JP5106455B2 - Content recommendation device and content recommendation method - Google Patents

Description

  The present invention relates to a content recommendation device and a content recommendation method for recommending content.

  With the increase in capacity of recording equipment and the increase in the number of broadcast channels, broadcast programs that can be viewed by users are increasing and diversifying. Therefore, the user needs to select which program is to be viewed / recorded from a large number of programs.

  However, it is difficult to select a program that the user wants to watch from a large number of programs. For example, EPG (Electronic Program Guide) is displayed in the program guide as the contents of the program, but it is not possible to see all of them.

  On the other hand, there is a technique for recommending other programs based on the metadata of a specific scene of a currently viewed program (see Non-Patent Document 1, for example).

  However, if a recommended program is searched based only on the metadata of the program that the user is viewing, it is not possible to make a recommendation using a new vocabulary that is not present in the program. In addition, when the vocabulary is acquired from the entire program, a program unrelated to the scene that the user is currently viewing may be recommended.

  In addition, not all recommended programs can be viewed on-air, but they are often reserved and recorded when there is time.

  When a program is recorded, a program title, broadcast date and time, thumbnail images representing contents, EPG, etc. are often displayed as a recorded program list.

  However, simply viewing the recorded program list after recording the program will not allow you to remember what you wanted to see when you reserved the recording.

  Also, even if you remember what you wanted to see, it is difficult to find the location of the content in the recorded program and play it from the right place.

http://www.nhk.or.jp/strl/publica/giken_dayori/jp2/mt-0901.html (Hideki Sumiyoshi, “Metadata Production and Utilization Technology for Video Media”, 6th “Information Search” New viewing style CurioView that uses ”, NHK Science & Technology Research Laboratories, NHK Giken News, No. 46, January 2009)

  Conventionally, while recommending content such as a wide range of programs scheduled to be broadcast while related to the scene that the user was interested in while viewing the content, the portion that the user would have wanted to watch at the time of recording reservation after recording the content It was difficult to make it easy to watch.

  The present invention has been made in consideration of the above circumstances, and recommends content such as a wide range of broadcast-scheduled programs related to scenes that the user is interested in while viewing the content, and after recording the content, It is an object of the present invention to provide a content recommendation device and a content recommendation method that allow a user to easily view a portion that the user would have wanted to view at the time of recording reservation.

  The content recommendation device according to the present invention includes, for each of the reservation-recorded contents, metadata including information related to the contents of the contents, a first feature keyword that characterizes the contents, and a plurality of pieces that constitute the contents A first storage unit that stores content information including information on scenes, a second feature keyword that characterizes each scene, and a reason why the content is reserved for recording, and a scene that constitutes the content being played When a specific scene is selected, a third characteristic keyword related to the specific scene is acquired from the content information related to the content being reproduced stored in the first storage unit. Information on a plurality of net moving images acquired using the first acquisition unit and the third characteristic keyword A second acquisition unit that acquires an extended keyword obtained by expanding the third feature keyword and acquires at least an evaluation value for the net moving image as an attribute of the extended keyword; A first selection unit that presents a list and accepts user selection for any of the presented extended keywords, and a list for recommending content that can be reserved for recording acquired using the user-selected extended keywords A second selection unit that presents and receives a user selection for the presented arbitrary content; a second storage unit that stores a reservation list in which information for recording reservation of the user-selected content is registered; For content that has already been recorded based on the reservation list, use the content information. Te is a presentation about the scene of the content, characterized in that a presentation unit that performs a presentation of why the recording reservation was made including at least the extended keywords and attributes.

  According to the present invention, while recommending content such as a wide range of broadcast scheduled programs while relating to a scene that the user was interested in while viewing the content, the user wanted to view after recording the content at the time of recording reservation You will be able to easily watch the part that will be.

It is a block diagram which shows the structural example of the content recommendation apparatus which concerns on one Embodiment of this invention. It is a flowchart which shows an example of the process sequence of scene division, scene feature extraction, and program feature extraction. It is a figure which shows an example of the scene division | segmentation result by a scene division part. It is a figure which shows an example of the document which is metadata of a program. It is a figure which shows an example of a specific expression extraction result. It is a figure which shows an example of a weight table. It is a figure which shows an example of the video information accumulate | stored in the video information storage part. It is a flowchart which shows an example of the process sequence from scene selection during the reproduction | regeneration of a program to accumulation | storage of a scene reservation list. It is a figure which shows an example of the viewing-and-listening scene information acquisition result by a viewing-and-listening scene information acquisition part. It is a figure which shows an example of the search result of a net moving image. It is a figure which shows the initial state of the co-occurrence matrix of the tag of FIG. It is a figure which shows the state which added the number of times of co-occurrence to the co-occurrence matrix of FIG. It is a figure which shows an example of a recording reason information selection screen. It is a figure which shows an example of a reservation program selection screen. It is a figure which shows an example of a scene reservation list. It is a figure which shows an example of the video information to which the reason for recording was added. It is a figure which shows an example of the recorded program list screen. It is a figure which shows an example of the scene list displayed on the recorded program list screen. It is a figure which shows an example of a scene menu.

