JP6446987B2 - Video selection device, video selection method, video selection program, feature amount generation device, feature amount generation method, and feature amount generation program - Google Patents

Description

  The present invention relates to an information processing technique for selecting an image using a computer or the like.

  When reproducing a video together with audio, if the content of the reproduced audio and the content of the video are far from each other, a viewer who views the audio and video feels uncomfortable. However, manually selecting a video that is related to the audio content is a very complicated task. Examples of the reproduced voice include narration, reading, announcement, and lyrics of music.

  An example of a technique for displaying an image in accordance with music is disclosed in Patent Document 1. The slide show creation server disclosed in Patent Literature 1 includes an original slide show including lyrics for each line of music, a series of images displayed for each line, and an overall impression word estimated using the lyrics. Data is stored for multiple original slide shows. The slide show creation server extracts image feature amounts from a plurality of images designated by the user. The slide show creation server adds a tag to the image data using the extracted image feature amount. The slide show creation server assigns the entire impression label to all the designated images using the assigned tag. The slide show creation server selects an original slide show that matches the assigned overall impression label. The slide show creation server creates a new slide show by replacing the image data of the selected original slide show with an image designated by the user.

  Patent Document 2 describes a content search device that calculates a distance indicating a degree of similarity of a plurality of contents classified into one or more types that are not necessarily the same among a plurality of types. The content search device derives a feature amount of a derivatable content feature (one or more of an image feature, an acoustic feature, and a semantic feature). The content search apparatus estimates the feature amount of an unknown feature of the content using the derived feature amount based on the correlation between different features calculated in advance. For example, the content search apparatus estimates the acoustic feature amount of the image content based on the image feature amount and the semantic feature amount derived from the image content and its metadata. The content search device calculates the similarity based on all the derived and estimated feature quantities.

  Patent Document 3 describes a music classification device that classifies music into categories. The music classification device generates, by learning, a category classifier that classifies music to a classification destination specified by a category name based on the lyrics data of the music. The music classification device classifies music classified into categories into subcategories by clustering.

JP 2014-115729 A International Publication No. 2010/053160 JP2013-214326A

  The content of the reproduced voice can be represented by text.

  The slide show creation server of Patent Document 1 creates a new slide show by replacing the image data of the original slide show created in advance with images designated by the user. Therefore, the slide show creation server cannot select a video that matches the text represented by the music, for example.

  In addition, the original slide show may be created by music and image data that does not match the text represented by the lyrics of the music. In that case, the new slide show created by the slide show creation server is composed of music and image data that does not match the text represented by the lyrics of the music. That is, the slide show creation server does not always create a slide show composed of music and image data that matches the text represented by the music.

  The content search device of Patent Literature 2 estimates the similarity between contents by calculating the distance between the contents based on the calculated or estimated feature amount. For example, when the content search device searches for content having a small distance from the specified text, a video is not always obtained. In addition, since the distance between contents also depends on the acoustic feature amount, even if the semantic feature amount and the facial image feature amount of the matching text and video are close to each other, the difference in the calculated or estimated acoustic feature amount is different. If it is large, the distance between the text and the video will not be small. Therefore, the content search device cannot select a video that matches the text.

  The technique of Patent Document 3 is a technique for classifying music pieces. Therefore, the video matching the text cannot be selected by the technique of Patent Document 3.

  An object of the present invention is to provide a video selection device and the like that can reduce the load of matching text and video.

  The video selection system according to an aspect of the present invention performs a text mining process on text associated with a video to obtain a video feature amount that is a feature amount of the video for each of the plurality of videos. For each of the video feature generation means, the target feature generation means for generating the target feature quantity that is the feature quantity of the target text by executing the text mining process on the target text, and the video feature quantity Similarity degree deriving means for deriving a degree of similarity representing the degree of similarity of the video feature quantity with respect to the target feature quantity, and the video having a high degree of similarity with respect to the target feature quantity based on the derived similarity degree Video selection for selecting a feature and selecting the video associated with the text from which the selected video feature was derived It includes a stage, a.

  The feature amount generation device according to an aspect of the present invention is a phrase that modifies the attribute from the text based on an attribute that is a phrase related to at least one of a plurality of videos each associated with the text. A feature extraction unit that extracts a feature and stores the extracted feature in the feature storage unit; and for each of the plurality of videos, the text stored in the feature storage unit in a text associated with the video Video feature generation means for detecting each feature and generating a feature quantity representing the detected feature as a video feature quantity of the video;

In the video selection method according to one aspect of the present invention, a video mining process is performed on the text associated with the video, and the video feature quantity that is the video feature quantity is determined for each of the plurality of videos. Generate
By executing the text mining process on the target text, a target feature amount that is a feature amount of the target text is generated, and for each of the video feature amounts, the video feature amount is similar to the target feature amount A degree of similarity representing the degree of the image is derived, and based on the derived degree of similarity, the video feature quantity having a high degree of similarity to the target feature quantity is selected, and the selected video feature quantity is derived Select the video associated with the text.

  A feature value generation method according to an aspect of the present invention is a phrase that modifies an attribute from the text based on an attribute that is a phrase related to at least one of a plurality of videos each associated with the text. A feature is extracted, the extracted feature is stored in a feature storage unit, and each of the features stored in the feature storage unit in each of the plurality of videos is associated with the video. A feature amount representing the detected feature is detected and generated as a video feature amount of the video.

  The video selection program according to an aspect of the present invention performs a text mining process on a text associated with a video by a computer, thereby obtaining a video feature amount that is a feature amount of the video. Video feature generation means for generating each of the target feature generation means for generating a target feature quantity that is a feature quantity of the target text by executing the text mining process on the target text, and the video feature quantity A similarity degree deriving unit for deriving a degree of similarity representing the degree of similarity of the video feature amount with respect to the target feature amount, and a degree of similarity with respect to the target feature amount based on the derived degree of similarity. The video associated with the text from which the high video feature is selected and the selected video feature is derived A video selecting means for selecting operates in.

  The feature value generation program according to one aspect of the present invention modifies the attribute from the text based on an attribute that is a phrase related to at least one of a plurality of videos each associated with the text. A feature extraction unit that extracts a feature that is a phrase to be stored and stores the extracted feature in the feature storage unit; and for each of the plurality of videos, a text associated with the video is stored in the feature storage unit Each feature is detected and a feature amount representing the detected feature is operated as a video feature generation unit that generates a video feature amount of the video.

  The present invention has the effect of reducing the load for matching text and video.

FIG. 1 is a block diagram showing a configuration of a video selection device 1 according to the first embodiment of the present invention. FIG. 2 is a diagram illustrating an example of accompanying information according to the first embodiment of this invention. FIG. 3 is a flowchart showing an example of an operation of extracting features based on a plurality of texts and accompanying information in the feature generation device 100. FIG. 4 is a diagram schematically illustrating an example of extracted features. FIG. 5 is a diagram schematically illustrating an example of extracted features. FIG. 6 is a diagram schematically illustrating an example of a feature list. FIG. 7 is a flowchart illustrating an example of an operation for generating video features of the feature generation device 100 according to the first embodiment of this invention. FIG. 8 is a diagram schematically illustrating an example of text associated with a video for each video. FIG. 9 is a diagram schematically illustrating an example of a video feature vector. FIG. 10 is a flowchart illustrating an example of an operation of receiving a video feature vector of the video selection device 110 according to the first embodiment of this invention. FIG. 11 is a flowchart illustrating an example of an operation of selecting a video in response to receiving the target text of the video selection device 110 according to the first embodiment of this invention. FIG. 12 is a diagram schematically illustrating an example of the target feature vector. FIG. 13 is a diagram schematically illustrating an example of the target feature amount and the video feature amount from which the similarity deriving unit 113 derives the similarity. FIG. 14 is a diagram schematically illustrating an example of similarity. FIG. 15 is a block diagram illustrating an example of a configuration of a video selection system 1A according to a first modification of the first embodiment of the present invention. FIG. 16 is a block diagram illustrating an example of a configuration of a video selection system 1B according to a second modification of the first embodiment of the present invention. FIG. 17 is a block diagram illustrating an example of a configuration of a video selection system 1C according to a third modification of the first embodiment of the present invention. FIG. 18 is a block diagram illustrating an example of a configuration of a video selection system 1D according to the second embodiment of this invention. FIG. 19 is a block diagram illustrating an example of the configuration of a computer that can be used to implement the video selection device and the feature generation device according to each embodiment of the present invention.

