JP4878591B2 - Content search device, content search method, program, and recording medium - Google Patents

Description

  The present invention relates to a content search apparatus and method for searching for content by designating a tag from a content database to which a keyword representing the content content is assigned as a tag.

In particular, tags that are conceptually related to each other display a tag cloud that is placed close to the two-dimensional space, and by devising a background image, various structural relationships between tags can be inferred It is related with the content search apparatus and method which show the tag cloud which can select easily the tag which a user is interested by doing.

  With the increase in the amount of information on the Internet and the advancement of search engine technology, various information can be acquired from the Internet. However, in order to select the information required by the user from the vast amount of information, it is necessary for the user to appropriately provide a keyword for search, and the burden on the user to think about the keyword is small. Absent. In the first place, it is inevitable to input characters from the keyboard when inputting keywords.

  In recent years, in order to reduce the burden on these users as much as possible, it has been used by setting metadata consisting of short words and phrases representing the content of the searched content as tags. Attempts have been made to make it easier to get to the information that people want.

  For example, in Flickr (registered trademark) of a photo sharing site and Youtube (registered trademark) of a moving image sharing site, keywords, categories, and genres that express the subject can be set as tags for photos and moving images. Also, in many blogs, tags can be set by the blog author. Technorati (registered trademark) blog search provides a system that can search for blog articles by specifying tags.

  There is what is called a tag cloud as an interface for a user to access tagged content. This “tag cloud” displays a set of tags having hyperlinks to tag search using tags as search conditions in a rectangular area.

  A tag cloud allows a user to click a tag displayed on the tag cloud via a mouse or a Wii (registered trademark) remote control of Nintendo (registered trademark) Co., Ltd. Content can be searched, and the work of inputting keywords by the user for the search can be reduced.

In addition, when a tag is added to content, there is a method in which the author or reader of the content explicitly gives it. In addition, a method for extracting a person name, an organization name, a unique expression of a place name, and the like included in the text in the content, and automatically extracting those that appear characteristically from these as tags (for example, Patent Document 1) is known. In addition, a set of texts to which a tag is explicitly given is used as learning data, and a tag is automatically assigned to text that is not assigned a tag based on tags assigned to similar texts in the learning data. Has been proposed (for example, see Non-Patent Document 1).
JP 2005-339139 A Shigeru Fujimura et al., Automatic multi-tagging for blog articles, Proceedings of the Japan Society for Artificial Intelligence, 3G8-3, 2007

  Any of the above prior arts is a technique for assigning a tag, and is not a technique for assisting a user in selecting content to which a tag is attached. Therefore, when various tags are added to a large number of contents, the area of the tag cloud increases. For example, in a general resolution display, it is limited to display a tag cloud containing about 500 tags. It is. However, in order to classify various contents, it is thought that 5,000 to 10,000 types of classification are necessary, and if 5,000 to 10,000 classifications are displayed as a simple tag cloud In this case, there is a problem that it is difficult to find an appropriate tag from the displayed tag cloud. In other words, many of today's tag clouds are arranged in alphabetical order or character code order, and tags placed nearby are not related to each other, so it is difficult to find an appropriate tag from among them. There is.

  For this reason, in the above conventional example, when about 5,000 tags are enumerated, it is no longer possible for a user to manually find a tag of interest, and tag search must be used. In this case, the advantage of tag cloud that can be searched by just clicking without keyboard input is lost, and it falls to the end.

An object of the present invention is to provide a content search apparatus and method that allow a user to easily select an appropriate tag when displaying a tag cloud composed of a large number of tags.

The present invention registers content with a tag in the content database in a content search device that searches content from a content database configured by a content set to which a tag that is a classification keyword is assigned, using the tag as a search condition. Based on the tag registration information, based on the tag statistical information, the inter-tag similarity calculation means for measuring the similarity between the tags, which is the similarity between the tags, and the inter-tag similarity Tag coordinate calculation means for calculating tag coordinates, which are coordinates on a two-dimensional plane of the tags, and storing them in a storage device so that the tags having high similarity between the tags are close to each other, and the tag Tag map generating means for generating a tag map in which each tag is arranged at the coordinates and storing it in a storage device; Tag map display means for displaying a map on a user terminal, tag selection means for acquiring a tag selected by a user from tags displayed on the tag map, and storing the tag in a storage device, and the tag Using the tag selected through the selection means as a search condition, the content to which the tag is attached is searched from the content database and stored in the storage device, and the search result searched by the content search means And a search result display means for displaying on the user terminal, the tag coordinate calculation means using the inter-tag similarity calculated by the inter-tag similarity calculation means, and the inter-tag similarity seen from each tag i Tags with a high degree are arranged in descending order, a high-order tag set N (i) is selected, a link from i to j (εN (i)) is extracted, and the similarity between the tags is defined as a link. Altitude coordinate calculating means for calculating the centrality score of the network as the altitude coordinates of the node as the tag, and the tag map generating means creates a three-dimensional solid model using the altitude coordinates, The content search apparatus includes a three-dimensional model generation unit that generates a tag cloud image with a feeling.