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

  In the following, a case where the content recommendation device according to the present embodiment is applied to a recording program recommendation / reservation function in a video recording device will be described as an example. However, the present invention is not limited to this. It can be applied to various functions other than the program recommendation / reservation function. Here, the content is, for example, a broadcast program. Here, the video recording device is not limited to an independent device, but may be a device incorporated in a television receiver or a device incorporated as hardware or software in a PC incorporating a television reception function.

  In general, the content recommendation device according to the present embodiment extracts a keyword that characterizes a scene being viewed, expands the keyword using information on the Internet video site, and searches for a broadcast program searched using the expanded keyword. Etc., and the information (recording reason information, etc.) from which the recommended content is recorded after the recommended content is recorded is presented.

  A configuration example of the content recommendation device according to the present embodiment will be described with reference to FIG.

  As shown in FIG. 1, the content recommendation device of this embodiment includes a video playback / scene designation unit 11, a recording reason information designation unit 12, a scheduled recording program designation unit 13, a recorded program designation unit 14, and a program feature extraction unit. 15, scene division unit 16, scene feature extraction unit 17, viewing scene information acquisition unit 18, recording reason information acquisition unit 19, program search unit 20, scene reservation list storage unit 21, video storage unit 22, and video information storage unit 23 I have. Reference numeral 101 in the figure denotes a portion related to the user interface.

  The outline of each part is as follows.

  The scene reservation list storage unit 21 relates to a program (recording reservation program) selected by the user to make a recording reservation from among the programs recommended based on a certain scene of a certain program by the recommendation processing of the present embodiment. A list of information (scene reservation list) is accumulated.

  The video storage unit 22 stores programs recorded by a recording processing unit (not shown).

  In the present embodiment, a recording processing unit (not shown) performs recording processing according to information in the scene reservation list stored in the scene reservation list storage unit 21 (records a program and stores it in the video storage unit 22). To do). The recording processing unit (not shown) is a recording reservation list created by a conventional method (for example, a recording reservation list created by direct input / editing by a user or input / editing using an electronic program guide). Therefore, the program may be stored in the video storage unit 22. In this case, the scene reservation list of this embodiment may be provided separately from the existing recording reservation list, or the scene reservation list may be integrated with the recording reservation list. The recording processing unit (not shown) further outputs video / audio to a screen / speaker (not shown) at that time, for example, when a remote control or a recording button on the GUI screen is pressed by the user. It may have a function of accumulating existing programs in the video accumulation unit 22.

  The program feature extraction unit 15 extracts a feature keyword (program feature keyword) of the program stored in the video storage unit 22 (that is, a recorded program).

  The scene division unit 16 divides a program into scenes for the programs stored in the video storage unit 22 (that is, recorded programs). A conventional method may be used for scene division of a program. However, if there is recording reason information when the program is reserved for recording, the program scene is divided using the recording reason information. (The scene dividing unit 16 (dividing unit) may divide the video information (content information) into scenes (divide) using the recording reason information (reason for recording reservation) included in the video information (content information). You may do it). For programs that do not have recording reason information (for example, programs recorded by a conventional method), scene division of the program may be performed by a conventional method.

  The scene feature extraction unit 17 extracts feature keywords (scene feature keywords) of individual scenes from the scene-divided program.

  The video information storage unit 23, for each program stored in the video storage unit 22, for example, metadata such as EPG, program feature keywords extracted by the program feature extraction unit 15, and scene division by the scene division unit 16. The program information (content information) including the result, the scene feature keyword of each scene extracted by the scene feature extraction unit 17 and the recording reason information when the program is reserved for recording is accumulated (however, the recording reason information For programs that do not have, this field may be left blank). Note that the metadata may be, for example, data received by a metadata acquisition unit (not shown) from a broadcasting station or a server, or may be extracted from program content as necessary. They may be used in combination.

  The video playback / scene designation unit 11 plays back the video of the program stored in the video storage unit 22 (or a program that is not a recorded program), and becomes the first trigger of the recommendation process of the present embodiment during the playback. Receive user input. Here, as a specific example of “user input serving as the first trigger”, a case where the “search” button is pressed will be described as an example, but the present invention is not limited to this.

  When the user wants to receive a recommended program based on the currently viewed scene, the user may press a “search” button, for example.

  Note that the program that was being played when the “first user input as a trigger” was received is referred to as the “original program”.

  The method for selecting a program to be viewed by the user may be the same as the conventional method, such as selecting from a list of recorded programs.