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

<First Embodiment>
First, a first embodiment of the present invention will be described in detail with reference to the drawings.

  FIG. 1 is a block diagram showing the configuration of a video selection system 1 according to the first embodiment of the present invention.

  Referring to FIG. 1, a video selection system 1 according to the present embodiment includes a feature generation device 100 and a video selection device 110. The feature generation device 100 and the video selection device 110 are connected to be communicable. In the example illustrated in FIG. 1, the feature generation device 100 and the video selection device 110 are implemented as different devices. However, the video selection device 110 may include the feature generation device 100. The video selection device 110 may operate as the feature generation device 100.

  The feature generation device 100 includes an accompanying information receiving unit 101, an accompanying information storage unit 102, a teacher data receiving unit 103, a teacher data storage unit 104, an attribute extraction unit 105, a feature extraction unit 106, and a feature storage unit 107. And a video feature generation unit 108. The video selection system 1 may include a user terminal (not shown) that is used by the user of the video selection system 1 to input instructions and the like.

  The video selection device 110 includes a target reception unit 111, a target feature generation unit 112, a similarity derivation unit 113, a video selection unit 114, an output unit 115, a video feature reception unit 116, and a video feature storage unit 117. including. The video selection device 110 may further include a video reception unit 118 and a video storage unit 119.

  The teacher data receiving unit 103 receives a plurality of texts. In each embodiment of the present invention, data representing text is also simply referred to as text. For example, receiving text means receiving data representing the text. The plurality of texts may be input to the teacher data receiving unit 103 by the video selection device 110, for example. The plurality of texts may be input to the teacher data receiving unit 103 by, for example, a video server (not shown) that stores a plurality of videos described later. The plurality of texts may be input to the teacher data receiving unit 103 by a user terminal (not shown), for example.

  Each text received by the teacher data receiving unit 103 is associated with at least one of a plurality of videos related to the content of the text. Each of the plurality of videos only needs to be associated with at least one text. For example, the user of the video selection system 1 may associate the text and the video determined to be related to the content of the text in advance for each of the plurality of texts. The video determined to be related to the text content is, for example, a video determined to match the text content visually, bodily, or visually and bodily. The user of the video selection system 1 adds, for each of a plurality of videos, one or more texts determined to match the contents of the video visually, bodily, or visually and bodily. You may associate with. The text associated with the video may include the planning intent of the video. As will be described later, the planning intention is, for example, a phrase or a sentence representing the concept of the video, the purpose of producing the video, or the like.

  The teacher data receiving unit 103 may receive a plurality of sets of text and at least one video ID (Identifier) that is an identifier of the video. The teacher data receiving unit 103 stores the received text associated with at least one of the plurality of videos in the teacher data storage unit 104.

  The text is, for example, lyrics. The text may be text representing the content of a narration, reading, or announcement. In the following, the video selection system 1 when the text is lyrics will be mainly described.

  The teacher data storage unit 104 stores a plurality of texts associated with at least one of the plurality of videos.

  The accompanying information receiving unit 101 receives the accompanying information of each of the plurality of videos described above. The accompanying information may be input to the accompanying information receiving unit 101 by the video selection device 110, for example. The accompanying information may be input to the accompanying information receiving unit 101 by the above-described video server (not shown), for example. The accompanying information may be input to the accompanying information receiving unit 101 by a user terminal (not shown), for example. The accompanying information receiving unit 101 stores the received accompanying information in the accompanying information storage unit 102.

  The accompanying information storage unit 102 stores accompanying information for each of a plurality of videos.

  The accompanying information is information including a word representing the content of the video. The accompanying information may be represented by a word for each item representing a viewpoint for classifying videos (that is, the type of video), for example. The accompanying information may be expressed by sentences or phrases for each item. The accompanying information may be represented by one or both of a word and a sentence or a phrase for each item.

  FIG. 2 is a diagram illustrating an example of accompanying information. The accompanying information includes, for example, a plurality of types of information representing the content of the video. The plurality of types of information represent, for example, classification viewpoints and classification results when videos are classified into a plurality of categories from a plurality of viewpoints. In this case, for example, a classification viewpoint is expressed as an item, a name specifying the classification viewpoint is expressed as an item name, and a classification result (that is, a category in which videos are classified) is expressed as an item value. The accompanying information may not be a classification viewpoint and a classification result. The “item name” is a name representing the viewpoint of classification of the video whose contents are represented by the accompanying information. In FIG. 2, “example (content)” represents a specific example or content of information represented by each item name. A word not surrounded by parentheses is a specific example of a value for each item included in the accompanying information. Words enclosed in parentheses represent the contents of values for each item included in the accompanying information. The type of video whose content is represented by the accompanying information is represented by a value for each item. Referring to FIG. 2, the accompanying information includes, for example, values of items represented by item names such as planning intention, model, clothes, place, season, weather, time of day, and event. In the following description, for example, the value of an item whose item name is “planning intention” is expressed as “planning intention value” or the like.

  In the example illustrated in FIG. 2, the plan intention value is represented by, for example, a phrase or a sentence. The values of items other than the planning intention are represented by words. The phrase or sentence, which is the value of the planning intention, represents, for example, the concept of the video or the purpose of producing the video. The value of the model is, for example, the sex of a person appearing in a video such as a man, woman, man or woman. The model value may be an animal, a plant, or an object other than a person. The value of clothes is, for example, the clothes of a person appearing in the video, such as clothes and Japanese clothes. The value of clothes may simply be “clothes”. The value of the place is, for example, a word representing the place where the video is shot or the place where the subject of shooting exists. The value of the place may be a word representing a division of the place, such as a city or the sea, or a specific area name. The value of the season is a word that represents the season in which the video was shot. The value of the weather is a word representing the weather when the video is taken. The time value is a word indicating the time zone when the video was taken. The value of the event is a word representing an event occurring in the scene shot as a video. The accompanying information may include two or more words per item for the item represented by the word. There may be an item that does not include a word (that is, an item that does not have a value). The accompanying information is not limited to the example shown in FIG. The accompanying information may not include the items shown in FIG. The accompanying information may include information on items other than the items shown in FIG.

  The attribute extraction unit 105 extracts an attribute that is a word representing at least part of the content of the video from the accompanying information. When the accompanying information includes a word as a value, the attribute extraction unit 105 may extract the word. When the accompanying information includes a sentence or a phrase as a value, the attribute extraction unit 105 represents the feature of the picture for each picture from the sentence or the phrase by, for example, a TF-IDF (Term Frequency-Inverse Document Frequency) method. What is necessary is just to extract a word.

  For each of the extracted attributes, the feature extraction unit 106 extracts, from the text stored in the teacher data storage unit 104, a word / phrase related to the attribute (that is, modifies the attribute) as a feature. For each of the extracted attributes, the feature extraction unit 106 may extract, for example, a word or phrase relating to the attribute (that is, modifying the attribute) from the text associated with the video having the attribute as a feature. . The attribute of the video is an attribute included in the accompanying information of the video. The feature extraction unit 106 stores the extracted features in the feature storage unit 107. The feature extraction unit 106 may generate a feature list that is a list of all extracted features. The feature extraction unit 106 may store the generated feature list in the feature storage unit 107.

