei expression from a set of input texts and links the extracted Kansei expression to a search object;
Kansei information extraction device comprising a Kansei information generation unit that receives the linked search target and Kansei expression as input, generates Kansei information for each side information with respect to the search target using the Kansei expression database, and stores the Kansei information in the search target database ,
And
A sensitivity search device that performs an information search using a search target database that stores at least sensitivity information for each side information for a search target,
A sensitivity expression database storing at least the sensitivity information for the sensitivity expression and the side information to which the sensitivity expression belongs;
A condition input unit for receiving search conditions;
A search unit that performs a search close to the search condition of the search target database using the received search condition and the sensitivity expression database,
A sensitivity search device comprising a result output unit for outputting the search results in a predetermined format;
Propose.

  According to the present invention, Kansei information about a search target is extracted from a text and made into a database, and the Kansei expression and the search target are calculated using this database, and information search is performed using the distance. By doing this, it is possible to perform a sensuous search that results can be obtained simply by inputting a sensibility expression, and there is no need to manually add metadata related to the target, thereby saving labor and reducing operational costs. Yes (solve the problem (1)).

  Further, according to the present invention, it is necessary to consider which aspect of the search object the sensitivity expression relates to both when generating the sensitivity information of the search object and when actually performing the search using the sensitivity expression. Thus, it is possible to avoid the feature of different aspects from becoming noise, leading to improvement of search accuracy (solving problems (2) and (3)).

The block diagram which shows the outline | summary of the Kansei information extraction apparatus and Kansei search apparatus of this invention Explanatory drawing showing an example of Kansei expression database Explanatory diagram showing an example of the search target database The block diagram which shows an example of embodiment of the sensitivity information extraction part Flow chart showing sensitivity information extraction processing in the sensitivity information extraction unit The block diagram which shows an example of embodiment of a sensitivity search part Flow chart showing Kansei Search Process in Kansei Search Unit Explanatory drawing which shows an example of the concrete process in the sensitivity information extraction part Explanatory drawing which shows the other example of the concrete process in the sensitivity information extraction part

  Hereinafter, the present invention will be described in detail with reference to the illustrated embodiments.

  FIG. 1 shows an outline of a sensitivity information extraction device and a sensitivity search device according to the present invention. The sensitivity information extraction device is composed of a sensitivity expression database (DB) 1, a search target database (DB) 2, and a sensitivity information extraction unit 3. The sensitivity search device includes a sensitivity expression DB 1, a search target DB 2, and a sensitivity search unit 4.

<Kansei expression database>
The sensitivity expression DB 1 is a database that stores at least sensitivity information for the sensitivity expression and side information to which the sensitivity expression belongs, and each record has at least a sensitivity expression ID, sensitivity information, and side information. As described later, the sensitivity expression DB 1 is used to add sensitivity information and side information to the sensitivity expression extracted from the text by the sensitivity information extraction unit 3 and the sensitivity expression input as a search condition by the sensitivity search unit 4. It is done.

  The sensitivity expression ID is an ID for uniquely identifying each sensitivity expression. For example, a standard form of the sensitivity expression can be used. At this time, if the same emotion expression ID is assigned to several emotion expressions, it is also possible to combine a plurality of images of emotion expressions and express them as one piece of sensitivity information.

  Sensitivity information represents an image of an emotional expression with some quantitative value. For example, a word vector created using co-occurrence information between the emotional expressions may be used.

  Aspect information is information that indicates the type of image represented by emotional expressions, such as “appearance features”, “appearance”, “personality”, etc., according to a predetermined classification system such as existing ontology, etc. At least one aspect information is given.

  FIG. 2 shows an example of a sensitivity expression database. In this example, a standard form of sensitivity expression is used as the sensitivity expression ID. In this example, the sensitivity information is a sensitivity vector having a predetermined value for each dimension axis (sensitivity axis) in a vector space having a predetermined number of dimensions. Here, the dimension number of the sensitivity vector is 7, the first dimension axis = beauty (the value is larger as it is more beautiful), the second dimension axis = silence,..., The seventh dimension axis = strength It shall be defined. Further, in this example, a category of a general noun semantic attribute system of a large Japanese vocabulary system (high quality, soft: “quality”, bright: “personality”, etc.) is used as side information. However, in the following description, a category name that is difficult to understand may be changed for convenience.