According to the present invention, in the tag cloud, the similarity between tags is calculated based on the statistical frequency of words included in the text to which the tag is attached. The tag coordinate calculation means arranges tags calculated by the means and having high similarity at positions close to each other on the two-dimensional plane, so that when the user finds one tag with high interest, The presence of the related tag can be easily noticed, and the related content can be easily found.

  The best mode for carrying out the invention is the following examples.

  FIG. 1 is a block diagram showing the overall configuration of a content search apparatus 100 that is Embodiment 1 of the present invention.

  The content search device 100 is a device that searches for content from a content database configured by a content set to which a tag that is a classification keyword is assigned using a tag as a search condition.

  Further, the content search apparatus 100 includes a content registration unit 10, a content database 11, tag statistical information 12, an inter-tag similarity calculation unit 20, a tag coordinate calculation unit 30, a tag map generation unit 40, a user interface. Unit 50 and content search means 60.

  The content registration unit 10 registers tagged content in the content database 11.

  The tag similarity calculation means 20 measures the similarity between tags based on the tag statistical information 12. That is, the inter-tag similarity calculation unit 20 generates a tag feature vector based on the frequency information of the words included in the content to which the tag is attached, and according to the similarity of the tag feature vector, The similarity between tags is calculated.

  The tag coordinate calculation means 30 calculates the tag coordinates, which are the coordinates on the two-dimensional plane of the tag, so that the tags having high similarity are close to each other based on the similarity between the tags. That is, the tag coordinate calculation means 30 is provided with network information that uses a tag as a node and a similarity between tags as a link, and places nodes having strong links on the two-dimensional plane at positions close to each other (below a predetermined distance). And the tag attribute is calculated as altitude coordinates.

  The tag map generation means 40 generates a tag map 41 in which the respective tags are arranged at the tag coordinates. That is, the tag map generation means 40 creates a three-dimensional surface model from a data set (x, y, z) set at an arbitrary position, applies light from an oblique direction, and captures a stereoscopic effect photographed from above. A certain two-dimensional image is generated as the tag map 41.

  The user interface unit 50 includes tag map display means 51, tag selection means 52, and search result display means 53.

  The tag map display means 51 displays the tag map 41 on the user terminal. That is, the tag map display means 51 displays on the user terminal device the tag map 41 that can scroll the tag map 41 generated by the tag map generation means 40 and click a tag on the tag map 41.

  The tag selection means 52 acquires the tag selected by the user from the tags displayed on the tag map 41. That is, the tag selection unit 52 includes a marker coordinate acquisition unit. This marker coordinate acquisition means is means for acquiring a change in the position of the tag map 41 moved by the user, such as mouse dragging, and acquiring coordinates near the center of the tag map 41. Further, the tag selection unit 52 acquires a list of tags near the center of the tag map 41 acquired by the marker coordinate acquisition unit, and searches the content database 11 for tags.

  The content search unit 60 searches the content database 11 for the content to which the tag is assigned using the tag selected by the tag selection unit 52 as a search condition.

  The search result display means 53 displays the search result on the user terminal. That is, the search result display unit 53 displays the search result searched by the content search unit 60 on the user terminal device.

  Note that the content in the above embodiment includes objects in real space that can be tagged, such as Web pages on the Internet, digital content of images, sounds, and moving images, products, services, and the like.

  Moreover, in the said Example, although a tag is a collection of the keywords used as a label for classifying content, and is normally given to content manually, the said patent document 1 or the said nonpatent literature 1 Alternatively, the automatic tagging technique described in the above may be used to automatically apply the information.

  Next, detailed operation of each means will be described.

  FIG. 2 is a flowchart illustrating processing for registering content in the first embodiment.

  The process for registering the content shown in FIG. 2 is repeatedly executed once started until it is explicitly stopped by an operator operating the system.

  FIG. 3 is a diagram illustrating an example of content registered in the content database 11.

  In S <b> 1, the user explicitly registers content in the content database 11 via the content registration unit 10 as shown in FIG. 3. The content registration means 10 is a general search engine, and may automatically collect and register resources on the Internet by a crawler conventionally used. However, in any method, the content to be registered is tagged content.