  For example, the feature extraction unit 106 may detect a phrase related to the attribute by basic text processing such as morphological analysis or syntax analysis on the text, and estimate the part of speech or characteristic of the detected phrase.

  The length of the phrase detected by the feature extraction unit 106 may be limited. In that case, the length of the phrase may be the number of units constituting the phrase. The unit that constitutes the phrase may be, for example, an adjective, an adjective verb, and a combination of a noun and a particle that constitute the phrase. For example, when the attribute is “person” and “tall” is detected as a phrase related to “person”, the feature extraction unit 106 uses, for example, “ What is necessary is just to specify "tall" and "high". Then, the feature extraction unit 106 may determine that the length of the phrase “tall” is 2. The feature extraction unit 106 may detect a word or phrase having a length equal to or less than a predetermined length (for example, two). The characteristic of the phrase is, for example, a characteristic indicating whether or not there is a possibility of modifying the attribute.

  The feature extraction unit 106 may use the text processing result to determine whether the detected word / phrase relating to the attribute is a word / phrase that can modify the attribute. If the feature extraction unit 106 determines that the detected phrase relating to the attribute can modify the attribute, the feature extraction unit 106 may extract the phrase as a feature.

  For example, if the detected word / phrase is a phrase that can modify other words, such as an adjective, an adjective verb, or a combination of a noun and a particle, the phrase can modify the attribute. What is necessary is just to determine that it is. In that case, the feature extraction unit 106 may extract the detected word / phrase as a feature.

  The feature extraction unit 106 further uses, for example, a dictionary including a combination of a word type and a word used to modify the word of the type, and a word detected as a word related to the attribute is You may determine whether it is used as a phrase which modifies the attribute. The feature extraction unit 106 extracts a phrase as a feature when the detected phrase is a phrase that can modify another word, and the detected phrase is used to modify the attribute that the phrase relates to. May be.

  The feature is not limited to the feature extracted by the feature extraction unit 106 from the text associated with the video. The feature may include, for example, a word / phrase representing the name of an event or a word / phrase representing a place, which is selected in advance. Features other than the features extracted from the text associated with the video by the feature extraction unit 106 may be stored in the feature storage unit 107 in advance by the administrator of the video selection system 1, for example. A dictionary in which phrases representing events and places are collected is created in advance by the administrator of the video selection system 1, for example, and accessed by the feature extraction unit 106 such as the feature dictionary storage unit (not shown) or the feature storage unit 107. It may be stored in a storage unit that can. Then, the feature extraction unit 106 may read a phrase from such a dictionary. The feature extraction unit 106 may store the phrase read from the above dictionary in the feature storage unit 107 as a feature.

  As described above, the video is associated with text such as lyrics, for example, which has been determined to be related to the video. A word (for example, the above-described attribute) extracted from information representing the content of the video (for example, the accompanying information described above) often appears in text associated with the video, such as lyrics. It is empirically known that the phrases related to the words extracted from the information representing the contents of the video appearing in the text associated with the video are “visual” and “sensible” phrases. ing. Therefore, the phrases extracted by the feature extraction unit 106 as the features are “visual” and “experienced” phrases. In other words, the feature extraction unit 106 can extract words such as “visual” and “sensible” as features.

  For each video, the video feature generation unit 108 performs a text mining process on the text associated with the video, thereby generating a video feature amount that is a feature amount of the video. More specifically, the video feature generation unit 108 detects, for each video, each extracted feature in text associated with the video. Then, the video feature generation unit 108 generates a video feature amount representing the feature appearing in the text associated with the video for each video. The text mining process described above represents, for example, extracting features as described above and detecting each of the extracted features. The video feature generation unit 108 may repeat generation of the video feature amount of the selected video while sequentially selecting videos until all of the plurality of videos are selected. Specifically, for example, the video feature generation unit 108 may select a video and detect a phrase (that is, a feature) extracted as a feature in text associated with the selected video.

  The video feature generation unit 108 generates a feature amount representing the detected feature as a video feature amount based on the result of detecting the feature. The video feature amount may be, for example, a set in which the detected feature is an element. The video feature amount may be expressed by a vector in which all extracted features are associated with any element, for example. In that case, in the following description, the video feature amount is also expressed as a video feature vector. The video feature generation unit 108 may associate the feature and the vector element so that the feature and the vector element are associated one-to-one, for example. The video feature generation unit 108 may associate the features with the elements of the video feature vector in the order of the features in the feature list, for example. The video feature amount is, for example, a value (for example, 1) indicating that the feature associated with the element has appeared, or a value (for example, 0) indicating that the element has not appeared. It may be a video feature vector.

  The video feature generation unit 108 may detect the appearance frequency of the feature for each feature in the text associated with the selected video. In that case, the video feature generation unit 108 may generate a feature quantity representing the appearance frequency of the feature detected for each feature as the video feature quantity. The video feature amount may be a set in which elements are combinations of detected features and appearance frequencies of the features. The video feature quantity may be a video feature vector whose element value is the appearance frequency of the feature associated with the element. In that case, the video feature generation unit 108 may normalize each video feature vector so that the size of the video feature vectors of a plurality of videos is constant.

  The size of the video feature vector is, for example, the length (that is, the square root of the sum of the squares of the values of the respective elements). The video feature generation unit 108 may normalize each video feature vector so that the size of the feature vector of each video is 1.

  The video feature generation unit 108 transmits the generated video feature amount of each video to the video selection device 110. The video feature generation unit 108 may associate the video ID with the video feature amount for each of the plurality of videos, and transmit the video ID and the video feature amount associated with each other to the video selection device 110. The video feature generation unit 108 further transmits the extracted features to the video selection device 110. The video feature generation unit 108 may read the feature list from the feature storage unit 107 and use the read feature list as the video selection device 110.

  The video feature receiving unit 116 of the video selection device 110 receives the video feature amount of each video from the video feature generating unit 108. The video feature generation unit 108 may receive the video ID and the video feature amount associated with each other from the video feature generation unit 108 for each of the plurality of videos. The video feature receiving unit 116 stores, in the video feature storage unit 117, the video ID and the video feature amount associated with each other of the received videos. The video feature receiving unit 116 further receives the extracted feature set, for example, as a feature list from the video feature generating unit 108, for example, and stores the received feature set (eg, feature list) in the video feature storage unit 117. Store.

  The video feature storage unit 117 stores a video ID and a video feature amount associated with each other of each of the plurality of videos. The video feature storage unit 117 further stores a set of features (for example, a feature list).

  For example, the target receiving unit 111 receives text specified by a user from a user terminal (not shown) used by the user. The designated text is, for example, text that causes the video selection device 110 to select a video corresponding to the text. In the following description, the specified text is expressed as target text. The target text is, for example, lyrics. The target text may be, for example, text representing content such as narration, reading, or announcement.

  The target feature generation unit 112 generates a target feature amount that is a feature amount of the target text by performing a text mining process on the target text. More specifically, the target feature generation unit 112 detects features included in a set of features (for example, a feature list) stored in the video feature storage unit 117, for example, in the target text. Then, the target feature generation unit 112 generates a target feature amount, which is a feature amount representing a feature that appears in the target text, based on the result of detecting the feature.

  When the video feature amount generated by the video feature generation unit 108 is a feature amount indicating whether or not a feature appears in the text associated with the video for each feature, the target feature generation unit 112 includes the target text in the target text. What is necessary is just to detect whether each feature appears. Then, the target feature generation unit 112 may generate a target feature amount that represents whether or not a feature appears for each feature. The target feature amount may be a set of detected features. The target feature amount is a value indicating that a feature associated with the element has appeared (for example, 1), or a value indicating that an element associated with the element has not appeared. It may be a vector (target feature vector) that is (for example, 0).