<Search target database>
As will be described later, the search target DB 2 is a database that stores at least sensitivity information for each side information for the search target that is output from the sensitivity information extraction unit 3, and each record includes a search target ID and at least one side information. And sensitivity information for each side information. As will be described later, the search target DB 2 is referred to together with the sensitivity expression DB 1 when searching in the sensitivity search unit 4 and is used to obtain a search result.

  The search target ID is an ID for uniquely identifying each search target. For example, a standard form of the search target can be used. The side information is information similar to that used in the emotional expression DB 1 that indicates the type of image that the search target has. Sensitivity information is an image of a search target represented by some quantitative value for each side information, and uses the same expression format as the sensitivity expression DB1.

  FIG. 3 shows an example of the search target database 2. In this example, the standard form of the search target is used as the search target ID. In this example, the same information as the case of the sensitivity expression DB 1 in FIG. 2 is used as the side information. Further, in this example, a sensitivity vector having the same number of dimensions and axes as the sensitivity information of the sensitivity expression DB 1 in FIG. 2 is used as sensitivity information.

  Taking “American short hair” in FIG. 3 as an example, the sensibility information is generated for each of the four aspect information of [quality] [appearance features] [good faith] [personality]. To avoid mixing the images. The emotional expression written after the side information in FIG. 3 is an example of the emotional expression belonging to each side information.

<Sensitivity information extraction unit 3>
The sensibility information extraction unit 3 receives a set of texts that are directly input from an input unit (not shown), read from a storage unit, or input from another device or the like via a communication medium. Sensitivity information is generated and output to the search target DB 2 and includes a search target information extraction unit 31 and a sensitivity information generation unit 32 as shown in FIG.

  The search target information extraction unit 31 extracts a sensitivity expression from the set of input texts, and outputs the extracted sensitivity expression in association with the search target.

  The sensibility information generation unit 32 receives the search object and the sensibility expression linked by the sensibility expression extraction unit 31, generates the sensibility information for each side information for each search object using the sensitivity expression DB 1, and stores it in the search object DB 2. Store.

  FIG. 5 shows the flow of the sensibility information extraction process in the sensibility expression extraction unit 3, and its operation will be described in detail below.

  (Step S1) The search target information extraction unit 31 extracts a sentiment expression representing the search target image from the set of input texts, and associates the extracted sentiment expression with the search target according to a predetermined method. Proceed to

  Kansei expression is an expression such as “quiet”, “cool”, “solid”, “cool”, etc. that expresses the impression and feeling of a certain object.

  Kansei expressions are extracted from texts after performing language analysis processing including well-known morphological analysis processing (generating word information) and then using rules that use word information such as notation, parts of speech, readings, word boundaries, etc. It is possible to use existing techniques, for example, the processing of the evaluation expression extraction unit described in Japanese Patent Application Laid-Open No. 2008-140359 (“Evaluation Information Extraction Device, Evaluation Information Extraction Method and Program”).

  In addition, as a method of linking the extracted emotional expression to the search target, if a text that is clearly written only for a specific search target (for example, a picture book) can be used, the emotional expression that appears in the text can be used. Everything can be tied to that particular search target. Even if such text is not available, if there is a name list (target word list) to be searched, for each search target in the list, co-occurrence information and dependency between the search target and the emotional expression in the text In addition to the method of using the relationship information, the same processing can be performed by using a relationship extraction technique (for example, Japanese Patent Application Laid-Open No. 2008-225565 “Correlated Expression Combination Extraction Device and Method”). Any method can be selected.

  (Step S2) In the sensitivity information generation unit 32, from the search target created by the search target information extraction unit 31 and the emotional expression associated therewith, the sensitivity information of the search target is generated for each side information of each emotional expression, The result is stored in the search target DB2.

  Sensitivity information to be searched is obtained from the sensitivity expression DB 1 for each of the sensitivity expressions associated with the search object, and for each sensitivity information having the same side information (for each sensitivity vector constituting the sensitivity information). Sensitivity information of the side information is calculated by taking the average of the values, and this is performed for all side information units. When taking the average, weighting may be performed according to the co-occurrence frequency of the search target and the emotional expression.

<Sensitivity search unit 4>
The sensibility search unit 4 inputs search conditions inputted directly from an input means (not shown), read from a storage means, or inputted from another device or the like via a communication medium, and outputs a search result. As shown in FIG. 6, it comprises a condition input unit 41, a search unit 42 and a result output unit 43.

  The condition input unit 41 receives and interprets the input search condition.

  The search unit 42 searches the search target DB 2 using the search condition and the emotional expression DB 1, here, the search target close to the search condition.