  The content to which the tag is attached is, for example, del. icio. Like a social bookmark service such as us, there is bookmark information to which a tag is attached when a user bookmarks a Web page. In the blog article, since a keyword is specified in the <dc: subject> field of RSS, the program article to which such RSS is given is an example of content to which a tag is assigned.

  As shown in FIG. 3, the content database 11 stores content IDs, tags, contents, and the like as data associated with content units. Here, the content ID is an identifier uniquely given to each content. The tag is a word or phrase that directly represents the content of the text set by the document creator. The content is the content itself. As shown in FIG. 3, additional information such as a URL, registration date and time, and a title may be added to hold the data.

  In addition, in this way, instead of explicitly assigning a tag, the user's name, organization name, place name specific expression, etc. included in the text in the content are extracted and characteristic from these That is automatically extracted as a tag (see Patent Document 1), or a text set to which a tag is explicitly given as learning data, and the similarity of text to text that has not been tagged Depending on the degree, the tag may be automatically assigned by a method of automatically assigning the tag (see Non-Patent Document 1) or the like.

  FIG. 4 is a diagram illustrating an example of the tag statistical information 12.

  Further, the content registration means 10 extracts the tag statistical information 12 from the tagged content, and creates a table of the tag statistical information 12 as shown in FIG. However, it is not always necessary to acquire the tag statistical information 12 from the content database 11 registered in the content registration means 10, and the tag statistical information 12 may be acquired from another information source. For example, if the content to be registered in the content database 11 is content that does not include text information such as a photo or video, a tag feature vector necessary for calculating the similarity between tags described later cannot be generated. .

  In such a case, the tag statistical information 12 may be acquired from information (such as a blog article) different from the content database 11. However, the following description is based on the assumption that the content stored in the content database 11 is a blog article and the tag statistical information 12 is acquired from the blog article.

  The tag statistical information 12 is created based on tag setting information stored in the content database 11. As shown in FIG. 4, the tag ID, the tag, the number of documents, the tag feature vector, and the similar tag are stored as data in a format in which each tag is associated. The tag ID is an identifier uniquely given to each tag. The number of documents is the number of contents for which the tag is set in the content database 11.

  The tag feature vector is a score vector of words that appear characteristically in the content to which the tag is attached, and is list-type data by a set of certain real values. As a method for generating a tag feature vector, the frequency of words that frequently appear in a document to which a tag is attached may be used, and there is also a method for obtaining a characteristic vector reflecting the contents. This is described below.

Similar tags are calculated by calculating the cosine similarity between vectors by using the tag feature vector and arranging the high similarity items in descending order until reaching a specific threshold. It is a list type data described by a set of pairs of numerical values. For example, when C _i and C _j are feature vectors of a tag, the similarity similarity is expressed by the following equation (1).

Similarity = C _i · C _j / | C _i || C _j | Expression (1)
As a method of generating a tag feature vector in which tag features are more expressed, there is a method of using a phrase vector having a pair of a word / phrase in a document set in which the tag is set and a residual DF value as an element. Yes (see Shigeru Fujimura, Ko Fujimura, Ryoji Kataoka, Masahiro Oku, Scoring of similarities between tags of Blog, Japan Database Society Letters, Vol.15, No.4, pp.33-36, 2006).

The “residual DF value” is represented by a difference in document frequency in the document set estimated by the document frequency and Poisson distribution in the document set of the phrase. For example, the total number of search result document sets is n, the document frequency of phrase i in the document set is df _i , the total number of documents is N, and the global frequency of phrase i in all documents is Assuming F _i , the residual DF value is obtained by the following equation (2).

  As an example, the residual DF value of the phrase “J-League” in the set of blog articles tagged with “soccer” is the number of collected blog entries, 400000, and the blog tagged with “soccer” If the number of entries is 1000, the document frequency of “J-League” in this article set is 300, and the global frequency of “J-League” in the entire collected blog articles is 700, the residual DF is 300− 1000 (1-exp (−700/400000)) = 298.25.

  The “residual DF value” is an index based on the following assumptions. Since document sets to which the same tag is assigned are highly likely to be similar in terms of both contents and phrases used, the document frequency of feature words representing the contents of the document is high. On the other hand, it is known that the probability distribution of the number of appearances in a document of a general phrase that appears in any document regardless of the meaning of the sentence is well approximated by the Poisson distribution. Therefore, by taking the difference between the two (the actual document frequency and the frequency approximated by the Poisson distribution), the influence of the general word can be canceled and the feature word value can be made more prominent.

