JP2017208047A - Information search method, information search apparatus, and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an information search apparatus capable of allowing a user to search information matching an information request without requesting the user to consider on a search query.SOLUTION: An information search apparatus 100 performs a series of processing including: extracting feature quantity of each of question sentences based on contents of plural question sentences which are written in a natural language included in question collection; calculating display coordinates for displaying question sentences on a screen based on the feature quantity extracted from the question sentences; displaying each of question sentences on the screen according to the display coordinates; when receiving specified coordinates on the screen, calculating the distance between the display coordinates of the respective question sentences and the specified coordinates; generating a search query based on a keyword which is included in a question sentence the calculated distance of which satisfies a predetermined condition; and inputting the generated search query to a search engine to display a search result relevant to the search query.SELECTED DRAWING: Figure 3

Description

  The present invention relates to an information search method, an information search device, and a program for text.

  In an information search system represented by a Web search, for example, keyword matching or the like is performed on a keyword set (hereinafter referred to as “search query”) entered by a user in a search window to search for a document that matches the search query. ing. However, in the case of search processing using keyword matching, the keyword of the search query must match the keyword in the document completely, and there is a problem that the recall rate (Recall) of the search decreases. There are things to do. Thus, there has been disclosed a query expansion technique or the like that expands and increases keywords included in a search query and matches a search query input by a user in a search window with a wider range of documents (Patent Documents 1 and 2).

  In addition to methods for improving retrieval recall, research is also being conducted on methods for improving information retrieval interfaces and performing information retrieval with simpler operations. For example, a method for facilitating visually searching for information using a two-dimensional plane is disclosed (Non-Patent Document 1).

JP 2010-123036 A JP2013-45182A

Daisaku Hayashi, Tomohiro Fukuhara, Tetsuji Sato, “Proposal and Evaluation of Interactive Search Interface to Support Information Arrangement”, 4th Forum on Data Engineering and Information Management DEIM Forum 2012 Proceedings, IEICE, E7-1, 2012

  As described above, in an information search system represented by Web search, for example, a user acquires desired information by inputting a search query in a search window.

  However, the act of inputting a search query in the search window in order to obtain desired information is often difficult for a user who is unfamiliar with information search.

  First, when performing an information search, the user needs to clearly define an information request that he wants to know. If the information request is vague or you do not know what to look for, the user may not know how to create a search query. Second, even if the user clearly understands the information request, it is difficult to create a search query for obtaining a search result that satisfies the information request.

  The present invention has been made in view of the above circumstances, and an information search method, an information search apparatus, and a program that enable a user to search for information that meets an information request without considering a search query. The purpose is to provide.

  In order to achieve the above object, the information search method according to the present invention extracts each feature amount of the question sentence based on the contents of a plurality of question sentences described in a natural language included in the question collection, For each of the question sentences, based on the feature amount extracted from the question sentence, a display coordinate when displaying the question sentence on the screen is calculated, and each of the question sentences is calculated on the screen. When the designated coordinates on the screen specified by the user operation are received, the display coordinates calculated for each of the question sentences and the distance between the designated coordinates are calculated, Based on a keyword included in the question sentence in which the calculated distance satisfies a predetermined condition, a search query is generated, the generated search query is input to a search engine, and searched by the search engine. Displaying search results for the search query.

  The information search apparatus according to the present invention includes a feature amount extraction unit that extracts each feature amount of the question sentence based on the contents of a plurality of question sentences described in a natural language included in the question collection, and the question sentence A coordinate calculation unit that calculates display coordinates when displaying the question sentence on the screen based on the feature amount extracted from the question sentence by the feature amount extraction unit, and each of the question sentences. When receiving the display coordinates displayed on the display coordinates calculated by the coordinate calculation section and the specified coordinates specified by the user operation on the screen, the coordinate calculation is performed for each of the question sentences. A distance calculation unit that calculates a distance between the display coordinates calculated by the unit and the designated coordinates, and a keyword included in the question sentence that satisfies a predetermined condition for the distance calculated by the distance calculation unit, Search query And an information search unit that inputs a search query generated by the query generation unit to a search engine and obtains a search result for the search query searched by the search engine. The display unit displays a search result for the search query acquired by the information search unit on the screen.

  A program for an information search method according to the present invention causes a computer to execute each step of the information search method.

  As described above, according to the information search method, information search apparatus, and program of the present invention, there is an effect that the user can search for information that meets the information request without considering the search query. can get.