  The result output unit 43 outputs the search result to a display screen of the display device in a predetermined format.

  FIG. 7 shows the flow of the sensibility search process in the sensibility search unit 4, and the operation will be described in detail below.

  (Step S11) In the condition input unit 41, at least one sensitivity expression is acquired as a search condition, and the process proceeds to S12. At this time, in addition to the search condition, a presentation method (display only the top X items, display a search result whose distance is within a set threshold value, or the like) may be input.

  Various methods are conceivable as the search condition input format. For example, it is possible to input a plurality of emotional expressions. In addition, in order to perform a sensitivity search for a specific application, a part of the search conditions is fixed in advance (for example, one of the emotional expressions is designated as “light”), and the search conditions can be arbitrarily input. Alternatively, the user may be input using a graphical user interface or the like.

  Hereinafter, the search condition will be described as one word. If there are two or more words, the side information about all the emotional expressions is obtained in (S14) and the logical product is obtained. It is assumed that the sensitivity information of the search condition is created by adding the sensitivity information for each aspect information in the same manner as in S2) and taking the average.

  (Step S12) In the search unit 42, it is checked whether or not the sentiment expression as the input search condition is registered in the sentiment expression DB1, and if it is registered, the process proceeds to S14, and if not registered, the process proceeds to S13.

  (Step S <b> 13) The result output unit 43 displays a message indicating that there is no result because the sensibility search cannot be performed. Alternatively, the results of a web keyword search may be displayed as an alternative.

  (Step S14) In the search unit 42, the sensitivity information X and the side information Y corresponding to the sensitivity expression input as the search condition are extracted from the sensitivity expression DB 1, and the process proceeds to S15.

  (Step S15) A record including the side information Y is searched from the search target DB 2, the distance between the sensitivity information of the record and the sensitivity information X is calculated, output to the result output unit 43, and the process proceeds to S16.

  When the sensitivity information is expressed as a vector as shown in FIG. 2 and FIG. 3, the distance calculation for determining whether it is close to the sensitivity is the sensitivity vector representing the sensitivity information X corresponding to the sensitivity expression input as a search condition. And the sensitivity vector representing the sensitivity information in the search target DB 2 can be performed using an arbitrary distance scale such as Euclidean distance and cosine similarity.

  (Step S16) In the result output unit 43, search target IDs and other information are arranged in the order of records having a short distance and output to a screen or the like, and the process ends.

  Hereinafter, a specific processing example in the above-described embodiment will be described in detail by taking as an example a case of searching for a pet. Note that the cosine distance is used for the calculation of the sensibility distance between the search condition in the search unit 42 and the search target registered in the search target DB 2. The text used in the search target information extraction unit 31 will be described for both (1) a case where only the search target is clearly written and (2) a case where it is not.

<Processing performed by the sensitivity information extraction unit 3>
(1) When the text is written about the search target Search for realizing a pet search (search for pets from the input emotional expression) when all the Kansei expressions appearing in the text can be linked to the search target An example of how to create a target database is as follows.

  (Step S1) FIG. 8A shows an example of input to the search target information extraction unit 31 at a certain point in time (when document ID = 100). (A) A text document is an example of text having a pet name and a document ID as document meta information. An example of the text analysis result is a part excluding the underline in FIG. 8B, and the result of extracting the emotional expression from the text is the underline part in FIG. 8B.

  The emotional expression extracted from the text is held in association with the pet name in the meta information. When there are a plurality of texts related to the same search target, the emotional expressions extracted from each text are collected for each search target. An example of the extracted contents is shown in FIG. When the search target is “American short hair” as in the example of FIG. 8, emotional expressions such as “strong” and “easy” are extracted from texts other than document ID = 100, and (c) This is reflected in the total number of extractions. Proceed to S2.

  (Step S2) In the sensitivity information generation unit 32, the sensitivity information and the side information corresponding to the sensitivity expression extracted in S1 are obtained from the sensitivity expression DB1, and the sensitivity information is generated for each side information. As a result of extracting emotional expressions from all texts, if all the records whose search target is “American short hair” are only the four records in FIG. 8C, the sensitivity of the search target “American short hair” The vector obtains sensibility information and side information corresponding to each of the sensibility expressions “fluffy”, “powerful”, “gassy”, and “easy” from the sensibility expression DB 1 (FIG. 2), and the sensibility information of the sensibility expression regarding the same aspect Calculate by taking the average and adding. At this time, it is also possible to weight by the number of extractions in (c).