  The viewing scene information acquisition unit 18 stores, in the video information storage unit 23, the feature information (viewing scene information) of the scene that has been played when the “search” button is pressed for the program being played. Further, it is obtained from the scene feature keyword of the scene of the program (original program).

  The recording reason information acquisition unit 19 acquires video content from a net video site (for example, a posting site) 30 based on the scene feature information obtained by the viewing scene information acquisition unit 18, and acquires the acquired video. Based on the content information, an extended keyword is acquired as a candidate for recording reason information, and further an attribute of the extended keyword (for example, an evaluation value, a date, etc.) is acquired.

  The recording reason information specifying unit 12 presents recording reason information (at least an extended keyword) as a candidate, and displays a screen for the user to select one or a plurality of recording reason information from the presented recording reason information candidates. Generate (this screen presents candidates for recording reason information and accepts selection by the user for the candidates for recording reason information presented).

  If there is desired recording reason information on the screen, the user can select it.

  When the recording reason information is selected by the user on the screen, the program search unit 20 uses the recording reason information to perform an EPG search for content such as a broadcast-scheduled program.

  The scheduled recording program specifying unit 13 presents (recommends) the search result (program) of the program search unit 20 to the user, and a screen for the user to select one or more recommended programs from the presented recommended programs. (This screen presents the recommended program and accepts selection by the user for the presented recommended program).

  The user can select a desired program among the presented recommended programs in order to reserve recording of the program.

  The selected scheduled recording program and information such as the recording reason information at that time are stored in the scene reservation list storage unit 21.

  The recorded program specifying unit 14 presents information on the recorded program and generates a screen for selecting a user to view from the presented programs (this screen displays information on the recorded program). And accepting selection by the user for the presented program). When presenting information related to a recorded program, for a program recorded based on the scene reservation list, the recorded program list is acquired from the scene reservation list storage unit 21 and stored in the video information storage unit 23. Information is presented using information such as program scene information and recording reason information.

  By the way, the content recommendation device of the present embodiment can also be realized by being installed as an application in a PC (Personal Computer) capable of viewing and recording video. Further, the PC used for the content recommendation device may have any shape such as a notebook type, and may be realized by using a video viewing device such as a TV other than the PC, or a video recording device such as an HDD recorder. .

  Various methods can be used as input means for operations such as selecting a recommended program. When using a PC, for example, a mouse or a keyboard can be used. In the case of a video viewing device such as a TV, it is only necessary to input at least up / down / left / right, decision / cancel, etc. Below, it demonstrates centering on the case where it operates with a remote control toward TV screen.

  Here, the timing at which the content recommendation device of the present embodiment performs processing will be described.

  The processing of the program feature extracting unit 15 / scene dividing unit 16 / scene feature extracting unit 17 is performed when a program is recorded. The processing may be performed while recording, or may be performed after the recording is completed.

  The processes of the viewing scene information acquisition unit 18, the recording reason information acquisition unit 19 and the recording reason information designation unit 12 cause an action such as the user viewing the video and pressing the “search” button through the video playback / scene designation unit 11. When you do. Following this, the recording reason information acquisition unit 19 and the recording reason information designation unit 12 are processed. Subsequently, the program search unit 20 and the recording reservation program designation unit 13 are processed.

  The processing of the recorded program specifying unit 14 and the like is performed after the program reserved for recording by the user through the content recommendation device is actually recorded.

  The video playback / scene designation unit 11 plays back a video of a program that is not a recorded program (for example, a program currently being broadcast or a program such as a DVD), and during the playback, the recommendation process of the present embodiment is performed. It may be possible to receive user input as an initial trigger. In this case, the scene division process, the scene feature keyword extraction process, and the program feature keyword extraction process are performed when the program is viewed. In the following, a description will be mainly given of a case where a user input serving as a first trigger of the recommendation process of the present embodiment is received during playback of a recorded program.

  Hereinafter, the process performed by the content recommendation device of the present embodiment will be described in more detail.

  In the following description, an example in which a user reserves recording of another recommended program while viewing a recorded TV program and views the program after recording will be described.

  First, an example of a procedure for scene division and feature extraction (program feature extraction and scene feature extraction) will be described with reference to FIG. In FIG. 2, KW represents a keyword (the same applies to KW in other figures).

<Scene division processing>
When recording a program, first, the scene dividing unit 16 divides the recorded program into a plurality of scenes (step S1).

<Scene feature extraction processing>
Next, the scene feature extraction unit 17 extracts, for each scene, a scene feature keyword that characterizes the scene.

Specifically, first, it takes out (step S3) one scene of the program, extracts the scene feature keyword candidate scene S _i taken out (step S4). Then, one scene feature keyword candidate is extracted (step S6), and the feature amount of the extracted scene feature keyword candidate w _{i, j} is calculated (step S7). Repeated this (step S5) after the calculation of the characteristic quantity for all candidate feature keyword of the scene S _i is completed, selecting the feature keyword of the scene S _i (step S8). Finally, the characteristics keywords of the selected scene S _i, accumulates in the video information storage section 23 (step S9).