It is a figure which shows the structural example of a search screen. It is a figure which shows the example of a display of a search screen. It is the schematic which shows the structural example of an information search device. It is a flowchart which shows an example of the process of the coordinate information creation process routine in an information search device. It is a flowchart which shows an example of the process of the information search process routine in an information search device.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. In the following description, the same reference numerals are given to the components or processes having the same functions throughout the drawings, and the repeated description is omitted as appropriate.

  FIG. 1 shows a configuration example of a search screen 50 of the information search apparatus 100 according to this embodiment. The search screen 50 includes a category selection area 51, a question display area 52, and a search result display area 53. The search screen 50 displays various types of information handled by the information search apparatus 100, and for example, designates a designated position of a mouse operated by a user with a pointer. It functions as a search interface that displays and accepts the position where the user clicked the mouse.

  The category selection area 51 is an area for switching the question display area 52 for each category. For example, a tab is arranged for each category. In the example of FIG. 1, three tabs 51 </ b> A to 51 </ b> C are arranged in the category selection area 51, and when the user clicks one of the tabs with the mouse, a question sentence regarding the category associated with the clicked tab is displayed. It is displayed in the question display area 52.

  Here, the category is, for example, a classification when a plurality of items are classified based on a certain standard, and in this embodiment, a plurality of question sentences are classified for each category as will be described later.

  The question display area 52 is an area in which a question text related to the category selected in the category selection area 51 is displayed. When the user clicks a region in the vicinity of the question display area 52 where a question sentence that seems to indicate the content close to the information request held by the user is displayed, the search result display area 53 The search result for the question sentence corresponding to the area clicked with the mouse is displayed.

  For example, as shown in FIG. 2, a case where three categories “PC”, “baseball”, and “music” are prepared in the category selection area 51 and the category “PC” is selected by the user will be described.

  In this case, in the question display area 52, a question text relating to a PC, that is, a personal computer is displayed. The user moves the mouse pointer 54 to an area where a question sentence that seems to indicate content close to the information request held by the user is displayed, and clicks the mouse. In addition, when the question sentence which shows the content close | similar to the information request | requirement which he owns is not displayed on the question display area 52, the user is the vicinity of the question sentence which is comparatively similar to the information request | requirement which he owns. Move the mouse pointer 54 to the area and click the mouse.

  For example, when a user has an information request that “a notebook computer used for work should be light, if possible,” vaguely, a question sentence indicating the information request itself exists in the question display area 52 shown in FIG. do not do. However, in the question display area 52 shown in FIG. 2, a question message “notebook is heavy” and a question message “I want a notebook that can be used at work” related to the user's information request are displayed. Moves the pointer 54 to an area in the vicinity where these question sentences are displayed, and clicks the mouse. In this case, the search result display area 53 displays search results mainly related to two question sentences, a question sentence “notebook is heavy” and a question sentence “I want a notebook that can be used at work”. Will be.

  In addition, as shown in FIG. 2, each question sentence displayed in the question display area 52 may be displayed so that the size of a character differs according to the importance of the question sentence.

  Hereinafter, for example, the information search apparatus 100 that displays the search screen 50 as shown in FIG. 2 and displays the search result based on the question sentence corresponding to the area of the question display area 52 clicked by the user in the search result display area 53 will be described. To do.

<System configuration>
FIG. 3 is a diagram illustrating a system configuration example of the information search apparatus 100 according to the present embodiment. The information retrieval apparatus 100 is composed of a computer including a CPU, a RAM, and a ROM that stores a program for executing a coordinate information creation processing routine and an information retrieval processing routine, which will be described later. It is configured as shown.

  The information search device 100 includes an input unit 10, a calculation unit 20, a storage unit 30, and a display unit 40. The calculation unit 20 includes a coordinate information creation unit 20A and an information search unit 20B.

  The input unit 10 of the information search apparatus 100 having the above configuration is pointed by a question collection including a plurality of question sentences written in, for example, a natural language, and a pointer 54 when the user clicks the mouse. The coordinates on the search screen 50, that is, “designated coordinates” are input.

