JP2018180914A - Content search device, content search method, program, and data structure - Google Patents

Abstract

PROBLEM TO BE SOLVED: To enable accurate content search.SOLUTION: A content search device 1 comprises: a collation parameter setting unit 23 configured to set a collation parameter that is used when collating candidate content with collation source content based on a feature of a candidate content being a candidate of content or editing technique; and a content DL/collation unit 24 configured to download the candidate content from an acquisition source of the candidate content, and collate the downloaded content with the content at the collation source using the collation parameter set by the collation parameter setting unit.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a content search device, a content search method, a program, and a data structure.

  In recent years, with the development of communication network technology and the progress of communication terminal devices represented by smartphones, personal computers (PCs) and the like, a large number of content distribution sites capable of uploading and downloading contents such as moving pictures have been established. In this type of content distribution site, users can easily upload content, and the number of content uploaded to the content distribution site is increasing year by year. In such content distribution sites, there is a problem that the content is uploaded by a non-rightholder who does not have the content right. Therefore, there is a need for a method for searching for illegal content uploaded by non-rightholders.

  Various methods are known as a method of searching for content from a content distribution site to which a large amount of content is uploaded. For example, in Non-Patent Document 1, a set of words related to a query input by a user is acquired using an inter-word relationship dictionary that defines causal relationships between words, upper-lower relationships, attribute relationships, etc. There is described a method of executing a search using not only the query input by the user but also a word acquired as being related to the query input by the user.

  Further, Non-Patent Document 2 uses a word-to-word relationship dictionary that defines causal relationships between words, upper-lower relationships, attribute relationships, etc., and a summary sentence of content selected by the user and a summary sentence of other content. A technique is described that evaluates the similarity with the user and presents highly similar content as content related to the content selected by the user.

  In addition, in Non-Patent Document 3, in order to collect an opinion on content, a tweet including a content name registered in Twitter (registered trademark) is posted, and then the same content name posted within a predetermined time is A method is disclosed that collects tweets (adjacent tweet group), extracts words having a high co-occurrence frequency in the adjacent tweet group as related words, and searches for tweets using the related words as a query.

  Further, Non-Patent Document 4 discloses a method of searching for harmful sites from a large number of sites that can be accessed via a communication network. In this method, a character string included in HTML (Hyper Text Markup Language) of a harmful site is statistically learned by SVM (Support Vector Machine), and a site including the character string extracted based on the learning in HTML is harmful. Determined as a site.

Miyazaki Taro, 6 others, "TV program search using inter-word relation dictionary," Proceedings of the 22nd Annual Conference of the Association for Language Processing Proceedings, March 2016, p. 917-920 Ichiro Yamada, 4 others, "Program similarity evaluation using random walk", Information Processing Society research report, Vol. 2012-ML-207, No. 12, July 27, 2012 Nakazawa Masami, 3 others, "Proposal of program related tweet collection method considering program viewer and topic change in program," Proceedings of Information and Systems Lecture 1, 2013 IEICE General Conference, March 2013 19th Ikeda, Kazufumi, 5 others, "Hazard site detection method based on HTML element", Journal of Information Processing Society, Vol. 52, no. 8, p. 2474-2483

  By searching using related queries as in the above-mentioned prior art, a large amount of content is extracted, and accordingly, the user is more likely to extract desired content. However, when a poster uploads illegal content, the content of the content illegally obtained from the collation source is lowered in order to reduce the accuracy of the automatic collation and to avoid being discovered by the right holder having the legitimate right to the content. I have to edit. In addition, the genre of the huge number of contents posted to the content distribution site is various. The appropriate setting in the matching process differs depending on the editing method and the genre, and if the appropriate setting is not made, it may not be possible to accurately search for content.

  Therefore, an object of the present invention made in view of such a point is to provide an illegal content search device, an illegal content search method, a program, and a data structure capable of searching for illegal content accurately.

  In order to solve the above problems, a content search apparatus according to the present invention is a content search apparatus for searching for content, and collates the candidate content based on the feature or editing method of the candidate content which is a candidate for the content. A collation parameter setting unit for setting a collation parameter when collating with the original content, downloading the candidate content from the acquisition source of the candidate content, and using the collation parameter set by the collation parameter setting unit A content DL / collation unit that collates the downloaded content with the content of the collation source.