  Next, a process for generating the tag map 41 will be described.

  FIG. 5 is a flowchart showing a process for generating the tag map 41.

  The process for generating the tag map 41 shown in FIG. 5 is repeatedly executed at regular intervals with a frequency of once a day or once a week.

  In S11, the inter-tag similarity calculation unit 20 creates a similar tag list for each tag, and in S12, the tag coordinate calculation unit 30 determines that tags having high similarities are close to each other on a two-dimensional plane (a predetermined value or less). The coordinates (x, y) of each tag such that the distance) is obtained are calculated. In this case, when it is regarded as a network having a tag as a node and a link from each tag i to a similar tag j as an edge, this is a general network. Therefore, conventionally proposed network visualization techniques Can be used.

  When calculating the coordinates (x, y) of each tag such that tags having a high degree of similarity are close to each other on a two-dimensional plane, "JP 2004-318739 A", "Kamada, T. and Kawai , S., An algorithm for drawing general undirected graphs, Information Processing Letters, Vol. 12, No. 31, pp. 7-15, 1989 ”may be used.

However, if all the tags j similar to each tag i are used and these algorithms are used, the number of links may be too large and the calculation time may be enormous. In this case, the number of links may be reduced and calculated by increasing a threshold for selecting a similar tag j. For example, if a set of tags having a degree of similarity equal to or greater than the threshold q with the tag i is N _q (i), the tags are arranged in descending order from the link with the high degree of similarity, and _q (i) Only the book links may be considered.

k _q (i) = αlog (1+ | N _q (i) |) +1 Equation (3)
Thereby, even when there are many strong links, the number of similar tags can be narrowed down to some extent. Even if there are few strong links, at least one of them can be adjacent to one another. As a result, it is possible to prevent a halfway position from being placed by a large number of weak links.

  Also, when the network visualization algorithm cannot consider the link weight, that is, when the link weight is only allowed to be 1 or 0, the link having a similarity higher than the above threshold is set to 1 (with link), and otherwise The tag coordinates can be calculated by giving the algorithm link information in which 0 (no link) is set to (link having a similarity equal to or less than the threshold value).

  The tag coordinate calculation means 30 calculates a z coordinate which is an altitude coordinate in addition to the (x, y) coordinate. Note that the z-coordinate is necessary for the tag map generating means 40 to display the relationship between the tags in an easy-to-understand manner by adding a shading or coloring to the two-dimensional tag map 41 in S13. However, this coloring is not essential, and it is not necessary to calculate the z-coordinate when not coloring.

  When calculating the z-coordinate, the logarithm of the number of contents to which a tag is simply attached may be normalized between [0, 1] and used. Since tags given by many people are general tags, the topic structure of the tag map 41 can be easily understood if the general tags are in a prominent position at a high altitude.

  Further, when a plurality of tags are auto-tagged in the content, for example, it is assumed that a high-level concept tag such as “baseball” or “sports” is simultaneously given to an article on the topic of “Hanshin”. As a result, the frequency of the upper tag conceptually increases, and the z coordinate is obtained in the relationship of sports> baseball> Hanshin.

  In addition, since tags of higher concepts such as “sports” have a large number of similar tags, there is a high possibility that they will be arranged at the center even on the xy plane.

  FIG. 6 is a diagram illustrating a display example of the tag map 41.

  As a result, as shown in FIG. 6, the conceptual relationship between the tags can be easily grasped from the height of the tags and the positional relationship.

  Another method for calculating the z coordinate is a method for calculating the centrality score of the network. For example, the link extracted by the above equation (3) is regarded as a directed graph from i to j, and from that link, “S. Brin and L. Page, The Anatomy of a Large-scale Hypertextual Web Search Engine, In Proceedings of It may be calculated using the PageRank (registered trademark) equation (4) described in “7th International World Wide Web Conference, 1998.”.

  Here, d is the probability of random jump, and a real number between [0, 1], for example, 0.15 is used. IN (i) is a set of tags having a link to the tag i. Out (j). It is the sum of the weights of the links coming out from the tag j. In this method, tags similar to many tags have many backlinks, but having many similar tags is likely to represent a more general idea. Therefore, similarly to the tag frequency, the z-coordinate is obtained in the relationship of sports> baseball> Hanshin.

  As another method of calculating the z-coordinate, a method of simply using the sum of similarity links of each tag, or a similarity link of each tag in a space where the tags are arranged at (x, y) coordinates. A method of using the total length for calculating the z coordinate is also conceivable.