  The question collection may be any text set including a question sentence written in a natural language, and there are no restrictions on the contents of the question sentence and the method of collecting the question collection. For example, a question sentence posted on a Web question answering site can be used, or a question sentence described in a frequently asked question (FAQ) such as a home page can be used. In this case, the question sentences included in the question collection may be processed within a range not departing from the spirit of the question sentences. For example, when the question sentence is a long sentence, or when the expression of the question sentence is different from a commonly used expression, processing such as removal of unnecessary clauses or words or replacement of words may be performed. . There is no particular limitation on the method for processing the question sentence. For example, the question sentence may be automatically processed based on the syntax analysis result of the question sentence, or the question sentence may be processed by the user.

  When receiving the question collection, the input unit 10 notifies the coordinate information creation unit 20A of the accepted question collection.

  Based on the question collection received from the input unit 10, the coordinate information creation unit 20 </ b> A displays the display coordinates of each question sentence included in the question collection in the question display area 52 and the question sentence in the question display area 52. Then, a question sentence score that defines the character size of each question sentence is created.

  Therefore, the coordinate information creation unit 20A includes a category division unit 21, a feature amount extraction unit 22, and a coordinate calculation unit 23.

  The category division unit 21 divides each question sentence included in the question collection for each category based on the contents of each question sentence included in the question collection received from the input unit 10.

  In addition, there is no restriction | limiting in the method of dividing | segmenting a question sentence into a category, A well-known category division method can be used. For example, the category information previously assigned to the question sentence at the collection source of the question collection may be used as it is, or a known clustering algorithm such as a K-average algorithm or Ward method is applied to the question collection, May be divided into categories. However, the category division unit 21 does not necessarily divide each question sentence into categories. For example, when the number of collected question sentences is relatively small or when the question sentences are concentrated on one topic, it is assumed that each question sentence is included in one category in advance, and the subsequent processing Execute.

  For each category, the feature amount extraction unit 22 extracts a feature amount from each question sentence based on each question sentence divided into the categories by the category division unit 21.

  In addition, there is no restriction | limiting in the extraction method of the feature-value from a question sentence, A well-known feature-value extraction method can be used. For example, morphological analysis is performed on a question sentence, the question sentence is divided into morphemes, and the morpheme occurrence distribution in the question sentence is expressed as a feature vector based on the bag-of-words format as the feature amount of the question sentence. Good. In addition, a semantic vector of each morpheme included in the question sentence is created using a neural network, and a synthesized vector obtained by synthesizing the semantic vectors of all the morphemes that appear in the question sentence by processing such as centroid calculation is used. It is good.

  For each category, the coordinate calculation unit 23 defines the display position of each question sentence divided into the category in the question display area 52 based on the feature amount extracted from each question sentence by the feature amount extraction unit 22. Calculate the display coordinates.

  Since the question display area 52 is an area represented in two dimensions, the coordinate calculation unit 23 performs a process of compressing the number of dimensions of the feature amount extracted from the question sentence by the feature amount extraction unit 22 into two dimensions.

  Specifically, when the feature amount extracted from each question sentence is expressed by a vector of three or more dimensions, the coordinate calculation unit 23 applies a vector dimension compression algorithm to the feature amount.

  In addition, there is no restriction | limiting in the dimension compression method with respect to a feature-value, A well-known dimension compression method can be used. For example, the coordinates of the question text in the question display area 52 can be obtained by normalizing the coordinates of the vector obtained by directly compressing the feature quantity into two dimensions using the singular value decomposition and the topic model. May be calculated. In addition, the degree of similarity of the corresponding question sentence is calculated from the feature amount using a multi-dimensional scaling method (Multi Dimensional Scaling: MDS), and a question sentence having a higher degree of similarity is displayed closer to the question display area 52. The question text display coordinates in the question display area 52 may be calculated.

  In addition to the display coordinates of the question sentence in the question display area 52, the coordinate calculation unit 23 defines a character size of the question sentence displayed in the question display area 52 according to the importance of the question sentence. That is, the question sentence score may be calculated.

  There is no limitation on the method of calculating the importance level of the question sentence, and a known calculation method can be used. For example, the coordinate calculation unit 23 applies a clustering algorithm such as Affinity Propagation to a data set whose elements are display coordinates of each question sentence in the question display area 52, and divides each display coordinate into a plurality of clusters. . At that time, the coordinate calculation unit 23 sets a characteristic question sentence representing each cluster, that is, a question sentence at the center of each cluster for each cluster. Then, the coordinate calculation unit 23 adds a predetermined value to the question sentence score of the set question sentence at the center of the cluster, and determines the character size of the question sentence at the center of the cluster as shown in FIG. You may make it display larger than another question sentence.