  The content search method according to the present invention is a content search method executed by a content search apparatus for searching for content, and collates the candidate content based on the feature or editing method of the candidate content which is a candidate for the content. A step of setting a collation parameter at the time of collation with the original content; downloading the candidate content from an acquisition source of the candidate content; and using the set collation parameter, the downloaded content and the collation source And collating with the contents of

  Further, in order to solve the above problems, a program according to the present invention is characterized in that a computer is caused to function as the above content search device.

  Further, in order to solve the above problems, a data structure of a setting parameter according to the present invention is a data structure of a setting parameter used in a content search device searching for content belonging to one or more genres. The genre of the candidate content and the frame length of the candidate content, or the editing method and the matching method of the candidate content, and the content searching device determines the frame length corresponding to the genre of the candidate content or the matching corresponding to the editing method A method is specified, the candidate content is downloaded from the acquisition source of the candidate content, and the downloaded candidate content and the content of the comparison source are collated using the frame length as a unit of collation or based on the collation method. Used for processing.

  According to the present invention, the matching process can be appropriately performed according to the characteristic of the content indicated by the content profile. Therefore, the user can accurately search for content matching the content of the collation source.

It is a functional block diagram showing an example of composition of an illegal content search device concerning this embodiment. It is a figure which shows the example of the search query production | generation rule and the search query produced | generated according to the said search query production rule. It is a figure which shows the search query production rule and the other example of the search query produced | generated according to the said search query production rule. It is a figure which shows the example of the title of a content, and a profile. It is a flowchart which shows an example of the content search method which concerns on this embodiment.

  First, referring to FIG. 1, the functional configuration of the present embodiment of the present invention will be described. FIG. 1 is a functional block diagram of the illegal content search device 1 according to the present embodiment. The illegal content search device 1 of the present embodiment is a device searching for illegal content posted by a non-right holder, but is not limited to illegal content, and is a content searching device searching for content matching the content of the collation source. You may

  As shown in FIG. 1, the illegal content search device 1 includes an illegal phrase model generation unit 11, an illegal phrase model storage unit 12, a search query generation rule storage unit 13, a search query generation unit 14, and a matching candidate acquisition unit 15, profile estimation model generation unit 16, profile estimation model storage unit 17, content profile acquisition / estimation unit 18, exception content removal unit 19, verification priority calculation unit 20, verification parameter setting unit 23, Content DL (Download: download) and collating unit 24, an illegal word / phrase model updating unit 25, and a profile estimation model updating unit 26.

  The illegal content search device 1 searches for illegal content stored in the content acquisition source on the Internet based on the regular title and the meta information of the content of the collation source. Illegal content is content posted by a non-right holder. Meta information is attribute information that accompanies the (regular) content of the verification source, and for example, subtitles, performers appearing in the content, character names, broadcast number, broadcast date, performer, abbreviation, content genre including.

  The content acquisition source is a site (for example, a content posting site, a site in which URLs (Uniform Resource Locators) of illegal content are summarized) on the Internet where content is posted. The content acquisition source accepts the posting of the content from the communication terminal based on the request of the posting user, and stores the posted content. Also, the content acquisition source causes the communication terminal to download the stored content based on the user's request. The content acquisition source is, for example, a server device that manages a content posting site or the like, a distributed system configured with a plurality of servers, a cloud service, or the like. Also, "posting" means uploading and storing content. Moreover, a "posting user" is a user who posts content among users who use a content acquisition source.

  The illegal phrase model generation unit 11 generates an illegal phrase model by machine learning using, as learning data, a title of content to which a label indicating illegal or non-illegal is added. The illegal phrase model is a model that outputs an illegal phrase assumed to be used for illegal content for an arbitrary phrase.

  The illegal word model storage unit 12 stores the illegal word model generated by the illegal word model generation unit 11.