  In the method in which the sum of the link lengths of the similarities of each tag is used as the z coordinate, if the sum of the link lengths is large, it means that similar tags are widely dispersed on the xy plane. . Therefore, it means that the tag is conceptually wide related to many concepts. As a result, it is possible to obtain the same effect as when using the tag frequency or PageRank (registered trademark).

  The tag map generation means 40 generates a three-dimensional solid model in S13 from the set of (x, y, z) coordinates of each tag acquired by the coordinate axis calculation described above, and applies a light source from an oblique direction. A tag map 41 with a feeling is created.

The tag map generation means 40 generates an image according to the following procedure.
(1) Generate equidistant grid data from a set of (x, y, z) coordinates generated by the tag coordinate calculation means 30.
(2) A three-dimensional model is generated from grid data, light is applied obliquely, and shading is applied.
(3) Color-coded according to altitude (altitude), photograph a three-dimensional model from directly above, and output a two-dimensional image.

  These processes are described in `` Wessel, P., and Smith, WHF., The Generic Mapping Tools, Version 4.2.0, Manual, http://www.soest.hawaii.edu/gmt/, 2007 ''. It can be easily realized by using such a tool. This tool includes (1) surface, (2) grdgradient, (3) grdimage, and the like as commands corresponding to the processes (1) to (3).

  Next, the user interface unit 50 will be described.

  First, as a preparation, a huge background image of about 10000 × 10000 pixels generated by the tag map generation means 40 is divided into tiles of about 250 × 250 pixels (in this case, divided into 1600 sheets), and a storage device on the server Keep in.

  FIG. 7 is a flowchart showing a processing flow of the user interface unit 50 for creating an HTML document for displaying the tag map 41 on the user terminal.

  The process of the user interface unit 50 shown in FIG. 7 is repeatedly executed once activated until it is explicitly stopped by the operator.

  The user interface unit 50 realizes a scrollable map by repeating the following five steps (1) to (5).

(1) [Display of tag map 41] (S21)
Of the tag map 41 divided on the tile, a tile including a default position or a center coordinate (hereinafter referred to as “marker coordinate”) designated by the user is used as a center, and the vicinity of the top, bottom, left, right (including diagonal) is also included. A total of nine tile images are sent to the terminal together with the eight tiles.

  Specifically, an HTML document that displays an area inside nine tiles centered on the marker coordinates is generated and sent to the Web browser of the terminal device, and the Web browser displays the nine documents described in the HTML document. The tile image is acquired from the server and displayed.

  Note that the tag map generation means 40 may embed the text of the tag such as “sports” in the image of the tag map 41 in advance, or the tag map 41 is used as the background image in the HTML document and the vicinity of the marker coordinates The tag text may be displayed by embedding a Java (registered trademark) Script code acquired from the server together with the coordinates, and placing the tag text in an upper layer of the tag map 41. .

(2) [Obtain marker coordinates] (S22)
The HTML map sent to the browser is embedded with Java (registered trademark) Script to acquire an operation event such as a drag of the user, and the tag map 41 is displayed by an intuitive operation of dragging the tag map 41. You may enable it to scroll and to move a marker position.

  This is facilitated by using the Java (registered trademark) Script library described in “Fuchs, t., Et.al., script.aculo.us, http://script.aculo.us/”. Can be realized. Also, along with the movement of the marker position, the marker position is transmitted to the server at any time, nine tiles on the top, bottom, left, and right are also acquired from the server at any time, and the tag map 41 centered on the marker position is always displayed. Request related content with a tag near the marker.

(3) [Tag selection] (S23)
On the server side, by calculating the Euclidean distance between the acquired marker coordinates and each element of the set of tag coordinates generated by the tag coordinate calculation means 30, the number of tags in the vicinity of the marker coordinates is reduced to K. get. K is an arbitrary number.

(4) [Content Search] (S24)
A query using the acquired K tags as an OR condition is generated, and the content search means 60 displays a search result such as the title, URL, and summary of the content to which the corresponding tag is attached from the content database 11. Get the information you need. These are transmitted to the terminal as a response to the related content request in step (2).

(5) [Display of search results] (S25)
FIG. 8 is a diagram illustrating a display example of the tag map 41 in the first embodiment.

  On the terminal side, the related content transmitted from the server is displayed as a periphery of the tag map 41 or a squirt on the tag map 41 as shown in FIG. Further, in some embodiments, instead of simply displaying the result of the OR condition search of K tags, the priority is set so that the closest content comes to the top by the marker coordinates among the K tags. You may make it display.