  Further, for example, the coordinate calculation unit 23 may extract a keyword from each question sentence according to a predetermined keyword extraction rule and calculate a question sentence score. In addition, there is no restriction | limiting in particular in the keyword extraction rule used in the coordinate calculation part 23, It can prescribe | regulate freely. In general, it is preferable to extract morphemes, which are independent words or content words such as nouns and verbs, from keywords as keywords.

  In this case, the coordinate calculation unit 23 applies a weight calculation method such as Term Frequency-Inverse Document Frequency (TF-IDF) method or BM25 method to calculate the weight of the keyword included in each question sentence. Then, the coordinate calculation unit 23 calculates the weight of each question sentence as a question sentence score based on the calculated keyword weight. The coordinate calculation unit 23 may consider that the question sentence is more important as the question sentence score is larger, and the question sentence having a larger question sentence score may be displayed in a larger size.

  The coordinate calculation unit 23 associates each question sentence with a question sentence ID that uniquely identifies the question sentence, associates the question sentence ID, the display coordinates, and the question sentence score with each question sentence, and associates the question sentence Each of the data sets having the ID, the question sentence, the display coordinates, and the question sentence score as a unit is stored for each category in the question collection coordinate group DB 30A provided in the storage unit 30 in advance.

  When the coordinate calculation unit 23 does not calculate the question sentence score, the coordinate calculation unit 23 assigns each of the data sets having the question sentence ID, the question sentence, and the display coordinates as a category to the question collection coordinate group DB 30A. Store every time. When the coordinate information creation unit 20A divides the question sentence into morphemes, the coordinate calculation unit 23 may include information on the morpheme included in the question sentence and the part of speech of each morpheme in the data set. Further, the coordinate calculation unit 23 may associate each category with a name (category name) representing the content of the category and store it in the question collection coordinate group DB 30A.

  When the user starts an information search with the information search device 100, the display unit 40 acquires each of the data sets for the category selected in the category selection area 51 from the question collection coordinate group DB 30A, and displays the question display area. 52. In this case, the display unit 40 displays a question sentence with a character size corresponding to the question sentence score at a position corresponding to the display coordinates of the question display area 52 for each data set corresponding to the selected category.

  The display unit 40 may display the question text in the question display area 52 by appropriately scaling the display coordinates according to the size of the question display area 52.

  On the other hand, when the user moves the mouse pointer 54 to the vicinity of the question sentence that seems to indicate the content close to the information request held by the user and clicks the mouse, the input unit 10 is designated by clicking the mouse. Coordinates corresponding to the selected position, that is, “designated coordinates” are input.

  When receiving the designated coordinates, the input unit 10 notifies the information retrieval unit 20B of the received designated coordinates.

  The information search unit 20B generates a search query based on the designated coordinates received from the input unit 10 and the display coordinates of the question text stored in the question collection coordinate group DB 30A, and performs a search with a search engine using the generated search query. The search results are output.

  Therefore, the information search unit 20B includes a distance calculation unit 24, a query generation unit 25, and an information search unit 26.

  The distance calculation unit 24 is included in the designated coordinates and the category selected in the category selection area 51 based on the designated coordinates received from the input unit 10 and the display coordinates of the question text stored in the question collection coordinate group DB 30A. The distance from the display coordinates of the question sentence is calculated for each question sentence. Then, the distance calculation unit 24 associates the question text included in the category selected in the category selection area 51 with the calculated distance. For example, the questions in which the distances are associated in ascending order of the associated distances. Each sentence is output to the query generator 25.

  In addition, there is no restriction | limiting in the calculation method of the distance of a designated coordinate and the display coordinate of a question sentence, For example, well-known distance calculation algorithms, such as a Eugrid distance or a cosine distance, can be used. Further, the calculated distance may be associated with the question sentence ID instead of the question sentence, and may be output to the query generation unit 25.

  Based on the question sentence and distance received from the distance calculation part 24, the query generation part 25 extracts keywords from the question sentence that the associated distance satisfies a predetermined condition, and generates a search query.

  A plurality of search query generation methods are implemented in advance in the query generation unit 25, and the query generation unit 25 determines a search query generation method to be applied, for example, according to the content of the category selected in the category selection region 51. be able to.