  When the video feature amount represents the appearance frequency of the feature for each feature, the target feature generation unit 112 may detect the appearance frequency of the feature for each feature in the target text. Then, the target feature generation unit 112 may generate a target feature amount representing the appearance frequency of each feature. The target feature amount may be, for example, a set in which an element is a combination of a feature and an appearance frequency of the feature. The target feature amount may be a vector (target feature vector) in which the value of each element represents the appearance frequency of a feature associated with the element. The target feature generation unit 112 may normalize the target feature vector. The target feature generation unit 112 may not normalize the target feature vector.

  The degree-of-similarity deriving unit 113 derives the degree of similarity, which is an index indicating the degree of similarity of the target feature amount with respect to the image feature amount of the image, for each of the plurality of images. That is, the similarity derived by the similarity deriving unit 113 is an index indicating how similar the video feature quantity and the target feature quantity are. In the following description, the fact that two feature quantities are similar is expressed as high similarity. The fact that the two feature quantities are not similar is expressed as low similarity. The degree to which two feature quantities are similar is expressed as high similarity. The similarity may be larger as the similarity is higher. The similarity may be smaller as the similarity is higher.

The similarity is, for example, a cosine similarity expressed by Equation 1. When the similarity is a cosine similarity, the similarity is higher as the similarity value is larger. In Expression 1 and the following expressions, “x” is an identifier representing multiplication. The vector q is a target feature vector, and q _i is the i-th element of the target feature vector. The vector d ^x is the video feature vector of the x th video, and d ^x _i is the i th element of the vector d ^x . The number of images is N (N is a natural number).

  The similarity may be, for example, a Euclidean distance expressed by Equation 2. When the similarity is the Euclidean distance, the similarity is higher as the similarity value is smaller. The similarity may be a reciprocal of the Euclidean distance. In this case, when the Euclidean distance is 0, the similarity deriving unit 113 may set, for example, the maximum numerical value that can be processed by the similarity deriving unit 113 as the similarity. When the similarity is the reciprocal of the Euclidean distance, the similarity is higher as the similarity value is larger. In Equation 2, “t” represents transposition of a vector, and “*” represents a product (inner product) of vectors. In Equation 2, each vector is a row vector.

The similarity may be a Jackard coefficient expressed by Equation 3. The Jackard coefficient is a value obtained by dividing the number of non-zero common elements of two feature vectors by the number of elements in which at least one element of those feature vectors is not zero. When the similarity is a Jackard coefficient, the similarity is higher as the similarity value is larger. In Equation 3, | Q∩D ^x | represents the number of common elements of vector q and vector d ^x whose values are not zero. | Q∪D ^x | This is the number of elements in which the value of at least one of the vector q and the vector d ^x is not zero. Q is, for example, a set of element numbers of vector q whose values are not 0. D ^x is, for example, a set of numbers of elements whose values are not 0 of the vector d ^x . “∩” represents a product set. “∪” represents a union.

Similarity = | Q∩D ^x | / | Q∪D ^x | (Formula 3)
The video selection unit 114 selects a video whose derived similarity is high. The video selection unit 114 may select a similarity indicating that the similarity is the highest from the similarities calculated for each of the plurality of videos. Then, the video selection unit 114 may select a video having the video feature amount used for deriving the selected similarity (that is, associated with the video feature amount).

  The target receiving unit 111 may receive an exclusion keyword in addition to the target text. An exclusion keyword is one or more words. In that case, the video selection unit 114 does not select a video including any of the exclusion keywords in the accompanying information regardless of the derived similarity value. In that case, for example, the video storage unit 119 may store the same accompanying information as the accompanying information stored in the accompanying information storage unit 102. For example, the video reception unit 118 may receive the accompanying information from a video server (not shown) that stores a plurality of videos, a user terminal (not shown), the feature generation device 100, or the like. The video receiving unit 118 may store the received accompanying information in the video storage unit 119.

  The output unit 115 outputs data representing the result of selection by the video selection unit 114. The data representing the selection result is, for example, the video ID of the selected video.

  As described above, the video selection device 110 may include the video reception unit 118 and the video storage unit 119.

  The video reception unit 118 receives video data of the plurality of videos from a video server (not shown) that stores the plurality of videos. Then, the video receiving unit 118 stores the received video data in the video storage unit 119. The video storage unit 119 stores video data of a plurality of videos stored by the video receiving unit 118.

  When the video storage unit 119 stores a plurality of video data, the output unit 115 may output the video data of the video selected by the video selection unit 114.

  Next, the operation of the video selection system 1 of the present embodiment will be described in detail with reference to the drawings.

  FIG. 3 is a flowchart showing an example of an operation of extracting features based on a plurality of texts and accompanying information in the feature generation device 100.

  First, the teacher data receiving unit 103 receives one or more texts associated with a video for each of a plurality of videos (step S101). One or more texts associated with the video are also described as teacher data. The teacher data receiving unit 103 stores the received teacher data in the teacher data storage unit 104 (step S102).

  Next, the accompanying information receiving unit 101 receives the accompanying information (step S103). FIG. 2 shows an example of the accompanying information received by the attached information receiving unit 101. The attached information receiving unit 101 receives the accompanying information illustrated in FIG. 2 for each of a plurality of videos. The accompanying information receiving unit 101 stores the received accompanying information in the accompanying information storage unit 102 (step S104).

  The operations in steps S103 and S104 may be performed before the operations in steps S101 and S102. The operations in step S103 and step S104 may be performed in parallel with the operations in step S101 and step S102.

  Next, the attribute extraction unit 105 extracts attributes from the accompanying information stored in the accompanying information storage unit 102 (step S105).

  Next, the feature extraction unit 106 extracts a word or phrase that modifies the attribute extracted by the attribute extraction unit 105 from the learning data as a feature (step S106).

  FIG. 4 is a diagram schematically illustrating an example of extracted features. In the example shown in FIG. 4, video 1 and text A are associated. Video 2 and text B are associated with each other. Video 3 and text C are associated with each other. The phrases in the right block schematically represent part of the text. In these blocks, words enclosed in parentheses are attributes included in the text among the attributes extracted by the attribute detection unit 105. An underlined phrase is a feature extracted as a phrase related to an attribute. For example, the text A includes “spring” and “male” as attributes. The feature extraction unit 106 extracts “warm” as the feature related to “spring”. The feature extraction unit 106 further extracts “tall” as the feature related to “male”.

  FIG. 5 is a diagram schematically illustrating an example of extracted features. In FIG. 5, “attribute” is an attribute detected in any text. In FIG. 5, “feature” is a feature extracted as a phrase related to a detected attribute.

  Next, the feature extraction unit 106 generates a feature list that is a list of extracted features (step S107).

  FIG. 6 is a diagram schematically illustrating an example of a feature list. In FIG. 6, “attribute” is an attribute detected in any text. In FIG. 6, “feature” is a feature extracted as a phrase related to a detected attribute. In FIG. 6, a part surrounded by a thick line represents a feature list.

  Next, the feature extraction unit 106 stores the generated feature list in the feature storage unit 107 (step S108).

  Next, an operation for generating video features of the feature generation device 100 according to the present embodiment will be described in detail with reference to the drawings.

  FIG. 7 is a flowchart illustrating an example of an operation of generating a video feature of the feature generation device 100 according to the present embodiment.

  Referring to FIG. 7, the video feature generation unit 108 reads a feature list that is a list of features extracted by the feature extraction unit 106 from the feature storage unit 107. (Step S111).

  Next, the video feature generation unit 108 identifies text associated with the video for each video (step S112).

  FIG. 8 is a diagram schematically illustrating an example of text associated with a video for each video. As described above, the video ID is an identifier for specifying a video. In FIG. 8, “text” represents an identifier of one or more texts associated with each video specified by the video ID.