  Note that the tag map 41 is not realized as a scrollable map that is moved by dragging the mouse or the like as in steps (2) and (3) above, but the tag is simply clicked by the user in step (3). A tag on the map 41 may be selected. In this case, the marker coordinate acquisition means is unnecessary.

  Note that the tag map display means 51 can realize various additional functions. For example, in order to search for a position on the tag map 41 that the user is interested in, first, as shown in FIG. 8, a search window for tag search displayed in the upper left is prepared and displayed in the middle left. The ranking of popular tags with a lot of content is displayed, and the frequently visited position for each user is bookmarked as shown in the lower left, and the tag map 41 is displayed for the second time. You may be able to jump to that position by clicking. Further, the default position may be displayed from the past history of the user.

  In addition, for a tag bookmarked by the user, an icon in the shape of a house, for example, is placed on the tag map 41, and an icon designed by a person at a position on the tag map 41 currently being viewed by the user. By arranging (Avatar), a lively situation on the tag map 41 can be expressed. Furthermore, you may make it implement | achieve the function which communicates, such as chat, between avatars.

  According to the above-described embodiment, it is possible to easily search for content to which a tag is attached by scrolling and clicking the tag map 41 without inputting a query from the keyboard. However, such an interface using the tag map 41 is not only used for searching the content with a tag from the tagged content database 11 but also used for generating a search query for an external search engine. can do.

FIG. 9 shows a list of topic words (topic list) related to the position of the marker coordinates of the tag map 41 on the right side of the tag map 41, and performs a search for an external search engine simply by clicking the topic word. It is a figure which shows a screen.

  FIG. 10 is a block diagram showing a content search apparatus 200 that is Embodiment 2 of the present invention.

  In the content search device 100, the content search device 200 has tag statistical information 13 instead of the tag statistical information 12, and has a user interface unit 50 a instead of the user interphase unit 50.

  The user interface unit 50 a includes a topic list display unit 54 and an external search engine call unit 55 instead of the search result display unit 53 in the user interface unit 50.

  The topic list display means 54 displays a topic list. The external search engine calling means 55 sends a search request as a search query to the external search engine for the topic selected by the user.

  In the content search device 200, as described above, a tag search is performed from the content database 11 based on the marker coordinates, and a search result set of the tag search is acquired. Then, the topic list extracting / displaying means 54 uses the technique described in the above-mentioned Patent Document 1 to extract characteristic expressions (topic words) that are characteristically included from the search result set of the tag search.

  These topic words are displayed as a topic list. The external search engine calling means 55 can be realized by a normal browser by embedding a URL with a query of the external search engine as a hyperlink of a topic list.

  According to the above embodiment, the tag cloud calculates the similarity between tags based on the conceptual frequency of the tags based on the statistical frequency of words included in the text to which the tags are attached. The tags are calculated by the means 20, and the tags having high similarity are arranged at positions close to each other on the two-dimensional plane by the tag coordinate calculating means 30, so that the user finds one tag with high interest. If it is possible, the presence of other related tags can be easily recognized, and related content can be easily found.

  In addition, according to the above embodiment, if the tag cloud is displayed as a scrollable map that can be scrolled up and down, left and right, a tag having a large area can be used on a mobile phone or a TV with a low resolution without depending on the screen size. A part of the cloud can be displayed.

  If the user scrolls the tag cloud by operating the mouse or the like, related contents can be automatically displayed one after another in conjunction with the display position of the tag cloud.

  Further, according to the above-described embodiment, the search for content searches for tags close to the center coordinates of the tag map 41 under the OR condition. For example, tags that are semantically close such as “baseball” and “Hanshin” , There is no need for the user to explicitly specify and search. In addition, it is possible to easily specify which is preferentially displayed based on the coordinate position.

  Furthermore, according to the above embodiment, the tag coordinate calculation means 30 gives the generality of the tag as the altitude coordinates of the tag cloud, and the tag map generation means 40 forms a three-dimensional model and applies the light source from an oblique direction. Thus, since the tag map 41 having a three-dimensional feeling is created, it is possible to easily grasp which concept each tag belongs to from the surrounding terrain structure. For example, even if a certain tag is located at the same distance in both sports and politics tags, the concept of tags depends on whether they are included in the foot of the mountain of sports or politics tags. Can quickly grasp the relevant relationship.

  Furthermore, according to the above embodiment, feature words can be extracted from the content database 11 using coordinate positions on the tag map 41, and search word candidates can be presented to the user. The user can transmit a search request to another external search engine simply by selecting the presented search word candidate by clicking the mouse or the like.