  The above processing is sequentially performed for all scenes. After the above process is completed for all scenes (step S2), the process proceeds to the next program feature extraction unit process.

<Program feature extraction processing>
Next, the program feature extraction unit 15 performs program feature keyword extraction that characterizes the entire program.

Specifically, program feature keyword candidates are extracted for the program (step S10). Then, one program feature keyword candidate is extracted (step S12), and the feature amount of the extracted feature keyword candidate wg _{, j} is calculated (step S13). This is repeated, and after the feature amount calculation is completed for all program feature keyword candidates (step S11), a program feature keyword is selected for the program (step S14). Finally, the selected program feature keyword is stored in the video information storage unit 23 (step S15).

  In the above, the scene feature extraction process is performed first, and then the program feature extraction process is performed. However, the present invention is not limited to this. For example, the program feature extraction process is performed first, and then the scene feature extraction process is performed. The scene feature extraction process and the program feature extraction process may be performed in parallel.

  In the following, the “scene division processing (scene division unit 16)”, “scene feature keyword extraction processing (scene feature extraction unit 17)”, and “program feature keyword extraction processing (program feature extraction unit 15)” will be described in detail. To do.

  First, the scene division process will be described.

  The scene dividing unit 16 divides the video of the program into a plurality of scenes using character information and video information when the program is recorded.

  For example, an existing scene division method for video information can be applied to the video scene division. For example, a method may be used in which a video is divided for each shot and the video is divided for each scene according to the similarity.

  For scene division, an existing scene division method for text data instead of video data can be applied. In this method, scene division can be performed in consideration of the topic content. For example, a method of automatically detecting breaks between topics by comparing vocabularies included in texts may be used.

  The result of scene division is obtained, for example, by the time stamp of the division point.

  FIG. 3 shows an example of a scene division result by the scene division unit 16. In this example, the start time of each scene is recorded.

  The scene division result is stored in the video information storage unit 23 in association with the program.

  Next, the scene feature keyword extraction process will be described.

  The scene feature extraction unit 17 extracts a scene feature keyword that is a feature of each divided scene.

  First, the scene feature extraction unit 17 uses, for example, a keyword, a semantic attribute of the keyword, an appearance position of the keyword, as information on the scene feature keyword candidate from the metadata of each scene divided by the scene division unit 16. To extract.

  In many cases, video data is accompanied by text data called closed captions as metadata. This can be used as extraction source metadata for character information. Alternatively, metadata other than closed captions may be used. For video that does not include closed captions, it is also possible to apply automatic speech recognition technology to audio data in the video and use it as character information extraction source data.

  As a means for extracting a keyword, for example, a method of extracting a noun phrase using a technique such as an existing morphological analysis can be used.

  As a means for determining the semantic attribute of the keyword, for example, an existing specific expression extraction method using a dictionary or a rule can be used. For example, a rule such as “If the expression“ corporation… ”is applied to a certain character string, its semantic attribute is“ company name ”” is prepared. If the expression is “O”, the semantic attribute can be determined as “company name” by the keyword semantic attribute determination process.

  FIG. 4 shows an example of a document which is program metadata. FIG. 5 illustrates the result of extracting character information from the document of FIG. In the table example of FIG. 5, semantic attributes and appearance positions are recorded for each keyword.

  Next, the scene feature extraction unit 17 calculates the feature amount of each extracted feature keyword candidate.

  The feature amount is an index representing how much the keyword characterizes the scene designated by the user.

  As a method for calculating the feature amount, for example, a method using tf-idf often used in document search may be used. In this case, for example, tf-idf is calculated with a closed caption of one scene as one document and the entire program as a document group composed of them.

Here, each scene of the division result _{S 0, S 1, ...,} S i, ..., and _{S n,} scene _{S i} features keyword candidate _w at _{i, 0, w i, 1} , ..., w i, _m .

Formula (1) shows a method of calculating the feature quantity f _{i, j} of the j-th keyword candidate w _{i, j} in the scene S _i using tf-idf.
f _{i, j} = tf (w _{i, j} ) × log {n / df (w _{i, j} )} (1)
In Expression (1), tf (w _{i, j} ) is the number of occurrences of w _{i, j} in S _i , n is the total number of scenes, and df (w _{i, j} ) is the number of scenes in which the keywords w _{i, j} appear. It is.

  Further, as another existing method, for example, an index for keyword selection in relevance feedback can be used.

  In addition, the simple statistical information as described above is not used as it is for the feature amount, but a weight may be given to each type of feature keyword candidate. In this case, for example, as shown in FIG. 6, a table in which the weight of the semantic attribute is defined for each program genre may be prepared and referred to when calculating the feature amount. The genre of the program can be easily known from the metadata.