  For each video, the video feature generation unit 108 detects a feature included in the feature list in the text associated with the video (step S113). The video feature generation unit 108 generates, for each video, a video feature amount (for example, a video feature vector) representing a feature appearing in the text associated with the video based on the result of detecting the feature (step S114).

  FIG. 9 is a diagram schematically illustrating an example of a video feature vector. In FIG. 9, for example, a portion surrounded by a rectangle drawn by a thick line represents a video feature vector of video 1. Numerical value columns shown in two stages below the feature vector of the video 1 represent the video feature vectors of the video 2 and the video 3. The value of each element of the video feature vector shown in FIG. 9 represents the appearance frequency of the feature in the text associated with the video. In the example shown in FIG. 9, each video feature vector is not normalized.

  The video feature generation unit 108 may further normalize each video feature vector so that the video feature vectors of each video have the same size. For example, the similarity deriving unit 113 of the video selection device 110 may normalize each video feature vector instead of the video feature generation unit 108.

  The video feature generation unit 108 transmits the video feature vector generated for each video and the feature list to the video selection device 110 (step S115).

  Next, the operation of the video selection device 110 of this embodiment will be described in detail with reference to the drawings. First, an operation of receiving a video feature vector of the video selection device 110 according to the present embodiment will be described.

  FIG. 10 is a flowchart illustrating an example of an operation of receiving a video feature vector of the video selection device 110 according to the present embodiment.

  Referring to FIG. 10, the video feature receiving unit 116 receives a video feature amount (for example, a video feature vector) and a feature list from the video feature generating unit 108 of the feature generating device 100 (step S201). The video feature receiving unit 116 stores the received video feature quantity and feature list in the video feature storage unit 117 (step S202).

  Next, an operation of selecting a video in response to receiving the target text of the video selection device 110 of the present embodiment will be described.

  FIG. 11 is a flowchart showing an example of an operation of selecting a video in response to receiving the target text in the video selection device 110 of the present embodiment.

  Referring to FIG. 11, first, the target receiving unit 111 receives a target text (step S211). For example, the target receiving unit 111 may receive target text representing the content of audio content from a content distribution server, for example.

  Next, the target feature generation unit 112 extracts features included in the feature list, which is a list of features extracted by the feature extraction unit 106, from the target text received by the target reception unit 111 (step S212). As described above, the feature list is stored in the video feature storage unit 117 by the video feature receiving unit 116, for example.

  The target feature generation unit 112 generates a target feature amount (for example, a target feature vector) that represents a feature that appears in the target text based on the result of extracting the features (step S213).

  FIG. 12 is a diagram schematically illustrating an example of the target feature vector. A numerical string shown in FIG. 12 represents a countermeasure feature vector. The target feature vector shown in FIG. 12 represents the appearance frequency of each feature included in the feature list in the target text.

  Next, the similarity degree deriving unit 113 calculates a degree of similarity representing the degree of similarity of the target feature amount with respect to each of the video feature amounts (step S214).

  FIG. 13 is a diagram schematically illustrating examples of target feature vectors and video feature vectors from which the similarity deriving unit 113 derives similarities. In the example illustrated in FIG. 13, the target feature amount is a target feature vector. The video feature amount is a video feature vector. In the example shown in FIG. 13, the target feature vector and each video feature vector are normalized. As described above, the target feature vector may not be normalized. The similarity degree deriving unit 113 derives the degree of similarity representing the degree of similarity between the target feature vector and the video feature vector for each of the video feature vectors.

  FIG. 14 is a diagram schematically illustrating an example of similarity. FIG. 14 shows the similarity indicating the level of similarity between the video feature vector of each video shown in FIG. 13 and the target feature vector. In the example shown in FIG. 14, the similarity is a cosine similarity. Therefore, the greater the similarity value, the higher the similarity.

  The video selection unit 114 selects a video feature amount similar to the target feature amount using the calculated similarity (step S215). The video selection unit 114 may select a video feature amount indicating that the similarity is the highest. In the example shown in FIG. 14, the video feature amount having the largest similarity value is the video feature amount most similar to the target feature amount. The video feature quantity of video 3 is most similar to the target feature quantity. That is, in the example illustrated in FIG. 14, the similarity between the video feature quantity of the video 3 and the target feature quantity is the highest.

  The video selection unit 114 selects a video related to the selected video feature amount (step S216). In the example illustrated in FIG. 14, the selected video related to the video feature amount of the video 3 is the video 3. The video selection unit 114 selects the video 3.

  The output unit 115 outputs the result of selection by the video selection unit 114 (step S217). The output unit 115 may output, for example, an identifier of the selected video (that is, a video ID) as a result of selection by the video selection unit 114. In the example illustrated in FIG. 14, the output unit 115 may output the video ID of the video 3 that is the selected video. For example, the output unit 115 may output the video ID of the selected video to the device that has transmitted the target text to the target receiving unit 111. When the target receiving unit 111 receives the target text from a content distribution server (not shown), the output unit 115 may transmit the video ID of the selected video to the content distribution server. The content distribution server receives the video ID from the output unit 115.

  For example, when the content distribution server is a karaoke server that distributes karaoke music and video, the content distribution server receives the target text that is the lyrics of the music in response to receiving a request for music distribution, What is necessary is just to transmit to the image | video selection apparatus 110. The video selection device 110 transmits the video ID selected for the transmitted target text to the content distribution server. The content distribution server specifies the video represented by the received video ID. Then, the content distribution server may distribute the music whose target text is lyrics and the video represented by the received video ID, for example, to the terminal that requested the distribution of the music. The content distribution server may transmit in advance lyrics to the video selection device 110 as a target text for a plurality of music pieces that can be distributed. The content distribution server may receive the video ID of the selected video in advance. The content distribution server may store in advance a song ID that is an identifier of a song and a video ID of a video that is selected when the lyrics of the song are the target text in a storage area (not shown). Good. Then, in response to a request for music distribution, the content distribution server may read the video ID of the video selected for the lyrics of the music requested to be distributed. Then, the content distribution server may distribute the music requested to be distributed and the video represented by the read video ID to the terminal that requested the distribution of the music.

  The content distribution server may be a content distribution server that distributes audio content and video in response to a request for audio content such as narration, reading, or announcement. In this case, the content distribution server may transmit, for example, text data representing the contents of distributable audio content as the target text to the video selection device 110. The content distribution server receives the video ID from the video selection device 110, and transmits the audio content whose content is represented by the target text and the video represented by the received video ID to the terminal that requested the distribution of the audio content. do it. The content distribution server may transmit the target text, receive the video ID, and distribute the audio content and the video in response to receiving the content distribution request. The content distribution server transmits the target text and receives the video ID in advance, and stores the content ID of the content represented by the target text and the video ID selected for the target text. It may be. In response to receiving the request for distributing the audio content, the audio content and the video selected for the target text representing the content of the audio content specified based on the stored video ID, You may transmit to the terminal which requested | required delivery of audio | voice content.

  The present embodiment described above has a first effect that the load for matching text and video can be reduced.

  The reason is that the video selection unit 114 performs matching between the text and the video based on the degree of similarity between the video feature amount representing the video feature and the target feature amount representing the text feature. is there. The video feature amount is generated by the video feature generation unit 108 using text pre-associated with the video and accompanying information of the video. The target feature amount is the same type of feature amount as the video feature amount. Then, the target feature amount is generated by the target feature generation unit 112 using the text that is the target of matching (the above-described countermeasure text).

  In the present embodiment, there is an effect that the video can be selected for the text so that the content of the text and the content of the video are accurately matched.