  For example, as a first search query generation method, the query generation unit 25 extracts a keyword from a question sentence corresponding to a display coordinate having the shortest distance from a designated coordinate, and uses each extracted keyword as a search query. Apply.

  In addition, there is no restriction | limiting in the keyword extraction rule used in the query production | generation part 25, It can prescribe | regulate freely. Generally, a method is used in which morphemes that are independent words or content words such as nouns and verbs are extracted from a question sentence as keywords. In this case, for example, when the question text has already been divided into morphemes by the coordinate information creation unit 20A, the division result can be used.

  In addition, as a second search query generation method, the query generation unit 25 includes N pieces (N = N = in order from the shortest distance from the corresponding display coordinate to the designated coordinate among the question sentences received from the distance calculation unit 24. (1, 2, 3,...) Is selected, and a keyword extracted from each of the selected question sentences is used as a search query.

  Note that the value of N may be a predetermined value or a variable value. For example, by selecting all the question sentences corresponding to each of the display coordinates included within a predetermined distance from the designated coordinates, N question sentences that change as a result may be selected.

  Specifically, the query generation unit 25 extracts a keyword from each of the selected N question sentences according to a predetermined keyword extraction rule, and determines a keyword to be used as a search query from each of the extracted keywords.

  Although various methods can be used as a method for determining a keyword used as a search query, the query generation unit 25 has a high appearance frequency among keywords extracted from each of the N question sentences, and has designated coordinates. Are used as search queries in order from the keyword included in the question sentence having a shorter distance.

  In the second search query generation method, the query generation unit 25 uses the expression (1) for each keyword extracted from each of the selected N question sentences to obtain the keyword score Score (keyword) of the keyword keyword. And M keywords (M = 1, 2, 3,...) Are used as a search query in descending order of the keyword score Score.

  Here, df (keyword) is the appearance frequency of the keyword keyword in the N question sentences, and tf (n, keyword) is the keyword in the question sentence corresponding to the display coordinate whose distance from the specified coordinate is the nth closest. The frequency of occurrence of the keyword.

  Note that a predetermined value may be used as the value of M, or a variable value may be used. For example, by selecting a keyword having a keyword score Score equal to or higher than a predetermined value, M keywords that change as a result may be used as a search query.

  Then, the query generation unit 25 outputs a search query including the generated M keywords to the information search unit 26.

  The information search unit 26 inputs the search query received from the query generation unit 25 to a search engine (not shown), and obtains the search result searched by the search engine using the search query input from the information search unit 26 from the search engine. To do.

  The search engine to which the information search unit 26 inputs a search query is, for example, a search engine in an arbitrary search system that operates on a device (external device) different from the information search device 100 connected to the information search device 100 via a network. Can be used. In this case, the information search unit 26 can use the search engine by using a search interface (Application Programming Interface: API) provided for searching the Web or the like using the search engine.

  Then, the information search unit 26 outputs the search result for the search query acquired from the search engine to the display unit 40.

  The display unit 40 displays the search result received from the information search unit 26 in the search result display area 53 shown in FIG.

  As described above, the search result based on the search query generated from the question display area 52 clicked by the user with the mouse or the like is displayed in the search result display area 53.

<Operation of information retrieval device>
Next, the operation of the information search apparatus 100 according to this embodiment will be described.

  When the information search apparatus 100 accepts the question collection by the input unit 10, the information search apparatus 100 notifies the coordinate information creation unit 20A of the accepted question collection. Then, the information retrieval apparatus 100 executes a coordinate information creation processing routine shown in FIG.

  First, in step S100, a known category dividing method is applied to each question sentence included in the question collection received from the input unit 10, and each question sentence is divided for each category. Then, each divided question sentence is associated with a category and stored in, for example, the storage unit 30.

  In step S102, one unselected category is selected from the categories of the question sentence divided in step S100. Hereinafter, the category selected in step S102 will be referred to as “attention category” in particular.

  In step S104, one unselected question sentence is selected from each question sentence included in the attention category selected in step S102. In the following, the question sentence selected in step S104 will be referred to as “attention question sentence”.

  In step S <b> 106, feature amounts are extracted from the attention question sentence selected in step S <b> 104 using the above-described known feature amount extraction method and stored in the storage unit 30, for example.