  The search query generation rule storage unit 13 stores a search query generation rule which is a rule for generating a search query for searching for illegal content from the content acquisition source from the regular title of the content (regular content) of the collation source. . As described above, the title of the illegal content is, for example, all of the regular titles of the regular content, in order to make the user recognize the relevance to the regular content while avoiding the illegal content being found from the rights holder. Alternatively, some may be converted (rephrased) into similar terms such as Kana conversion, Kanji conversion, Roman characters or abbreviations. The search query generation rule is a rule for generating a word or phrase obtained in the above-described paraphrase from the word or phrase contained in the title of the content. Note that the paraphrasing as described above can be output using, for example, a language processing method such as Word2Vec. In addition, the search query generation rule is a rule that generates a phrase including a fluctuation in the expression of the phrase included in the title of the content. The search query generation rule storage unit 13 may generate a word or phrase based on a search query generation rule having a tendency different for each genre of content (in the case of a moving image, a drama, an animation, a movie, etc.).

  The search query generation rule is, for example, a rule that Kana conversion, Romaji conversion, and Kanji conversion of an arbitrary word / phrase. In addition, the search query generation rule is a rule to translate an arbitrary phrase into a foreign language. In addition, the search query generation rule is a rule that any given word is converted into a written term.

  The search query generation unit 14 inputs, based on the operation of the operator of the illegal content search device 1, the content of the comparison source, and the regular title and meta information of the content.

  The search query generation unit 14 uses the illegal phrase model stored in the illegal phrase model storage unit 12 and the search query generation rule stored in the search query generation rule storage unit 13 to detect illegal phrases related to the regular title. Generate a search query that contains

  Specifically, the search query generation unit 14 generates a search query including an illegal phrase whose probability value calculated from the illegal phrase model described above is equal to or higher than a threshold value, based on the regular title. In addition, the search query generation unit 14 generates a search query including terms in different expressions having the same or similar semantic content as the terms included in the regular title. For example, the search query generation unit 14 generates a search query by performing kana conversion, kanji conversion, or romanization conversion on the terms included in the regular title according to the above-described search query generation rule.

  Further, the search query generation unit 14 may generate the input regular title as a search query.

  Further, the search query generation unit 14 can generate a search query including meta information. The search query generation unit 14 includes one or more of a regular title and meta information using a search query generation rule, for example, “title subtitle”, “title date”, “title broadcast number”, “performer”, Generate "abbreviated date" etc. as a search query. In the example illustrated in FIG. 3, for example, when the regular title is “Tuesday drama“ Trio ”” and the search query generation rule is the number-of-speaks notation (1), the search query generation unit 14 “Triio 1 episode” Generate a search query

  Further, the search query generation unit 14 uses the illegal phrase model or the search query generation rule to search for a phrase that paraphrases the regular title according to the tendency according to the genre of the content (drama, animation, movie, etc. Can be generated as

  The search query generation unit 14 outputs the search query generated as described above to the matching candidate acquisition unit 15.

  The collation candidate acquisition unit 15 searches the content acquisition source on the communication network based on the search query output by the search query generation unit 14, and the content matching the search query is a candidate having the possibility of illegal content. As content, identification information of the candidate content is acquired from a content acquisition source. The identification information is information for uniquely identifying the candidate content, and is, for example, an address of the content on the Internet, that is, a URL or the like. Further, the matching candidate acquisition unit 15 acquires, from the content acquisition source, the title of the content matching the search query and the profile (accompanying profile) attached to the content. Here, the profile acquired by the matching candidate acquiring unit 15 includes the content length, the posting time, the posting user name and the like shown in FIG. 4.

  Specifically, the collation candidate acquisition unit 15 extracts one or more contents having a title including the search query from the content group at the content acquisition source by the search query output from the search query generation unit 14. And collation candidate acquisition part 15 acquires each discernment information, a title, and an accompanying profile by making a fixed number of contents into candidate contents among contents extracted by contents acquisition origin. For example, the matching candidate acquisition unit 15 acquires identification information, a title, and an accompanying profile of each of the upper fixed number of contents among the contents extracted as having a high degree of matching with the search query.

  Also, among the content acquisition sources, not only the content extracted by the search using the search query, but also the related content having high relevance to the content extracted by the search using the search query (for example, the user who views and listens) Some layers have a function of extracting one or more same content (e.g., the same content), and presenting a title, a thumbnail, an accompanying profile, etc. of the extracted related content. In such a case, the matching candidate acquisition unit 15 may set the related content extracted by the content acquisition source as the candidate content, and may acquire identification information, a title, and an accompanying profile of the candidate content. At this time, the matching candidate acquisition unit 15 may acquire the degree of association indicating the degree of association of the related content with the content extracted by the search query.