  The reason is that the video feature generation unit 108 and the target feature generation unit 112 use the features extracted by the feature extraction unit 106 to generate feature amounts. As described above, it is experienced that a feature that is a word that modifies a word (the above-described attribute) that expresses the content of a video is a visual, bodily sensation, or a visual bodily sensation. Has been revealed. The text pre-associated with the video is, for example, text determined by the user when it matches the video visually, bodily, or visually and bodily. Therefore, when the above-mentioned feature appearing in the text associated with the video is similar to the feature appearing in the target text, the video and the target text are visually, bodily, or visually There is a high possibility that the match will be intuitive. The video feature generation unit 108 generates a video feature amount representing a feature that appears in text associated with a video. The target feature generation unit 112 generates a target feature amount representing a feature appearing in the target text. The similarity degree deriving unit 113 derives such a degree of similarity representing the degree of similarity between the video feature quantity and the target feature quantity. Using the similarity, the video selection unit 114 selects a video for the target text so that the video feature amount and the target feature amount are similar to each other. Accordingly, the video selection unit 114 can accurately select a video that visually, bodily, or visually matches the target text.

The second effect described above will be described in more detail.

  As described above, the feature extraction unit 106 extracts a phrase that modifies an attribute that is a word extracted from the accompanying information, thereby extracting a phrase that is expressed visually and bodily as a feature. The target feature generation unit 112 generates a target feature amount (for example, a target feature vector) based on the appearance frequency of features in the target text such as lyrics. The video feature generation unit 108 generates a video feature vector based on the appearance frequency of features in text such as lyrics associated with the video and accompanying information (particularly planning intention) of the video. The similarity degree deriving unit 113 calculates a degree of similarity, such as a cosine similarity degree, for example, indicating the level of similarity that is the degree to which the target feature vector and the video feature vector are similar. The video selection unit 114 compares the target feature vector with the video feature vector by using a similarity such as a cosine similarity, and determines the relevance representing the depth of the relationship between the target text and the video. The above-described determination based on the similarity between the target feature vector and the video feature vector is a determination of the relationship between the target text and the video using visual and physical characteristics. Therefore, by using such similarity and selecting a video whose target feature vector and video feature vector are similar to the target text, a video similar to the target text is visually and physically experienced. You can expect to be chosen.

  For example, in the feature extraction, it is expected that visual and bodily words such as “blue” and “sunny” will be extracted as features by extracting words related to the attribute “sky”. it can. When the appearance frequency of such a feature in the text associated with the video is derived, for example, in the text associated with the video with the characteristic of “blue sky”, such as “blue” and “sunny” It can be expected that the number of words having visually the same characteristics is high. Furthermore, when the target text includes the expression “blue sky” or “sunny sky”, the target text and the video with a high frequency of appearance of at least one of the words “blue” and the term “sunny” The degree of similarity is increased.

  As described above, in the video selection system 1 according to the present embodiment, for example, when the target text has a visual and bodily expression such as “blue sky” or “harsh winter”, the target text is You can match images with features such as “clear sky” and “cold season”. Thus, the video selection system 1 of the present embodiment can perform matching that cannot be realized by matching using synonyms.

  That is, the video selection system 1 of the present embodiment can select an appropriate video that is visually uncomfortable with respect to the text.

<First Modification of First Embodiment>
Next, a first modification of the first embodiment of the present invention will be described in detail with reference to the drawings.

  FIG. 15 is a block diagram illustrating an example of the configuration of the video selection system 1A according to the present modification. Comparing FIG. 15 with FIG. 1, the video selection system 1 </ b> A of the present modification includes a video selection device 110 </ b> A instead of the video selection device 110. The video selection device 110A includes a teacher data reception unit 121, a teacher data storage unit 122, a teacher data transmission unit 123, and an accompanying information reception unit 124, in addition to the components of the video selection device 110 of the first embodiment. And an accompanying information storage unit 125 and an accompanying information transmission unit 126. The video selection system 1A may include a user terminal (not shown) that is used by the user of the video selection system 1A to input instructions and the like.

  The teacher data receiving unit 121 receives the above-described teacher data (a plurality of texts associated with each video) from another device such as a content distribution server. The teacher data receiving unit 121 stores the received teacher data in the teacher data storage unit 122. The teacher data storage unit 122 stores teacher data. The teacher data transmission unit 123 transmits the teacher data stored in the teacher data storage unit 122 to the teacher data reception unit 103. The teacher data receiving unit 103 receives teacher data from the teacher data transmitting unit 123.

  The accompanying information receiving unit 124 receives the accompanying information described above from another device such as a content distribution server. The accompanying information receiving unit 124 stores the received accompanying information in the accompanying information storage unit 125. The accompanying information storage unit 125 stores accompanying information. The accompanying information transmitting unit 126 transmits the accompanying information stored in the accompanying information storage unit 125 to the accompanying information receiving unit 101. The accompanying information receiving unit 101 receives accompanying information from the accompanying information transmitting unit 126.

  Except for the above differences, the video selection system 1A of the present modification is the same as the video selection system 1 of the first embodiment.

<Second Modification of First Embodiment>
Next, a second modification of the first embodiment of the present invention will be described in detail with reference to the drawings.

  FIG. 16 is a block diagram illustrating an example of a configuration of a video selection system 1B according to the present modification. Referring to FIG. 16, the video selection system 1B includes a video selection device 110B. The video selection system 1B includes a feature generation device 100. The video selection system 1B may include a user terminal (not shown) that is used by the user of the video selection system 1B to input instructions and the like.

  Except for the above differences, the video selection system 1B of the present modification is the same as the video selection system 1 of the first embodiment.

<Third Modification of First Embodiment>
Next, a third modification of the first embodiment of the present invention will be described in detail with reference to the drawings.

  FIG. 17 is a block diagram illustrating an example of a configuration of a video selection system 1C according to the present modification. The video selection system 1C of the present modification includes a video selection device 110C. The video selection device 110C includes each component of the feature generation device 100 of the first embodiment. The video selection device 110 </ b> C may not include the video feature receiving unit 116. Then, the video selection device 110C operates as the feature generation device 100 of the first embodiment. The video selection system 1C may include a user terminal (not shown) that is used by the user of the video selection system 1C to input instructions and the like.

  The video feature generation unit 108 of the present embodiment stores the generated video feature amount in the video feature storage unit 117. The video feature generation unit 108 of the present embodiment may store the above-described feature list in the video feature storage unit 117. The feature extraction unit 106 according to the present embodiment may transmit the feature list to the target feature generation unit 112. The target feature generation unit 112 of this embodiment may read the feature list from the feature storage unit 107.

  Except for the above differences, the video selection system 1C of the present modification is the same as the video selection system 1 of the first embodiment.

<Second Embodiment>
Next, a second embodiment of the present invention will be described in detail with reference to the drawings. This embodiment is an embodiment conceptually representing each embodiment of the present invention.

  FIG. 18 is a block diagram illustrating an example of the configuration of the video selection system 1D of the present embodiment.

  Referring to FIG. 18, the video selection system 110D according to the present embodiment performs a text mining process on text associated with a video, thereby obtaining a video feature quantity that is the video feature quantity. A video feature generation unit 108 that generates each of the videos; a target feature generation unit 112 that generates a target feature amount that is a feature amount of the target text by executing the text mining process on the target text; For each video feature quantity, a similarity degree deriving unit 113 that derives a degree of similarity of the video feature quantity with respect to the target feature quantity, and the target feature quantity based on the derived similarity degree The video feature amount having a high degree of similarity is selected, and the selected video feature amount is associated with the derived text. Comprises a video selection unit 114 for selecting the image are, the.

  The present embodiment described above has the same effect as the first effect of the first embodiment. The reason is the same as the reason why the first effect of the first embodiment occurs.