  In step S108, the feature quantity extracted from the attention question sentence in step S106 is compressed into two dimensions using the known dimension compression method described above. Then, each element of the feature quantity compressed in two dimensions is normalized according to the size of the question display area 52, and each element of the normalized feature quantity is used as the attention question display coordinates in the question display area 52. It is associated with the sentence and stored in the storage unit 30.

  In step S110, it is determined whether or not all question sentences included in the attention category selected in step S102 have been selected in step S104. If there is a question sentence that has not been selected in step S104, step S104 is performed. Migrate to Then, an unselected question sentence is selected in step S104 until no unselected question sentences are included in the question sentence included in the attention category selected in step S102, and steps S104 to S110 are repeated, so that the attention category is selected. Display coordinates in the question display area 52 are associated with each of the included question sentences.

  On the other hand, if the determination process in step S110 is affirmative, the process proceeds to step S112.

  In step S112, a known clustering algorithm is applied to the data set having each of the display coordinates of each question sentence in the attention category calculated in step S108 as an element, and each of the display coordinates of each question sentence is included in the attention category. To divide into multiple clusters.

  In step S114, a question sentence score is set for each question sentence included in the target category.

  Specifically, first, a predetermined value is set as the question sentence score of each question sentence included in the attention category.

  Then, for each of the plurality of clusters in the target category divided in step S112, a question sentence that is the center of the cluster is set based on each display coordinate of each question sentence included in the cluster. The cluster center setting method is not limited. For example, the question sentence corresponding to the display coordinate having the shortest distance from the average value of the display coordinates in each question sentence included in the specific cluster is displayed in the specific cluster. It may be the cluster center.

  Then, for each cluster, by adding a predetermined value to the question sentence score of the question sentence associated with the cluster center, the question sentence score of the question sentence associated with the cluster center becomes a question sentence other than the cluster center. It should be larger than the question sentence score. Accordingly, for example, if the question text displayed in the question display area 52 is displayed larger as the value of the question text score associated with the question text increases, the question associated with the cluster center in the attention category. The character of the sentence is displayed in the question display area 52 larger than the character of the question sentence other than the cluster center.

  Therefore, the user can quickly grasp a characteristic question sentence from each question sentence displayed in the question display area 52 based on the size of the character of the question sentence. Can be improved.

  In step S116, the question sentence ID, the display coordinates calculated in step S108, and the question sentence score set in step S114 are associated with each of the question sentences included in the attention category, and the question collection coordinate group DB 30A. In category unit.

  In step S118, it is determined whether or not all categories divided in step S100 have been selected in step S102. If there is a category that has not been selected in step S102, the process proceeds to step S102. Then, until there is no unselected category among the categories divided in step S100, an unselected category is selected in step S102, and steps S102 to S118 are repeated, whereby a question sentence ID, a question sentence, display coordinates, and Each of the data sets associated with the question sentence scores is stored in the question collection coordinate group DB 30A for each category.

  With the above processing, the coordinate information creation processing routine shown in FIG. 4 ends.

  Next, when the information search device 100 provides an information search service to the user, the information search device 100 executes an information search processing routine shown in FIG. In executing the information search processing routine, it is assumed that the user has already selected a specific category in the category selection area 51.

  First, in step S200, each of the data sets stored in the question collection coordinate group DB 30A is read for each category by executing the coordinate information creation processing routine shown in FIG. 4 and included in the category selected by the user in the category selection area 51. Each question sentence is displayed in the question display area 52 of the search screen 50. In this case, the question text is displayed in the question display area 52 at the position represented by the display coordinates of the question text in the size of the character corresponding to the question text score. As described above, as an example, the larger the question sentence score, the larger the character size of the question sentence displayed in the question display area 52 is displayed.

  The category selected by the user in the category selection area 51 is particularly referred to as “selected category”.

  In step S202, whether or not the question display area 52 is clicked is determined based on the presence or absence of designated coordinates notified from the operation system of the information search device 100 when the user clicks the mouse in the question display area 52. In the case of negative determination, the process of step S202 is repeated until the user clicks the question display area 52 with the mouse. On the other hand, if the determination is affirmative, the process proceeds to step S204.

  In step S204, the designated coordinates notified from the operation system in step S202 are stored in the storage unit 30.

  In step S206, referring to the question collection coordinate group DB 30A, one unselected question sentence is selected from each question sentence included in the selected category.

  In step S208, display coordinates associated with the question sentence selected in step S206 are acquired from the question collection coordinate group DB 30A.