  If the plump side is [Quality] [Appearance Features], the “Strong” and “Something” sides are [Appearance Features], and the “Friendly” side is [Goodwill] [Personality], then “American” “Short hair” generates sensibility information for a total of four aspects of [quality] [appearance features] [good intentions] [personality] (search ID = “American short hair” in FIG. 3) (See Kansei expression in the side information column, which describes the extracted Kansei expressions that belong to the side information in the same column.)

  Similarly, sensibility information is obtained for all other search targets, stored in the search target DB 2, and the process ends. FIG. 3 shows the search target DB 2 that includes sensitivity information for each aspect obtained by the above method together with the search target.

(2) When using other texts When the text written about the search target is not available, a target word list is required.

  (Step S1) FIG. 9A-1 shows an example of input to the search target information extraction unit 31 at a certain point in time (when document ID = 300). (A-1) A text document is an example of text having only a document ID as document meta information. The target word list is shown in FIG. This is a list of notation of pets to be searched such as “American short hair”, “Singapura”, etc., and at least has a search target ID which is a standard form of search target. It holds notation fluctuation information for absorbing the notation fluctuation.

  The example of the text analysis result is a part excluding the underline in FIG. 9B, and the result of extracting the emotional expression from the text is the underline part in FIG. 9B. FIG. 9C shows an example of an extraction result created by linking Kansei expressions included in two sentences before and after a sentence in which the search target of the input target word list or its notation fluctuation appears in the text. Shown (in this example, the target word being processed is “American short hair”). Proceed to S2.

  Since the processing after (Step S2) is the same as the case of (1), it is omitted.

<Processing performed by the sensitivity search unit 4>
A search example of the above-described search target database (FIG. 3) for pet search will be described.

  (Step S11) Assume that the following four patterns of search conditions are input in the condition input unit (Note that the side information such as [measurement] [appearance, appearance features, personality] is described for convenience of explanation, Is not entered in.)

<Input 1> Small [Measurement and Measure]
<Input 2> Fashionable [Apparel, appearance features, personality]
<Input 3> Fashionable [Appearance, Appearance Features, Personality], Glamorous [Appearance Features]
<Input 4> Exotic Feeling <Input 1> (Step S12) Since the search condition “small” is registered in the emotional expression DB1, the process proceeds to S14.

  (Step S14) “Small” sensitivity information and side information [metrics] are obtained from the sensitivity expression DB1.

  (Step S15) A distance calculation is performed between the sensitivity information obtained in S14 and the sensitivity information of the record whose side is [metrics] in the search target DB2, and the process proceeds to S16.

  (Step S16) The search objects are output in the order from the shortest distance in S15. Assuming that the records in the search target DB 2 are only “American short hair”, “Singapura”, “Sham”, “Persia” in FIG. 3, “Singapura” , “Siam” and “Persian” (“American short hair” is not included in the search results because there is no sensitivity information with the [Measurement] aspect). The three search results are output to the screen and the process ends.

In the case of <Input 2> (Step S12) Since the search condition “fashionable” is registered in the emotional expression DB1, the process proceeds to S14.

  (Step S14) The “fashionable” sensitivity information and side information are obtained from the sensitivity expression DB1. At this time, the side information is [appearance, appearance features, personality].

  (Step S15) A distance calculation is performed between the sensitivity information obtained in S14 and the sensitivity information of the record corresponding to any one of the three aspects “Appearance, Appearance Features, Personality” in the search target DB2, and the process goes to S16 move on. When there are two or more corresponding side faces in one search target, the distance to the side face with the closest distance is used in the next step.

  (Step S16) The search objects are output in the order from the shortest distance in S15. Assuming that the records in the search target DB 2 are only “American short hair”, “Singapura”, “Siam”, “Persia” in FIG. 3, “Siam” , "Persia", "Singapura", "American short hair". Output the result to the screen and exit.

In the case of <Input 3> (Step S12) The search conditions are two words “fashionable” and “gorgeous”, both of which are registered in the emotional expression DB1, and therefore the process proceeds to the next step. If there are a plurality of word inputs and there is a word that is not registered in the emotional expression DB 1, only the information of the registered word is used to perform the subsequent steps.

  (Step S14) Sensitivity information and side information of both words “fashionable” and “gorgeous” are obtained from the sensitivity expression DB1. First, the side information is obtained by performing a logical product of the side information [dressing, appearance features, personality] and [appearance features] of each word, and only “appearance features” are obtained. Then, the sensitivity information of the search condition is calculated by taking the average of the sensitivity information whose side information of both words is [appearance feature]. Proceed to S15.