  In addition, the collation candidate acquisition unit 15 may extract a predetermined number of top contents (newly arrived contents) newly stored in the contents acquisition source. In this case, the matching candidate acquisition unit 15 further acquires identification information, a title, and an accompanying profile of the candidate content, using the newly arrived content as the candidate content. The newly arrived content is, for example, content posted to the content acquisition source between a predetermined time before the reference and the reference based on the timing at which the matching candidate acquisition unit 15 acquires the content. As a result, the collated candidate acquiring unit 15 can search for illegal content without omission when illegal content is included in content acquired based on a search query or content that can not be acquired only with related content.

  The number of candidate contents acquired by the collation candidate acquisition unit 15 is an appropriate number preset in a list format or the like as a setting parameter according to the contents acquisition source and the contents. For example, when there is a content acquisition source that tends to post a large amount of illegal content for a specific genre (drama, animation, etc.), the matching candidate acquisition unit 15 selects candidate content of a specific genre from the content acquisition source. Get more than the candidate content of other genres. In addition, the matching candidate acquisition unit 15 may acquire more contents related to a posting user having a high posting rate of illegal content than contents related to other posting users. This can increase the possibility that the content identified by the candidate content includes illegal content.

  The matching candidate acquisition unit 15 outputs the identification information of the candidate content, the title, and the accompanying profile acquired from the content acquisition source to the content profile acquisition / estimation unit 18.

  The profile estimation model generation unit 16 generates a profile estimation model that outputs a statistical profile, which is statistical information on the content, based on the title of the content and the accompanying profile attached to the content. The profile estimation model is a model that indicates the correspondence between the title and the illegality, and the correspondence between the incidental profile and the illegality.

  Specifically, the profile estimation model generation unit 16 learns a title illegal probability indicating the probability that the posted content is illegal content for each title, and profile estimates a statistical model indicating the correspondence between the title and the title illegal probability. Generate as a model. The profile estimation model generation unit 16 preferably generates a profile estimation model by learning using a statistical model (SVM, naive Bayes, etc.) capable of calculating the accuracy to be classified into each class. Classification may be either illegal / non-illegal binary classification, or multi-valued classification such as content title (s) or non-illegal. The specific method of generating the profile estimation model can be any known method. For example, an example of the method is described in "Introduction to machine learning for language processing (Manabu Okumura, Takaya Takamura, Corona, p101-117)." In this method, the text that is the learning data is subjected to morphological analysis, and the contained words are extracted to perform learning, but it is also possible to use the text as a character string for learning without performing the morphological analysis.

  In addition, the profile estimation model generation unit 16 learns the posting user illegal accuracy indicating the accuracy that the posted content is illegal content for each posting user, and a statistical model indicating correspondence between the posting user and the posting user illegal accuracy. Generate as profile estimation model. The posting user illegality probability is estimated based on the illegality of the content posted by each posting user in the past. Specifically, the profile estimation model generation unit 16 determines the content rate of the content having a title with a high probability of illegality in the content posted by the posting user in the past, the feature amount such as the deletion rate of the content itself, and the illegality probability of the posting user Create a profile estimation model based on illegal probability learning data showing correspondence with.

  The profile estimation model storage unit 17 stores the profile estimation model generated by the profile estimation model generation unit 16.

  The content profile acquisition / estimation unit 18 acquires a statistical profile of candidate content based on the accompanying profile output by the matching candidate acquisition unit 15. The statistical profile of the candidate content acquired by the content profile acquisition / estimation unit 18 is information on the content of the content and information on the posting user. The profile of the candidate content acquired by the content profile acquisition / estimation unit 18 is, as shown in FIG. 4, the content length, the posting time, the posting user name, the illegal probability of the title, the illegal probability of the posting user, The deletion rate of the content posted by the posting user, the user type, the editing tendency type (with image editing) of the posting user, and the editing tendency type (with voice editing) of the posting user are included. Among these profiles, the probability of illegality of a title estimated using a profile estimation model and the probability of illegality of a posting user, which will be described in detail later, are referred to as statistical profiles.

  The user type is a type indicating whether or not the content posting user is a legitimate user who has generated (or has the right of) the content of the above-mentioned matching source. The user type is determined based on a previously created regular user list or the like. The editing tendency type is a type that indicates the presence or absence of editing performed on the content, for example, editing with a cut, time expansion and contraction, and special processing such as PinP. The editing tendency type is determined based on a list of editing tendency types of the posting user, which is created in advance for at least a part of the posting users.