  In step S210, the distance between the designated coordinates acquired in step S204 and the display coordinates of the question sentence acquired in step S208 is calculated. As already described, there is no limitation on the method of calculating the distance between the designated coordinates and the display coordinates of the question text, and the distance is calculated using a known distance calculation algorithm such as the Eugrid distance or the cosine distance.

  The calculated distance is stored in the storage unit 30 in association with the question sentence selected in step S206, for example.

  In step S212, it is determined whether or not all question sentences included in the selected category have been selected in step S206. If there are question sentences in the selected category that have not yet been selected in step S206, the process proceeds to step S206. Transition. Then, until there is no unselected question sentence from the question sentences included in the selected category, an unselected question sentence is selected in step S206, and steps S206 to S212 are repeated, whereby the question sentence included in the selected category is selected. Each of them is stored in the storage unit 30 in correspondence with the distance from the designated coordinates in step S210.

  On the other hand, if the determination process in step S212 is affirmative, the process proceeds to step S214.

  In step S214, a question sentence is selected from the question sentences included in the selected category based on the distance associated with each question sentence included in the selected category in step S210. Then, keywords are extracted from the question sentence selected in this step, and a search query corresponding to the designated coordinates acquired in step S204 is generated based on the extracted keywords.

  For example, when the second search query generation method is used, N questions in order from the shortest distance associated with the question sentence in step S210 among the question sentences included in the selected category. Select a sentence. Then, for example, in accordance with the keyword extraction rules described above, keywords are extracted from the selected N question sentences, and for each of the extracted keywords, a keyword score Score is calculated for each keyword using equation (1). Then, M keywords are generated as search queries in descending order of the keyword score Score.

  Note that, as described above, the search query generation method executed in this step is not limited to this, and for example, corresponds to the shortest distance among the distances associated with each question sentence included in the selected category in step S210. The keywords may be extracted from the question sentence to be searched, and each of the extracted keywords may be used as a search query.

  In step S216, the search query generated in step S214 is input to a search engine in an arbitrary search system using an API provided in advance for searching the Web or the like, and the search result for the search query is acquired from the search engine. To do.

  In step S218, the search result acquired from the search engine in step S216 is displayed in the search result display area 53 of the search screen 50.

  In the present embodiment, a description has been given of an embodiment in which a question sentence in the question display area 52 is designated using a mouse. However, a touch panel is attached to a display device that displays the search screen 50, and a question sentence is designated with a finger or the like. Needless to say, you can.

  With the above processing, the information retrieval processing routine shown in FIG. 5 is completed.

  As described above, the information search device 100 according to the present embodiment displays the question text included in the collected question collection at a position based on the feature amount of the question text in the question display area 52, and the user can The specified coordinates are accepted by clicking the mouse in the vicinity of the question sentence that seems to indicate the content close to the information request held. Then, the information search apparatus 100 automatically generates a search query based on the received designated coordinates and the display coordinates of the question text in the question display area 52, and the search result searched by the search engine using the generated search query. Is output to the search result display area 53.

  Therefore, unlike the conventional search method using a search engine, the information search apparatus 100 does not require the user to input a search query into the search window of the search engine. Compared to the case, the operability related to the search can be improved. In particular, it is possible to improve the operability of a mobile terminal such as a smartphone or a tablet terminal in which character input is relatively difficult due to the size.

  In addition, the information search apparatus 100 automatically generates a search query by simply clicking the vicinity of a question sentence that seems to indicate content close to the information request that the information search apparatus has. Therefore, since it is possible to obtain a search result for a search query that is difficult for the user to come up with, it is possible to obtain a variety of information compared to when the user reviews the search query. As described above, since the content corresponding to the information request can be examined from the viewpoint that it is difficult for the user to come up with the information, the information retrieval method according to the present invention can be used particularly for an educational purpose application or the like.

  In addition, when the user clicks on the question display area 52 with the mouse, the information search apparatus 100 displays the search result of the question according to the designated coordinates. Compared to, it is possible to obtain a large amount of information at high speed. Therefore, the information search method according to the present invention can be used in a search terminal or the like of a call center that requires quick search.

  Further, the information search device 100 can automatically generate a search query based on a plurality of question sentences displayed in the vicinity of the area clicked by the user. Therefore, even when there is no question sentence suitable for the user's information request in the question display area 52, when the user clicks in the vicinity of the question sentence that seems to be similar to the information request, the search result suitable for the information request Can be obtained.