When the example of FIG. 6 is used, if w _{i, j} = “polar bear (semantic attribute: mammal)” and the program genre is “documentary / cultural / animal” _, the weight cw (wi _{, j)} of w _{i, j} ) Is 1.2, and cw (w _{i, j} ) × f _{i, j} = 1.2 × f _{i, j} is used as the feature amount.

  In the table illustrated in FIG. 6, even with the same semantic attribute, the weight can be changed according to the program genre. For example, the weight of the semantic attribute “person name” in the program genre “animal” is “0.9”, whereas the weight of the semantic attribute “person name” in the program genre “travel” is “1.2”. Further, in defining the semantic attribute weight, other elements may be used instead of or in addition to the program genre illustrated in FIG. 6, and the weight may be uniquely defined for each semantic attribute.

For the scene feature keyword candidates whose feature values have been calculated as described above, for example, the top _ns items with the largest feature values are set as the scene feature keywords of the scene. Note that n _s is a predetermined integer. Further, the user may be able to adjust _ns as appropriate.

  The extracted scene feature keyword is stored in the scene item in the video information storage unit 23 in association with the program.

  Next, the program feature keyword extraction process will be described.

  The program feature extraction unit 15 extracts keywords appearing in the entire program as program features.

  Similar to the scene feature keyword candidate extraction process of the scene feature extraction unit 17, the program feature extraction unit 15 uses, for example, a keyword and the meaning of the keyword as information on the program feature keyword candidate from the EPG data given to the program. The attribute and the appearance position of the keyword are extracted.

For the extracted program feature keyword candidates, a universal quantity indicating whether or not it appears in the entire program is calculated. The universal amount uses a value that increases if the program feature keyword candidates w _{G, 0} , w _{G, 1} ,..., W _{G, k} appear evenly in the metadata (eg, closed caption) of each scene. . For example, the closed caption of one scene may be set as one document and the above-described df (w _{G, i} ) may be used.

  Note that this alone may increase the universal amount of general keywords in the EPG. In such a case, in addition to this, tf-idf is calculated using EPG of one program as one document and EPG of many programs obtained so far as a document group. You may omit it.

For the program feature keyword candidates whose universal quantities have been calculated as described above, for example, the top n _p items having the largest feature quantity are set as the program feature keywords of the program. n _p is a predetermined integer. Further, the user may be able to adjust n _p as appropriate.

Note that n _p and n _s may always be set as the same value, or n _p and n _s may be set independently.

  The extracted program feature keyword is stored in the video information storage unit 23 in association with the program.

  The above processing is performed when a program is recorded (when a program that is not a recorded program is viewed), and the results are stored in the video information storage unit 23.

  FIG. 7 shows an example of video information stored in the video information storage unit 23. In this example, the video information storage unit 23 includes a program identifier (program ID) that uniquely indicates the program, and each piece of EPG information that is metadata of the entire program (in the example of FIG. 7, title, broadcast date, overview, genre). ), A program feature keyword, division result information corresponding to each scene (for example, the start time and end time of the scene), a scene feature keyword, a recording reason, and the like are stored as one record.

  In the example of FIG. 7, display label information is also accumulated for each scene. Various existing methods may be used to generate the label. For example, a method of selecting a sentence including a keyword extracted using a method such as the feature keyword extraction described above can be used. In addition, when an outline sentence in the EPG data includes an outline of each scene, it is possible to acquire the corresponding sentence from the EPG and use it as a label.

  The reason for recording will be described later.

  Next, an example of a processing procedure from scene selection to scene reservation list accumulation during program playback will be described with reference to FIG.

  The video playback / scene designation unit 11 starts playback of the video of the program (step S21).

  When the video playback / scene designation unit 11 receives a scene selection (ie, “user input as a first trigger for recommendation processing” such as a “search” button) (step S22), the viewing scene information acquisition unit 18 first. The feature information (scene feature keyword) of the selected scene is obtained, and then the recording reason information acquisition unit 19 is acquired from the site of the online video (for example, a posting site) 30 based on the feature information. Video content information information is used to acquire recording reason information candidates (extended keywords and their attributes), and the recording reason information specifying unit 12 presents the recording reason information (at least extended keywords) as candidates (step S22). .

  When the recording reason information specifying unit 12 accepts the selection of recording reason information (step S24), first, the program search unit 20 performs EPG search of a broadcast-scheduled program based on the selected recording reason information, and recording. The reserved program designating unit 13 presents the search result (recommended program) (step S25).

  When the scheduled recording program specifying unit 13 accepts the selection of the scheduled recording program (step S26), the selected scheduled recording program and information such as the recording reason information at that time are stored in the scene reservation list storage unit 21 (step S27). .

  Hereinafter, each of the above processes will be described in detail.