  Specifically, the content profile acquisition / estimation unit 18 acquires the content length and posting time included in the accompanying profile output by the matching candidate acquisition unit 15 as the content length and posting time of the candidate content.

  In addition, the content profile acquisition / estimation unit 18 determines the user type of the posting user based on the posting user included in the accompanying profile output from the matching candidate acquisition unit 15 based on the authorized user list stored in advance in the memory. presume. Specifically, when the posting user is included in the authorized user list, the content profile acquisition / estimation unit 18 estimates that the user type of the posting user is legitimate. In addition, when the posting user is not included in the authorized user list, the content profile acquisition / estimation unit 18 estimates that the user type of the posting user is non-normal.

  Further, the content profile acquisition / estimation unit 18 estimates the editing tendency type of the posting user based on the posting user included in the accompanying profile output by the matching candidate acquiring unit 15 based on the editing tendency type list. The editing tendency type list is a list stored in advance in the memory, and is a list showing the correspondence between the posting user and the editing tendency type. The editing tendency type is a type of editing method frequently performed on the content related to the posting user. The type includes, for example, edit by cutting, time expansion / contraction, presence / absence of special processing such as PinP, and the like. The content profile acquisition / estimation unit 18 acquires the estimated editing method as the editing tendency type.

  Also, the content profile acquisition / estimation unit 18 stores the title illegality probability or the posting user illegality probability in the profile estimation model storage unit 17 based on the title of the candidate content and the posting user name output from the matching candidate acquisition unit 15 respectively. Estimate based on the profile estimation model In the following description, the accompanying profile and statistical profile acquired or estimated by the content profile estimation unit 18 and the profile estimated using the various lists by the content profile estimation unit 18 may be simply referred to as “profile”.

  Furthermore, the content profile acquisition / estimation unit 18 outputs the profile acquired or estimated as described above to the exceptional content removal unit 19 together with the identification information and the title.

  The exception content removal unit 19 determines exception content to be removed from the illegal content candidate based on the profile output from the content profile acquisition / estimation unit 18. Specifically, when the profile acquired by the content profile acquisition / estimation unit 18 satisfies a predetermined condition, the exceptional content removing unit 19 removes the candidate content corresponding to the profile as the exceptional content. The predetermined condition may be, for example, that the user type included in the profile is normal. Also, the predetermined condition may be, for example, that the posting time included in the accompanying profile is earlier than the publication time of the content of the matching source, or the content length included in the accompanying profile has a predetermined length (for example, several seconds) Degree) may be shorter. The predetermined conditions are not limited to these, and may be any conditions indicating that the candidate content is not likely to be illegal content.

  The exceptional content removing unit 19 outputs the identification information, the title, and the profile of the candidate content that has not been removed to the matching priority calculation unit 20.

  The matching priority calculation unit 20 calculates the priority in the matching process to be described later, based on the identification information, the title, and the profile of the candidate content not removed by the exceptional content removing unit 19.

  Specifically, the matching priority calculation unit 20 calculates the priority using the priority calculation model based on the illegality probability of the candidate content.

  First, the matching priority calculation unit 20 calculates the editing distance between the character string indicating the title of the candidate content and the character string indicating the title of the content of the matching source. The edit distance is a kind of distance indicating how different two strings are, and the procedure necessary to transform one string into the other by inserting, deleting, or replacing one character. Minimum number of times. That is, the smaller the editing distance, the higher the relevance between the character string indicating the title of the candidate content and the character string indicating the title of the content to be collated. In addition, the matching priority calculation unit 20, instead of the character string indicating the title of the content of the matching source, for example, names of characters appearing in the content of the matching source included in the meta information of the content of the matching source, appearance The edit distance between a character string such as a person's name and a subtitle and a character string indicating the title of the candidate content may be calculated.

  Further, the matching priority calculation unit 20 extracts identification information, a title, and a profile of candidate content related to a title whose editing distance is smaller than a predetermined value.