  Furthermore, in the information search apparatus 100, the higher the importance of the question sentence, the larger the character size of the question sentence displayed in the question display area 52. Therefore, as compared with the case where the question sentence is displayed with the same size character. The user can quickly grasp a characteristic question sentence from a plurality of question sentences.

  Note that the present invention is not limited to the above-described embodiment, and various modifications and applications are possible without departing from the gist of the present invention.

  For example, the information search apparatus 100 described above has a computer system inside, but the “computer system” includes a homepage providing environment (or display environment) if a WWW system is used. .

  In the specification of the present application, the embodiment in which the program is preinstalled in the ROM has been described. However, the program may be provided by being stored in a computer-readable recording medium.

DESCRIPTION OF SYMBOLS 10 ... Input part 20 ... Operation part 20A ... Coordinate information creation part 20B ... Information search part 21 ... Category division part 22 ... Feature-value extraction part 23 ... Coordinate calculation part 24 ... distance calculation unit 25 ... query generation unit 26 ... information search unit 30 ... storage unit 40 ... display unit 50 ... search screen 51 ... category selection area 52 ... question Display area 53 ... Search result display area 100 ... Information search device 30A ... Question collection coordinate group DB

Claims (8)

Based on the contents of a plurality of question sentences described in a natural language included in the collection of questions, each feature amount of the question sentence is extracted,
For each of the question sentences, based on the feature amount extracted from the question sentence, the display coordinates when displaying the question sentence on the screen are calculated,
Displaying each of the questions at the calculated display coordinates on the screen;
When the designated coordinates on the screen designated by the user operation are received, the distance between the display coordinates calculated for each of the question sentences and the designated coordinates is calculated,
Generating a search query based on a keyword included in the question sentence in which the calculated distance satisfies a predetermined condition;
Enter the generated search query into a search engine,
An information search method for displaying a search result for the search query searched by the search engine. When calculating the display coordinates of the question sentence, the importance of the question sentence is calculated based on the content of the question sentence,
The said question sentence is displayed so that the size of the character of the said question sentence displayed on the said screen may become large, so that the importance of the said question sentence is high when displaying the said question sentence. Information retrieval method. Based on the display coordinates of each of the question sentences, classify each of the question sentences into a plurality of clusters,
When displaying the question sentence, the question sentence is displayed such that the size of the character of the question sentence that is the cluster center in each of the clusters is larger than the size of the character of the question sentence that is not the cluster center. The information search method according to claim 2. Based on the content of the question sentence, the feature amount of each of the question sentences is extracted for each category,
Based on the feature amount, the display coordinates when displaying the question sentence on the screen are calculated for each category,
The information search method according to claim 1, wherein each of the question sentences is displayed for each category on the calculated display coordinates on the screen. 5. The search query including a keyword included in the question sentence displayed at a display coordinate having a shortest distance from the designated coordinate is generated when generating the search query. The information search method described in the section. When generating the search query, the distance from the designated coordinate is within the threshold for each keyword included in each of the question sentences whose distance from the designated coordinate is within a predetermined threshold. The keyword score is calculated so that a keyword included in the question sentence having a higher appearance frequency in each of the included question sentences and having a shorter distance from the designated coordinates has a higher keyword score, and the keyword score The information search method according to any one of claims 1 to 4, wherein the search query includes the top M keywords. A feature amount extraction unit that extracts each feature amount of the question sentence based on the contents of a plurality of question sentences described in a natural language included in the question collection;
For each of the question sentences, a coordinate calculation unit that calculates display coordinates when displaying the question sentence on the screen based on the feature amount extracted from the question sentence by the feature amount extraction unit;
A display unit for displaying each of the question sentences at display coordinates on the screen calculated by the coordinate calculation unit;
A distance calculation unit that calculates a distance between the display coordinates calculated by the coordinate calculation unit and the specified coordinates for each of the question texts when the specified coordinates on the screen specified by a user operation are received; ,
A query generation unit that generates a search query based on a keyword included in the question sentence in which the distance calculated by the distance calculation unit satisfies a predetermined condition;
An information search unit that inputs a search query generated by the query generation unit to a search engine and acquires a search result for the search query searched by the search engine;
With
The display unit displays a search result for the search query acquired by the information search unit on the screen.   The program for making a computer perform each step of the information search method of any one of Claims 1-6.