  First, viewing scene information acquisition will be described.

  The viewing scene information acquisition unit 18 is currently viewed by the user when an action such as pressing the “Search” button on the remote controller or selecting the “Search” menu while the user is viewing the program. That is, the feature information of the scene currently being reproduced is acquired as three types of keywords (G feature keyword, P feature keyword, and N feature keyword).

  Here, the G feature keyword is a keyword that characterizes the entire program.

  The P feature keyword is a keyword that characterizes the selected scene.

  The N feature keyword is a keyword that characterizes a scene that has not been selected and does not appear in the selected scene.

  Hereinafter, a method for acquiring these three types of keywords will be described.

  When the user takes an action such as pressing the “search” button, the viewing scene information acquisition unit 18 acquires a record of the program from the video information storage unit 23 using the program ID of the program currently being played as a key. At the same time, a program feature keyword is acquired as a G feature keyword candidate.

  Next, the number of seconds after the program start time is the time at which the current playback is in the program (hereinafter, the time indicating the number of seconds after the program start time is referred to as “elapsed time”). )), Information on the scene currently being reproduced is acquired from the program information. Then, the feature keyword assigned to the scene information currently being reproduced is acquired as the P feature keyword.

  Further, a feature keyword assigned to a scene other than the scene currently being reproduced is used as an N feature keyword. However, if feature keywords for all scenes that are not being played are used, the number will increase too much.For example, only the feature keywords for the scene before and / or after the scene that is currently being played will be used. Is preferred.

  FIG. 9 is an example of acquiring information of a scene being viewed from the stored video information.

  In the example of FIG. 9, among the feature keywords “Polar bear” and “Ice area” of the currently playing scene i (i = 2 in this case), the feature amount of “Polar bear” is larger, and the feature amount is larger. “Polar bear” is selected as the P feature keyword.

  The time when the user presses the “search” button in the scene (i) instead of referring to the feature amount (that is, the elapsed time of the portion that was played back when the scene was selected). The feature keyword that appears at the closest elapsed time may be used as the P feature keyword.

  In the example of FIG. 9, the feature keywords “seal” and “walrus” of the scene (i−1) (i−1 = 1 in this case) immediately before the scene (i), and the scene (i + 1) ( In this case, the feature keywords “penguin” and “seabird” of i + 1 = 3) are selected as the N feature keywords of the scene (i).

  Next, acquisition of recording reason information will be described.

  The recording reason information acquisition unit 19 acquires an extended keyword from the site 30 of the online video based on the scene feature keyword acquired by the method described above.

  A large number of moving images exist on the Internet moving image site 30 by, for example, posting, and a plurality of character strings indicating the contents of the moving images called tags are given to them. The user can search for moving images using this tag as a query key. In addition, date and time information such as posting date and time or posting date, evaluation value as a result of evaluating a moving image by a person who viewed the moving image, and the like are often given to each moving image. When obtaining evaluation values from multiple Internet video sites, if the method of assigning the evaluation values is different, normalize the evaluation values of each site, etc., and obtain the evaluation values as attributes of extended keywords. Anyway. In addition, the evaluation value obtained from the Internet video site 30 is used, or alternatively, other information obtained from the Internet video site 30 (for example, the number of times of reproduction, etc.) is used as the extended keyword attribute. The evaluation value may be calculated. Furthermore, as an attribute of the extended keyword, only the evaluation value may be used, the evaluation value and the posting date / time may be used, and various other methods are possible.

  In the present embodiment, a case where moving image meta information in which a tag, an evaluation value, and a posting date / time are set is used as an information source of an extended keyword will be described as an example.

  Here, a case will be described as an example where the obtained extended keyword and its attributes (evaluation value, etc.) are generated as “recording reason information” that is information when recording a recommended program.

  First, a method for acquiring an extended keyword using the P feature and a method for acquiring an extended keyword using the N feature will be described, and then a method for acquiring the attribute of the extended keyword will be described.

  First, a video search is performed from the network video site 30 using the P feature keyword. In this case, the search query may simply be “(P feature keyword)” (for example, “polar bear” in FIG. 9). However, at this time, if the number of search result videos is large, another P feature keyword (for example, “ice area” in FIG. 9) or G feature keyword (for example, “animal”, “north pole” in FIG. 9). "," Antarctica ") can be used as an AND query to narrow down the search. Further, N feature keywords (for example, “seal” in FIG. 9) may be used as a NOT query as appropriate.

  Next, some tags are selected as expansion keywords from the set of tags attached to the search result moving image.

  Hereinafter, an extended keyword selection method will be described with reference to FIGS.