  Further, the matching priority calculation unit 20 may determine the relevancy based on both the editing distance for the title of the candidate content and the editing distance for the person's name. For example, the matching priority calculation unit 20 may calculate a score obtained by weighting each of the editing distance for the title of the candidate content and the editing distance for the name of the person and then adding them. In this case, the matching priority calculation unit 20 extracts candidate content related to a score smaller than a predetermined value.

  As described above, the collation candidate acquisition unit 15 acquires identification information of a wide range of candidate contents including newly arrived contents, in addition to the contents extracted based on the search query and the contents related to the contents. As a result, the candidate content identified by the identification information may contain a large amount of content having low relevance to the content of the matching source. Therefore, it is possible to reduce the processing load associated with the matching described later by the matching priority calculation unit 20 extracting only candidate content that is expected to have a small editing distance, that is, a high degree of relevancy. Become.

  When the candidate content is extracted based on the editing distance, the matching priority calculation unit 20 determines the priority based on the title illegalty probability of the extracted candidate content. At this time, the title illegal probability included in the profile estimated by the content profile acquisition / estimation unit 18 is used as the title illegal probability of the candidate content. Further, the matching priority calculation unit 20 may determine the priority using the posting user illegal probability instead of the extracted title illegalty of the candidate content. As the posting user illegality probability of the candidate content, the posting user illegality accuracy included in the profile estimated by the content profile acquisition / estimation unit 18 is used. Further, the matching priority calculation unit 20 may determine the priority based on both the title illegality probability and the posting user illegality probability. For example, the matching priority calculation unit 20 can set the sum of values obtained by weighting each of the title illegal probability and the posting user illegal probability as the priority. Further, the matching priority calculation unit 20 may determine the priority based on the combination of the previously calculated editing distance and each illegal probability.

  Further, the matching priority calculation unit 20 outputs the priority calculated by the matching priority calculation unit 20 to the content DL / matching unit 24 together with the identification information of the candidate content related to the priority, the title, and the profile.

  The collation parameter setting unit 23 sets a collation parameter to be used in the collation processing based on the feature of the candidate content, using the list of setting parameters stored in advance. The matching is to determine whether the candidate content matches the content of the matching source. The setting parameters are, for example, a frame length and a matching method. The frame length is a length of a frame which is a basic unit of matching in the matching process. The collation method includes a method of collating by voice, a method of collating by image, and the like.

  Any known method can be used as a matching method. For example, “Media fingerprint technology for identifying music and video and its application (Kawanishi Takahito, et al., The Japan Society for Industrial and Applied Mathematics, Applied Mathematics 21 (4), P.289-292, Dec. 22, 2011) Shows an example of the method.

  The setting parameter list is a list in which appropriate settings are described corresponding to profiles of candidate content or combinations of profiles. The profile of candidate content used in the setting parameter list is for estimating a frame length required to such an extent that the accuracy of matching is ensured, and is, for example, a genre. When the genre of the candidate content is sports mashup content, the candidate content is configured by editing a short moving image of about several seconds. Therefore, in the setting parameter list, for example, the setting of a short frame length (for example, about 2 seconds to 3 seconds) is described corresponding to the genre of mashup content of sports. As a result, the content DL / collation unit 24 can perform collation processing with the set short frame length, and can detect candidate contents matching the content of the collation source.

  On the other hand, when the genre of the candidate content is a drama or a movie, the content length is from several tens minutes to several hours. For this reason, in the setting parameter list, for example, setting of a long frame length (for example, about 5 minutes) is described corresponding to a genre such as drama or movie. As a result, the content DL / collation unit 24 can perform collation processing with the set long frame length, and can accurately detect candidate contents matching the content of the collation source.

  Also, the profile of candidate content used in the setting parameter list may be, for example, an editing method. The editing method is a method of editing performed on the content, and includes, for example, PinP for embedding an image in an image, time extension and the like, and the like. The collation parameter setting unit 23 may set the editing method of the candidate content as the editing tendency type of the posting user acquired by the content profile acquisition / estimation unit 18.

  For example, if the editing method of the candidate content is PinP and an image similar to the image of the content of the matching source is embedded in different images, the candidate content is recognized as the entire image as different from the content of the matching source . Therefore, it is difficult to determine that the candidate content has a high degree of matching with the content of the matching source by image matching. For this reason, in the setting parameter list, for example, the setting of collation by voice is described corresponding to the editing method of PinP. As a result, the content DL / collation unit 24 can perform voice collation to accurately detect candidate content that matches the content of the collation source.