  (Step S15) Distance calculation is performed between the sensitivity information obtained in S14 and the sensitivity information of the record whose side surface is [appearance feature] in the search target DB2, and the process proceeds to S16.

  (Step S16) The search objects are output in the order from the shortest distance in S15. Assuming that the records in the search target DB2 are only “American short hair”, “Singapura”, “Siam”, and “Persia” in FIG. "Persia", "Siam", "American short hair", "Singapura". Output the result to the screen and exit.

In the case of <Input 4> (Step S12) Since the search condition is “exotic feeling overflowing” and there is no registration in the emotional expression DB1, the process proceeds to S13.

  (Step S13) Since Kansei search cannot be performed, a message indicating that there is no result, for example, “Unfortunately Kansei search could not be performed. Normal search results are displayed.” Is output to the screen and the process is terminated.

  The present invention is shown in a program for realizing the functions shown in the configuration diagrams of FIGS. 1, 4 and 6 or a flowchart of FIGS. 5 and 7 via a medium or communication line on a known computer. It can also be realized by installing a program having a procedure.

  1: Sensitivity expression database (DB) 2: Search object database (DB) 3: Sensitivity information extraction unit 4: Sensitivity search unit 31: Search object information extraction unit 32: Sensitivity information generation unit 41: Condition input Part, 42: search part, 43: result output part.

Claims (10)

A sensitivity information extraction device that extracts sensitivity information from a set of texts and creates a search target database,
A search object information extraction unit that extracts a Kansei expression from a set of input texts and links the extracted Kansei expression to a search object;
Using the linked search target and the emotional expression as input, and using the emotional expression database that stores at least the emotional information for the emotional expression and the side information to which the emotional expression belongs, the emotional information for each aspect information for the search target is generated and searched. A sensibility information extraction device comprising a sensibility information generation unit for storing in a target database. In the search target information extraction unit,
The sensibility according to claim 1, wherein when the input text describes only a specific search object, all of the sensibility expressions appearing in the text are linked to the specific search object. Information extraction device. In the search target information extraction unit,
If the input text does not specify the search target, the emotional expression that appears in the text is shared with each search target in the predetermined search target list. The sensibility information extraction device according to claim 1, wherein the information is linked using starting information and dependency relationship information. In the Kansei information generator,
Kansei information and side information of each Kansei expression linked to the search target is obtained from the Kansei expression database, and the sensitivity information of the side information is generated by averaging the values for each Kansei information with the same side information. The sensibility information extraction device according to claim 1. A sensitivity search device that performs information search using a search target database that stores at least sensitivity information for each side information for a search target,
A condition input unit for receiving search conditions;
A search unit that performs a search that is sensibly close to the search condition in the search target database using the received search condition and a sensitivity expression database that stores at least the sensitivity information for the sensitivity expression and the side information to which the sensitivity expression belongs. When,
And a result output unit for outputting the search result in a predetermined format. The search part
Sensitivity information and aspect information corresponding to the sensitivity expression input as a search condition are extracted from the sensitivity expression database, a record including the same aspect information is searched from the search target database, and the sensitivity information of the record and the search condition are input. The sensibility search device according to claim 5, wherein a distance between the sensibility information corresponding to the sensibility expression is calculated. When there are multiple Kansei expressions input as search conditions, Kansei information and side information corresponding to each Kansei expression are extracted from the Kansei expression database, and after calculating the logical product of the side information, the side information is The sensitivity search apparatus according to claim 6, wherein sensitivity information as a search condition is created by taking an average of values. A method for extracting Kansei information by extracting Kansei information from a set of texts and creating a search target database,
A search target information extraction unit extracts a Kansei expression from a set of input text, and links the extracted Kansei expression to the search target;
Sensitivity information generation unit receives the linked search object and emotional expression as input, and uses the emotional expression database that stores at least the emotional information for the emotional expression and the aspect information to which the emotional expression belongs to And a method of generating information and storing it in a search target database. A sensitivity search method for performing an information search using a search target database that stores at least sensitivity information for each side information for a search target,
A condition input unit accepting a search condition;
The search unit uses the received search condition and the emotional expression database storing at least the sensitivity information for the emotional expression and the side information to which the emotional expression belongs to perform a search close to the search condition in the search target database. The steps of
A result output unit including the step of outputting the search result in a predetermined format.   The program for making a computer perform each process step of the method of Claim 8 or 9.

Images