Assume that the following M _n +1 moving images are obtained as a result of searching for moving images.
Movie M ₀ : tag t _0,0 , t _0,1 ,..., T _{0, T0} , evaluation value r ₀ , posting date and time d ₀
Video _{M 1:} tag _{t 1,0, t 1,1, ...,} t 1, T1, evaluation value _{r 1,} contribution date and time _{d 1}
. . .
Video M _n : tags t _{n, 0} , t _{n, 1} ,..., T _{n, Tn} , evaluation value r _n , posting date and time d _n
From the obtained tags t _0,0 ,..., T _{n, Tn} , the feature keyword used in the query (for example, “Polar bear” in FIG. 9) is removed, and the tag set t ₀ ,. , _{T T.}

Next, the tag set t ₀ ,..., _{T T} is divided into groups that co-occur between videos.

Here, a co-occurrence matrix is created, and the moving image is referred to in order from the moving image M _0, and is added to the numerical value of the corresponding item in the co-occurrence matrix for two tags attached to one moving image .

  For example, assuming that a moving image as shown in FIG. 10 is obtained from the result of searching with the P feature keyword “polar bear”, the number of co-occurrence is added to the initial co-occurrence matrix as shown in FIG. In an actual example, it is considered that a larger amount of moving images than in FIG. 10 is obtained as a search result.

In FIG. 10, for example, in the video M ₀ , when a tag other than “Polar Bear” that was the search query is referenced, in this case, “Tobe Zoo” and “Piece” are assigned as tags. Add 1 to the value of the term (2nd row, 2nd column) where “Tobe Zoo” and “Piece” intersect.

  Thus, if it adds to the term of a co-occurrence matrix in all the moving images, for example, a co-occurrence matrix as shown in FIG. 12 will be obtained.

  Next, a pair (or group) having a large number of co-occurrence is extracted from the co-occurrence matrix after addition. However, those included in the P feature keyword, the N feature keyword, or the G feature keyword are omitted because an extended keyword cannot be obtained.

  In the example of FIG. 12, the N feature keyword is “seal”, and the G feature keyword is “animal”, “north pole”, “antarctic”, so the resulting pairs are “Tobe Zoo-Peace”, “Asahiyama Zoo-Penguin” "," Global warming-endangered species ".

  Finally, for example, the expansion keywords are selected one by one from the obtained pairs.

  Any index may be used for this selection method.

  To give a simple example, for example, a broadcast schedule EPG that is a recommendation target may be searched with each keyword, and the larger number may be selected as an extended keyword. This can prevent the final recommended program from being reduced.

  In addition, various methods are conceivable, such as selecting an item that has a large number of co-occurrence with other words or phrases or selecting an item that is frequently searched by searching for an online video.

  The keyword obtained as described above becomes an extended keyword.

  Instead of using co-occurrence tags as extended keywords as described above, assuming that a set of tags assigned to one video is one document, an existing document clustering method is applied, and a representative is selected from these. You may take the method of selecting one tag.

  In addition to these methods, various methods can be used as an extended keyword acquisition method.

  Similarly, an extended keyword using N features is also acquired. In this case, the process of acquiring the extended keyword may be performed by switching the roles of the N feature keyword and the P feature keyword.

  That is, the search query may simply be “(N feature keyword)” (for example, “seal” in the example of FIG. 9). However, at this time, when the number of videos in the search result is large, other N feature keywords (for example, “walrus” in FIG. 9) or G feature keywords (for example, “animal”, “north pole” in FIG. 9). , “Antarctica”), and use the appropriate search. Further, a P feature keyword (for example, “polar bear” in FIG. 9) may be used as a NOT query as appropriate.

  The method for acquiring the extended keyword from the search result is the same as that for acquiring the extended keyword based on the P feature.

  The extended keyword acquisition method using the P feature and the extended keyword acquisition method using the N feature may be the same method or different methods.

  Preferably, for example, the recording reason information acquisition unit 19 (second acquisition unit) uses a scene feature keyword (second feature keyword) related to a predetermined scene other than the selected scene (specific scene) as a query, Alternatively, a scene feature keyword (second feature keyword) related to a predetermined scene other than the selected scene (specific scene) is used as a query, and a program feature keyword (first feature keyword) related to the content is used as an AND query. And / or using the scene feature keyword (second feature keyword) related to the selected scene (specific scene) as a NOT query to obtain a plurality of other net moving images, and related to the obtained other plurality of net moving images. Based on the information, other extended keywords are acquired and at least as attributes of the other extended keywords. And acquire the evaluation value for the net video recording reason information specifying section 12 (first selector) further may be also present a list of the other extended keywords obtained.

  Further, a configuration in which the acquisition of extended keywords using N features is omitted is also possible. In this case, in the subsequent processing, a portion related to the extended keyword using the N feature may be omitted.

  Next, a method for acquiring the extended keyword attribute will be described.

  In addition to the tags, information such as evaluation values and posting dates / times is given to the moving image meta information.

  Here, the evaluation value is useful as an index for determining which expansion keyword is recommended by the user, and the posting date is useful as an index for determining the topicality of the expansion keyword. Include in recording reason information.