  In other words, in the above-described embodiment, in a content search apparatus that searches for content using a tag as a search condition from a content database configured by a content set to which a tag that is a classification keyword is assigned, the content database is tagged. Content registration means for registering content, similarity between tags based on the tag statistical information, and inter-tag similarity calculation means for storing in a storage device, Tag coordinates that are coordinates on the two-dimensional plane of the tag are calculated so that tags with high similarity exist close to each other, and the tag coordinate calculation means for storing in the storage device, and the tag coordinates, Tag map generating means for generating a tag map in which tags are arranged and storing the tag map in a storage device, and the tag map Tag map display means for displaying on a user terminal, tag selection means for acquiring a tag selected by a user from tags displayed on the tag map, and storing the tag in a storage device, and the tag selection means Using the tag selected via the search condition, the content to which the tag is attached is searched from the content database and stored in the storage device, and the search result searched by the content search unit is used. It is an example of the content search apparatus which has a search result display means displayed on a user terminal.

  In this case, the tag selection means displays a scrollable image in which the tag map is scrolled up and down and left and right by a mouse operation, and has marker coordinate acquisition means for acquiring the center position of the tag map as marker coordinates. The content search means is a tag arranged on the tag map, the tag existing in the vicinity of the marker coordinates is acquired from the tag map coordinate data, and the tag is used as a search condition. This is means for searching the content database for content to which a tag is attached.

  Further, the tag coordinate calculation means is means having an altitude coordinate calculation means for calculating a tag attribute score calculated based on the statistical information of the tag as an altitude coordinate of the node as the tag, and the tap map generation means This is a means having a three-dimensional model generation means for generating a tag cloud image with a three-dimensional effect by creating a three-dimensional solid model using altitude coordinates.

  Further, the tag attribute score is a score calculated based on the frequency information of the content to which the tag is attached.

  In addition, the tag attribute score uses the inter-tag similarity information calculated by the inter-tag similarity calculating means, and arranges tags having a high similarity in descending order as seen from each tag i, and sets the upper tag set N (i ), A link from i to j (εN (i)) is extracted, and the centrality score of the network having the similarity between the tags as a link is the altitude coordinate of the node.

  And the said tag attribute score calculated the altitude coordinate according to the sum total of the distance on the two-dimensional plane to the similar tag in each tag using the inter-tag similarity information calculated by the said inter-tag similarity calculating means It is a score.

  Moreover, the said Example can be grasped | ascertained as invention of a method. In other words, in the above-described embodiment, in the content search method for searching for content using a tag as a search condition from a content database configured by a content set to which a tag that is a classification keyword is assigned, the content database is tagged. Based on the content registration step for registering the content, the tag statistical information, the similarity between the tags is measured and stored in the storage device, the tag similarity calculation step, and the tag similarity between each other, Tag coordinates that are coordinates on the two-dimensional plane of the tag are calculated so that tags with high similarity are close to each other, and stored in a storage device, and the tag coordinates A tag map generation step of generating a tag map in which tags are arranged and storing the tag map in a storage device, and the tag map A tag map display step for displaying on the user terminal, a tag selection step for acquiring the tag selected by the user from the tags displayed on the tag map, and storing the tag in the storage device, and the tag selection step Using the tag selected via the search condition, the content to which the tag is attached is searched from the content database and stored in the storage device, and the search result searched by the content search step is used. It is an example of the content search method which has a search result display process displayed on a person's terminal.

  Further, the above embodiment can be grasped as a program invention. That is, the said Example is an example of the program which makes a computer implement | achieve the said content search apparatus. That is, the said Example is an example of the program which makes a computer implement | achieve the said content search method.

Moreover, the above embodiment can be grasped as an invention of a recording medium. That is, the above-described embodiment is an example of a computer-readable recording medium that records the above program. As a recording medium, a CD, a DVD, an FD, an HD, an optical disk, a magneto-optical disk, a semiconductor memory, and the like are conceivable.

1 is a block diagram illustrating an overall configuration of a content search apparatus 100 that is Embodiment 1 of the present invention. 6 is a flowchart illustrating processing for registering content in the first embodiment. It is a figure which shows the example of the content registered into the content database. It is a figure which shows the example of the tag statistical information. It is a flowchart which shows the process which produces | generates the tag map. It is a figure which shows the example of a display of the tag map. It is a flowchart which shows the flow of a process of the user interface part 50 which produces the HTML document for displaying the tag map 41 on a user terminal. It is a figure which shows the example of a display of the tag map 41 in Example 1. FIG. The figure which displays the list | wrist of the topic word (topic list) relevant to the position of the marker coordinate of the tag map 41 on the right side of the tag map 41, and performs the search with respect to an external search engine only by clicking a topic word. is there. It is a block diagram which shows the content search apparatus 200 which is Example 2 of this invention.