<Other embodiments>
The video selection device 110, the video selection device 110A, the video selection device 110B, the video selection device 110C, the video selection device 110D, the feature generation device 100, and the feature generation device 100D are respectively a computer, a program for controlling the computer, and a dedicated program. It can be realized by hardware or a combination of a computer and a program for controlling the computer and dedicated hardware.

  FIG. 19 illustrates a computer 1000 capable of realizing the video selection device 110, the video selection device 110A, the video selection device 110B, the video selection device 110C, the video selection device 110D, the feature generation device 100, and the feature generation device 100D. It is a figure showing an example of a hardware constitutions. Referring to FIG. 19, a computer 1000 includes a processor 1001, a memory 1002, a storage device 1003, and an I / O (Input / Output) interface 1004. The computer 1000 can access the recording medium 1005. The memory 1002 and the storage device 1003 are storage devices such as a RAM (Random Access Memory) and a hard disk, for example. The recording medium 1005 is, for example, a storage device such as a RAM or a hard disk, a ROM (Read Only Memory), or a portable recording medium. The storage device 1003 may be the recording medium 1005. The processor 1001 can read and write data and programs from and to the memory 1002 and the storage device 1003. The processor 1001 can access, for example, other devices via the I / O interface 1004. The processor 1001 can access the recording medium 1005. The recording medium 1005 is a program that causes the computer 1000 to operate as the video selection device 110, the video selection device 110A, the video selection device 110B, the video selection device 110C, the video selection device 110D, the feature generation device 100, or the feature generation device 100D. Is stored.

  The processor 1001 stores the computer 1000 stored in the recording medium 1005 as a video selection device 110, a video selection device 110A, a video selection device 110B, a video selection device 110C, a video selection device 110D, a feature generation device 100, or a feature. A program to be operated as the generation device 100D is loaded into the memory 1002. Then, when the processor 1001 executes the program loaded in the memory 1002, the computer 1000 causes the video selection device 110, the video selection device 110A, the video selection device 110B, the video selection device 110C, the video selection device 110D, and the feature generation. It operates as the device 100 or the feature generation device 100D.

  Accompanying information reception unit 101, teacher data reception unit 103, attribute extraction unit 105, feature extraction unit 106, video feature generation unit 108, target reception unit 111, target feature generation unit 112, similarity derivation unit 113, video selection unit 114, The output unit 115, the video feature receiving unit 116, the video receiving unit 118, the teacher data receiving unit 121, the teacher data transmitting unit 123, the accompanying information receiving unit 124, and the accompanying information transmitting unit 126 are, for example, recording media for storing programs It can be realized by a dedicated program that can be read from the memory 1002 to the memory 1002 and that can realize the function of each unit, and a processor 1001 that executes the program. The accompanying information storage unit 102, the teacher data storage unit 104, the feature storage unit 107, the video feature storage unit 117, the video storage unit 119, the teacher data storage unit 122, and the accompanying information storage unit 125 are included in the memory included in the computer 1000. 1002 and a storage device 1003 such as a hard disk device. Alternatively, the accompanying information receiving unit 101, the accompanying information storing unit 102, the teacher data receiving unit 103, the teacher data storing unit 104, the attribute extracting unit 105, the feature extracting unit 106, the feature storing unit 107, the video feature generating unit 108, and the target receiving unit 111, target feature generation unit 112, similarity derivation unit 113, video selection unit 114, output unit 115, video feature reception unit 116 video feature storage unit 117, video reception unit 118, video storage unit 119, teacher data reception unit 121, A part or all of the teacher data storage unit 122, the teacher data transmission unit 123, the incidental information reception unit 124, the incidental information storage unit 125, and the incidental information transmission unit 126 is realized by a dedicated circuit that realizes the function of each unit. You can also.

  Moreover, although a part or all of said embodiment can be described also as the following additional remarks, it is not restricted to the following.

(Appendix 1)
Video feature generation means for generating, for each of the plurality of videos, a video feature quantity that is a feature quantity of the video by performing a text mining process on text associated with the video;
Target feature generation means for generating a target feature quantity that is a feature quantity of the target text by executing the text mining process on the target text;
Similarity degree deriving means for deriving the degree of similarity of the video feature quantity with respect to the target feature quantity for each of the video feature quantities;
Based on the derived degree of similarity, the video feature amount having a high degree of similarity to the target feature amount is selected, and the video associated with the text from which the selected video feature amount is derived is selected. Video selection means to
A video selection system comprising:

(Appendix 2)
A feature extracting unit that extracts a feature that is a phrase that modifies the attribute of the video from the text based on an attribute that is a phrase related to at least one of the plurality of videos;
For each of the plurality of videos, the video feature generation means detects each of the extracted features in the text associated with the video, and calculates a feature amount representing the detected features. Generating as the video feature amount,
The video selection system according to claim 1, wherein the target feature generation unit detects each of the extracted features in the target text, and generates a feature amount representing the detected feature as the target feature amount.

(Appendix 3)
The video selection system according to claim 2, further comprising attribute extraction means for extracting the attribute from accompanying information that is associated with each of the videos and is information that characterizes the video.

(Appendix 4)
Based on an attribute that is a phrase related to at least one of a plurality of videos each associated with text, a feature that is a phrase that modifies the attribute is extracted from the text, and the extracted feature is stored as a feature A feature extraction means stored in the means;
For each of the plurality of videos, in the text associated with the video, each of the features stored in the feature storage means is detected, and a feature amount representing the detected feature is detected as the video of the video. Video feature generation means for generating as feature quantities;
A feature amount generating apparatus.

(Appendix 5)
The feature quantity generation device according to appendix 4, further comprising attribute extraction means for extracting the attribute from accompanying information that is associated with each of the videos and is information that characterizes the videos.

(Appendix 6)
Generating a video feature amount that is a feature amount of the video for each of the plurality of videos by performing a text mining process on text associated with the video;
By executing the text mining process on the target text, a target feature amount that is a feature amount of the target text is generated,
For each of the video feature quantities, derive a similarity indicating the degree of similarity of the video feature quantity to the target feature quantity,
Based on the derived degree of similarity, the video feature amount having a high degree of similarity to the target feature amount is selected, and the video associated with the text from which the selected video feature amount is derived is selected. To
Video selection method.

(Appendix 7)
Based on an attribute that is a phrase related to at least one of the plurality of videos, a feature that is a phrase that modifies the attribute of the video is extracted from the text;
For each of the plurality of videos, in the text associated with the video, each of the extracted features is detected, and a feature amount representing the detected feature is generated as the video feature amount of the video. ,
The video selection method according to claim 6, wherein each of the extracted features is detected in the target text, and a feature amount representing the detected feature is generated as the target feature amount.

(Appendix 8)
The video selection method according to claim 7, wherein the attribute is extracted from accompanying information that is associated with each of the videos and is information that characterizes the video.

(Appendix 9)
Based on an attribute that is a phrase related to at least one of a plurality of videos each associated with text, a feature that is a phrase that modifies the attribute is extracted from the text, and the extracted feature is stored as a feature Stored in the means,
For each of the plurality of videos, in the text associated with the video, each of the features stored in the feature storage means is detected, and a feature amount representing the detected feature is detected as the video of the video. Generate as feature quantity,
Feature generation method.

(Appendix 10)
The feature value generation method according to claim 9, wherein the attribute is extracted from accompanying information that is associated with each of the videos and is information that characterizes the video.

(Appendix 11)
Computer
Video feature generation means for generating, for each of the plurality of videos, a video feature quantity that is a feature quantity of the video by performing a text mining process on text associated with the video;
Target feature generation means for generating a target feature quantity that is a feature quantity of the target text by executing the text mining process on the target text;
Similarity degree deriving means for deriving the degree of similarity of the video feature quantity with respect to the target feature quantity for each of the video feature quantities;
Based on the derived degree of similarity, the video feature amount having a high degree of similarity to the target feature amount is selected, and the video associated with the text from which the selected video feature amount is derived is selected. Video selection means to
Video selection program to be operated.