  Also, for example, when the editing method of the candidate content is time expansion and contraction, the voice of the candidate content is largely different in the feature amount to be extracted from the voice before editing. Therefore, in the case where the candidate content is obtained by expanding and contracting the regular content in time, it is difficult to determine that the candidate content has a high degree of matching with the content of the comparison source by the audio matching process. For this reason, in the setting parameter list, for example, the setting of collation by image is described corresponding to the editing method of time expansion and contraction. As a result, the content DL / collation unit 24 can perform collation using an image, and can accurately detect candidate content that matches the content of the collation source.

  The content DL / collation unit 24 downloads candidate content from the content acquisition source in descending order of the priority calculated by the collation priority calculation unit 20. Then, the content DL / collation unit 24 determines whether the candidate content matches the content of the collation source by collating the downloaded candidate content with the content of the collation source according to the setting by the collation parameter setting unit 23 Do. The content DL / collation unit 24 outputs the identification information of the illegal content as the illegal content that is the candidate content that matches the content of the collation source.

  In addition, when the content length of the content is long, the content DL / collation unit 24 can efficiently search for illegal content by downloading in order from the candidate content with the highest priority and collating in order from the downloaded content. It is.

  In addition, the content DL / collation unit 24 may download candidate content having a long content length (for example, moving image content of several dozen minutes to several hours) and start collation from a portion downloaded in parallel. In this case, the content DL / collation unit 24 cancels the download of the remaining time of the candidate content when the candidate content matches the content of the collation source. The content DL / collation unit 24 outputs the identification information of the illegal content, with the matched candidate content as the illegal content. Then, the content DL / collation unit 24 downloads and collates the candidate content with the next highest priority. As a result, the content DL / collation unit 24 can shorten the time required to collate one candidate content, that is, increase the number of candidate contents collated per unit time.

  Further, the content DL / collation unit 24 outputs the title of the illegal content to the illegal phrase model updating unit 25 together with the label indicating the illegality. Further, the content DL / collation unit 24 outputs the profile of the illegal content to the profile estimation model updating unit 26 together with the label indicating the illegality.

  The illegal phrase model updating unit 25 updates the illegal phrase model based on the title of the illegal content output from the content DL / matching unit 24. Specifically, the illegal phrase model updating unit 25 causes the illegal phrase model generation unit 11 to update the illegal phrase model by machine learning with the title of the illegal content as new learning data. This is expected to increase the accuracy of the illegal word model.

  The profile estimation model updating unit 26 updates the profile estimation model based on the profile of the illegal content output from the content DL / collation unit 24. Specifically, the profile estimation model updating unit 26 causes the profile estimation model generation unit 16 to update the profile estimation model by machine learning with the profile of the illegal content as new learning data. This is expected to increase the accuracy of the profile estimation model.

  Subsequently, a content search method executed by the illegal content search device 1 according to the present embodiment will be described with reference to the flowchart shown in FIG. FIG. 5 is a flowchart showing an example of the content search method.

  First, the search query generation unit 14 inputs the content of the collation source, the title, and the meta information based on the operation of the operator of the illegal content search device 1 (step S1).

  When the content to be collated, the title, and the meta information are input in step S1, the search query generation unit 14 generates a search query using an illegal phrase model and a search query generation rule (step S2).

  When a search query is generated in step S2, the matching candidate acquisition unit 15 causes the content acquisition source to extract candidate content corresponding to the search query based on the search query, and the identification information and title of the extracted candidate content. And the accompanying profile are obtained (step S3).

  When the identification information, the title, and the incidental profile are acquired in step S3, the content profile acquisition / estimation unit 18 further acquires or estimates the profile of the candidate content based on the acquired incidental profile (step S4).

  When the profile is acquired or estimated in step S4, the exceptional content removing unit 19 removes candidate content that satisfies the predetermined condition from the candidate for illegal content based on the acquired profile (step S5).

  When the identification information of the exception content is removed in step S5, the matching priority calculation unit 20 determines each of the titles and the profiles of the candidate content identified by the pieces of identification information not removed by the exception content removing unit 19. The priority of the candidate content is calculated (step S6).