Explanation of symbols

100: Content search device,
10: Content registration means,
11 ... Content DB,
12 ... Tag statistics information,
20 ... means for calculating similarity between tags,
30: Tag coordinate calculation means,
40. Tag map generation means,
41 ... Tag map,
50 ... User interface part,
51. Tag map display means,
52. Tag selection means,
53. Search result display means,
200 ... content search device,
12 ... Content DB,
13 ... Tag statistics information,
50a ... User interface part,
54 ... Topic list extraction display means,
55: External search engine calling means.

Claims (5)

In a content search apparatus that searches for content using a tag as a search condition from a content database configured by a content set to which a tag that is a classification keyword is assigned.
Content registration means for registering tagged content in the content database;
An inter- tag similarity calculating means for measuring the inter-tag similarity, which is the similarity between the tags, based on the tag statistical information and storing it in the storage device;
A tag that calculates tag coordinates, which are coordinates on a two-dimensional plane of a tag, based on the above-described similarity between tags so that tags having high similarity between the tags are close to each other, and stores them in the storage device Coordinate calculation means;
Tag map generating means for generating a tag map in which each tag is arranged at the tag coordinates and storing the tag map in a storage device;
Tag map display means for displaying the tag map on the user terminal;
Tag selecting means for acquiring a tag selected by the user from the tags displayed on the tag map and storing it in a storage device;
Content search means for searching for a content to which the tag is attached from the content database using the tag selected via the tag selection means as a search condition;
Search result display means for displaying the search result searched by the content search means on the user terminal;
Have
The tag coordinate calculation means uses the inter-tag similarity calculated by the inter-tag similarity calculation means, arranges tags having a high inter-tag similarity as seen from each tag i in descending order, and sets the upper tag set N (i ), A link from i to j (∈N (i)) is extracted, and the centrality score of the network having the inter-tag similarity as the link is calculated as the altitude coordinate of the node that is the tag Having coordinate calculation means,
The content search apparatus, wherein the tag map generation unit includes a three-dimensional model generation unit that generates a three-dimensional tag cloud image by creating a three-dimensional solid model using the altitude coordinates . In claim 1,
The tag selection means is a means having marker coordinate acquisition means for displaying a scrollable image in which the tag map is scrolled up and down and left and right by a mouse operation, and acquiring the center position of the tag map as marker coordinates.
The content search means is a tag arranged on the tag map by calculating a distance between the acquired marker coordinates and the tag coordinates calculated by the tag coordinate calculation means , wherein the marker coordinates The content search apparatus is a means for acquiring a tag present in the vicinity of the tag from the tag map coordinate data, and searching the content database for the content to which the tag is attached using the tag as a search condition. . In a content search method for searching for content using a tag as a search condition from a content database configured by a content set to which a tag that is a classification keyword is assigned,
A content registration step of registering tagged content in the content database;
An inter-tag similarity calculation step of measuring an inter-tag similarity, which is a similarity between tags, based on the tag statistical information, and storing it in a storage device;
A tag that calculates tag coordinates, which are coordinates on a two-dimensional plane of a tag, based on the above-described similarity between tags so that tags having high similarity between the tags are close to each other, and stores them in the storage device A coordinate calculation step;
A tag map generating step of generating a tag map in which each tag is arranged at the tag coordinates and storing the tag map in a storage device;
A tag map display step of displaying the tag map on the user terminal;
A tag selection step of acquiring a tag selected by the user from the tags displayed on the tag map and storing the tag in a storage device;
A content search step of searching the content database for the content to which the tag is assigned using the tag selected through the tag selection step as a search condition;
A search result display step of displaying the search result searched in the content search step on the user terminal;
Have
The tag coordinate calculation step uses the inter-tag similarity calculated in the inter-tag similarity calculation step, arranges tags having a high inter-tag similarity as seen from each tag i in descending order, and sets the upper tag set N (i ), A link from i to j (∈N (i)) is extracted, and the centrality score of the network having the inter-tag similarity as the link is calculated as the altitude coordinate of the node that is the tag A coordinate calculation step,
The tag map generation step includes a three-dimensional model generation step of generating a three-dimensional tag cloud image by creating a three-dimensional solid model using the altitude coordinates . The program which makes a computer implement | achieve the content search apparatus of Claim 1 or Claim 2 . The computer-readable recording medium which recorded the program of Claim 4.

Images