(Appendix 12)
Computer
A feature extraction unit that extracts a feature that is a phrase that modifies the attribute of the video from the text based on an attribute that is a phrase related to at least one of the plurality of videos;
For each of the plurality of videos, in the text associated with the video, each of the extracted features is detected, and a feature amount representing the detected feature is generated as the video feature amount of the video. The video feature generating means;
The target feature generating means for detecting each of the extracted features in the target text and generating a feature amount representing the detected feature as the target feature amount;
The video selection program according to appendix 11, which is operated as described above.

(Appendix 13)
Computer
13. The video selection program according to appendix 12, wherein the video selection program is operated as attribute extraction means for extracting the attribute from accompanying information that is associated with each of the videos and is information that characterizes the video.

(Appendix 14)
Computer
Based on an attribute that is a phrase related to at least one of a plurality of videos each associated with text, a feature that is a phrase that modifies the attribute is extracted from the text, and the extracted feature is stored as a feature A feature extraction means stored in the means;
For each of the plurality of videos, in the text associated with the video, each of the features stored in the feature storage means is detected, and a feature amount representing the detected feature is detected as the video of the video. Video feature generation means for generating as feature quantities;
A feature generation program to be operated.

(Appendix 15)
Computer
15. The feature quantity generation program according to appendix 14, which is operated as attribute extraction means for extracting the attribute from accompanying information that is associated with each of the videos and is information that characterizes the videos.

  The present invention has been described above with reference to the embodiments, but the present invention is not limited to the above embodiments. Various changes that can be understood by those skilled in the art can be made to the configuration and details of the present invention within the scope of the present invention.

DESCRIPTION OF SYMBOLS 1 Image | video selection system 1A Image | video selection system 1B Image | video selection system 1C Image | video selection system 1D Image | video selection system 100 Feature generation apparatus 100D Feature generation apparatus 101 Accompanying information receiving part 102 Accompanying information storage part 103 Teacher data receiving part 104 Teacher data storage part 105 Attribute Extraction unit 106 Feature extraction unit 107 Feature storage unit 108 Video feature generation unit 110 Video selection device 110A Video selection device 110B Video selection device 110C Video selection device 110D Video selection device 111 Target reception unit 112 Target feature generation unit 113 Similarity derivation unit 114 Video selection unit 115 Output unit 116 Video feature reception unit 117 Video feature storage unit 118 Video reception unit 119 Video storage unit 121 Teacher data reception unit 122 Teacher data storage unit 123 Teacher data transmission unit 124 Accompanying information reception Part 125 accompanying information storing unit 126 accompanying information transmitting unit 1000 computer 1001 processor 1002 memory 1003 storage 1004 I / O interface 1005 recording medium

Claims (15)

Feature extraction means for extracting a feature that is a phrase that modifies the attribute of the video from text associated with the video having the attribute based on an attribute that is a phrase related to at least one of the plurality of videos; ,
Oite the text associated with the image, extracted by detecting each of the feature, the feature value representing the said detected feature, image feature generating means for generating a video image characteristic amount is a feature amount of the image When,
Detecting each of said feature that Oite, extracted into target text, the object characteristic generating means for generating a target characteristic quantity is a feature quantity representing the said detected feature, a feature quantity of the target text,
Similarity degree deriving means for deriving the degree of similarity of the video feature quantity with respect to the target feature quantity for each of the video feature quantities;
Based on the derived degree of similarity, the video feature amount having a high degree of similarity to the target feature amount is selected, and the video associated with the text from which the selected video feature amount is derived is selected. Video selection means to
A video selection system comprising: Before Symbol image feature generating means, for each of the plurality of images to generate the image feature
Image selection system according to 請 Motomeko 1. The video selection system according to claim 2, further comprising attribute extraction means for extracting the attribute from accompanying information that is associated with each of the videos and is information that characterizes the video. Based on an attribute that is a phrase related to at least one of a plurality of videos each associated with text, a feature that is a phrase that modifies the attribute is extracted from the text, and the extracted feature is stored as a feature A feature extraction means stored in the means;
For each of the plurality of videos, in the text associated with the video, each of the features stored in the feature storage means is detected, and a feature amount representing the detected feature is detected as the video of the video. Video feature generation means for generating as feature quantities;
A feature amount generating apparatus. The feature quantity generation device according to claim 4, further comprising attribute extraction means that extracts the attribute from accompanying information that is associated with each of the videos and is information that characterizes the videos. Computer
Based on an attribute that is a phrase related to at least one of a plurality of videos, a feature that is a phrase that modifies the attribute of the video is extracted from text associated with the video having the attribute,
Oite the text associated with the image, extracted by detecting each of the feature, the feature value representing the said detected feature, generates as image feature is a feature value of the image,
Oite the target text, extracted detects each of the feature, the feature value representing the said detected feature, generates a target feature quantity is a feature quantity of the target text,
For each of the video feature quantities, derive a similarity indicating the degree of similarity of the video feature quantity to the target feature quantity,
Based on the derived degree of similarity, the video feature amount having a high degree of similarity to the target feature amount is selected, and the video associated with the text from which the selected video feature amount is derived is selected. To
Video selection method. Computer
Generating the video feature for each of the plurality of videos;
Video selection method according to 請 Motomeko 6. Computer
The video selection method according to claim 7, wherein the attribute is extracted from accompanying information that is associated with each of the videos and is information that characterizes the video. Computer
Based on an attribute that is a phrase related to at least one of a plurality of videos each associated with text, a feature that is a phrase that modifies the attribute is extracted from the text, and the extracted feature is stored as a feature Stored in the means,
For each of the plurality of videos, in the text associated with the video, each of the features stored in the feature storage means is detected, and a feature amount representing the detected feature is detected as the video of the video. Generate as feature quantity,
Feature generation method. Computer
The feature value generation method according to claim 9, wherein the attribute is extracted from accompanying information that is associated with each of the videos and is information that characterizes the video. On the computer,
A feature extraction process for extracting a feature that is a phrase that modifies the attribute of the video based on an attribute that is a phrase related to at least one of the plurality of videos from text associated with the video having the attribute; ,
Oite the text associated with the image, extracted by detecting each of the feature, the feature value representing the said detected feature, image feature generation process of generating the image feature is a feature value of the image When,
Oite the target text, extracted detects each of the feature, the feature value representing the said detected feature, and the object characteristic generation process of generating a target characteristic quantity is a feature quantity of the target text,
For each of the video feature amounts, a similarity degree derivation process for deriving a degree of similarity representing the degree of similarity of the video feature amount to the target feature amount;
Based on the derived degree of similarity, the video feature amount having a high degree of similarity to the target feature amount is selected, and the video associated with the text from which the selected video feature amount is derived is selected. Video selection processing to
A video selection program that executes The video feature generation processing generates the video feature amount for each of the plurality of videos.
Video selection program according to 請 Motomeko 11. On the computer,
Attribute extraction processing for extracting the attribute from accompanying information that is associated with each of the videos and is information that characterizes the videos
The video selection program according to claim 12, wherein: On the computer,
Based on an attribute that is a phrase related to at least one of a plurality of videos each associated with text, a feature that is a phrase that modifies the attribute is extracted from the text, and the extracted feature is stored as a feature A feature extraction process stored in the means;
For each of the plurality of videos, in the text associated with the video, each of the features stored in the feature storage means is detected, and a feature amount representing the detected feature is detected as the video of the video. Video feature generation processing to generate as a feature amount;
A feature generation program that executes On the computer,
Attribute extraction processing for extracting the attribute from accompanying information that is associated with each of the videos and is information that characterizes the videos
15. The feature quantity generation program according to claim 14, wherein:

Images