  When the priority of each candidate content is calculated in step S6, the content DL / collation unit 24 downloads the candidate content from the content acquisition source in descending order of priority, and the downloaded candidate content is input in step S1. It collates with the content of collation origin (step S7).

  Note that a computer can be suitably used to function as the illegal content search device 1 described above, and such a computer is a computer that describes a program that describes processing content for realizing each function of the illegal content search device 1. It can be realized by storing it in a database and reading out and executing this program by the CPU of the computer.

  The program may also be recorded on a computer readable medium. Computer readable media are available to be installed on a computer. Here, the computer readable medium having the program recorded thereon may be a non-transitory recording medium. The non-transitory recording medium is not particularly limited, but may be, for example, a recording medium such as a CD-ROM or a DVD-ROM.

  As described above, according to the present embodiment, the illegal content search device 1 uses the list of setting parameters stored in advance to compare the candidate content with the candidate content based on the feature or editing method of the candidate content. Set the matching parameter when matching with. Then, the illegal content search device 1 collates the downloaded content with the content of the collation source using the collation parameter set by the collation parameter setting unit. Therefore, for example, even if the candidate content is an edited regular content, matching can be performed based on an element (for example, an image, a sound, etc.) that does not change due to the editing. Also, for example, accurate matching can be performed by appropriately setting the unit to be matched according to the length of the candidate content. Therefore, the illegal content search device 1 can accurately search for illegal content.

  In the above description, the illegal content search device 1 according to the present embodiment is not limited to illegal content, and may be a content search device that searches for content that matches the content of the collation source. Is the accuracy with which the content of the match source is matched.

  Although the embodiments described above have been described as representative examples, it will be obvious to those skilled in the art that many modifications and substitutions can be made within the spirit and scope of the present invention. Therefore, the present invention should not be construed as being limited by the above-described embodiments, and various modifications and changes are possible without departing from the scope of the claims.

1 Illegal Content Search Device 11 Illegal Word Model Generation Unit 12 Illegal Word Model Storage Unit 13 Search Query Generation Rule Storage Unit 14 Search Query Generation Unit 15 Matching Candidate Acquisition Unit 16 Profile Estimation Model Generation Unit 17 Profile Estimation Model Storage Unit 18 Content Profile Acquisition -Estimator 19 Exception content removing unit 20 Collation priority calculation unit 23 Collation parameter setting unit 24 Content DL / Collation unit 25 Illegal phrase model update unit 26 Profile estimation model update unit

Claims (6)

A content search apparatus for searching for content, comprising:
A collation parameter setting unit configured to set a collation parameter when the candidate content is collated with the content of the collation source based on the feature or the editing method of the candidate content serving as the candidate of the content;
A content DL / collation unit that downloads the candidate content from an acquisition source of the candidate content and collates the downloaded content with the content of the collation source using the collation parameter set by the collation parameter setting unit; ,
A content search apparatus comprising: In the content search device according to claim 1,
The content DL / collation unit may order the priority according to the degree of certainty that the candidate content matches the content of the collation source, or the editing distance between the title of the candidate content and the title of the content of the collation source. A content search device that matches the content with the content of the matching source. In the content search device according to claim 1 or 2,
The content DL / collation unit collates the content of the collation source sequentially from the downloaded portion without waiting for the download of the candidate content to be completed, and the content in which the candidate content matches the content of the collation source A content search device that stops downloading of the candidate content when it is determined that A content search method executed by a content search apparatus for searching content, comprising:
Setting a matching parameter when matching the candidate content with the content of the matching source based on the feature or editing method of the candidate content as the content candidate;
Downloading the candidate content from an acquisition source of the candidate content, and collating the downloaded content with the content of the collation source using the set collation parameter;
A content search method comprising:   A program for causing a computer to function as the content search device according to any one of claims 1 to 3. A data structure of setting parameters used in a content search apparatus for searching for content belonging to one or more genres, which is a data structure of
The genre and the frame length of the candidate content which is the candidate of the content, or the editing method and the matching method of the candidate content,
The content search apparatus specifies a frame length corresponding to the genre of the candidate content or the matching method corresponding to the editing method, downloads the candidate content from an acquisition source of the candidate content, and downloads the candidate content A data structure of a setting parameter used for processing of matching candidate content and content of matching source as a unit of matching using the frame length or based on the matching method.

Images