  The evaluation value as the attribute of the extended keyword is obtained by performing predetermined statistical processing on the evaluation value included in the information including the extended keyword among the information on the plurality of net moving images from which the extended keyword is acquired. The obtained value may be used.

  For example, the evaluation value of the extended keyword can be obtained simply by taking the median value of the evaluation values of the moving image group having the extended keyword as a tag. The evaluation value of the extended keyword does not necessarily have to be one value. For example, the quartile range of the evaluation value of the moving image group may be extracted as the evaluation value.

  Similarly, the posting date and time as the attribute of the extended keyword is a predetermined date with respect to the posting date and time included in the information including the extended keyword among the information on the plurality of net moving images from which the extended keyword is acquired. It may be a value obtained by performing statistical processing.

  For example, the median value and the quartile range may be used for the posting date of the extended keyword as well as the evaluation value.

  Of course, the present invention is not limited to the median or quartile range, and various statistical indicators may be used for both.

  The extended keywords and their attributes obtained in this way are referred to as recording reason information which is the reason for recording when the user makes a recording reservation for a program.

  Next, designation of recording reason information will be described.

  The recording reason information specifying unit 12 generates a screen for selecting what the user desires from the recording reason information obtained as described above.

  FIG. 13 shows an example of a recording reason information selection screen generated by the recording reason information specifying unit 12.

  FIG. 13 shows an example in which three pieces of recording reason information for “scene being viewed” are presented as candidates on the left of the screen example, and three pieces of recording reason information for “previous scene” are presented as candidates on the right side of the screen example. .

For example, recording reason information specifying section 12 performs a set R _P of recording reasons information obtained from the P characteristic keywords, the set operation for the set R _N recording reason information obtained from N feature keyword.

The recording reason information in the "scene being viewed" on the left side of the screen example, and outputs the sort the results of R _P \ R _{N. Here,} R P \ _{R N} represents a set excluding the set _{R N} from the set _{R P.}

  As a sorting criterion, for example, the order in which the evaluation values of the expansion keywords are high may be considered.

The recording reason information in the "previous scene" in the example screen right and outputs the result of R _N, but also preferentially output the R _P ∩ R _N therein. Here, _{R P} ∩ R _N represents the logical intersection of the set _{R P} the set _{R N.}

  The evaluation value and the posting date / time information in the recording reason information may be displayed as they are. For example, if the evaluation value has a certain height or more, a “specific mark (for example, a star mark) ( (See 121 in FIG. 13), change the color of the extended keyword depending on the evaluation value, add “!” (See 122 in FIG. 13) or “New” if the posting date / time distribution is recent. You can increase the information displayed on the screen.

  Further, as an index for the user to select the recording reason information, a part of the viewer's comment given to the original net moving image may be displayed.

  Various other variations are possible.

  On the screen for selecting recording reason information, the user can select one recording reason information using an operation device such as a remote controller.

  FIG. 13 illustrates a state in which “Totobe Zoo” is selected by the user among the three recording reason information in the “viewing scene” presented on the left side of the screen example (see 122 in FIG. 13).

  When the selection is determined, the content recommendation device shifts the processing to the program search unit 20.

  Next, program search, presentation of the results, and designation of a scheduled recording program will be described.

  The program search unit 20 searches for an EPG of a broadcast-scheduled program based on the extended keyword in the recording reason information selected by the user.

  In addition, when the EPG of the broadcast scheduled program is not held, for example, a program search service such as that found in a Web service may be used.

  The result of the program search is presented to the user by the recording reservation program designating unit 13.

  FIG. 14 shows an example of a screen (reserved program selection screen) for the user to select a reserved program, which is generated by the recording reserved program designating unit 13.

  FIG. 14 shows an example in which two programs are presented as search results by “Tobe Zoo”.

  When presenting a plurality of recommended programs (reserved program candidates), for example, how many P-feature keywords appear in the EPG, G-feature keywords appear, and N-feature keywords for a plurality of recommended programs. It may be rearranged considering whether it appears.

For example, P feature keyword appearance count P _P , G feature keyword appearance count P _G , N feature keyword appearance count P _N ,
α × P _P + β × P _G −γ × P _N (α, β, γ> 0)
May be sorted by this score as the score of each reserved program.

  The user can select a desired reserved program from the presented programs using an operation device such as a remote controller.

  FIG. 14 illustrates a state in which “Polar bear artificial breeding” is selected by the user from the presented program (see 124 in FIG. 14).

  Next, accumulation of the scene reservation list will be described.

  FIG. 15 shows an example of a scene reservation list stored in the scene reservation list storage unit 21.

  In the example of FIG. 15, when a recording reservation program is designated by the user, the program ID of the program reserved for recording and “user input as a first trigger for recommendation processing” such as a “search” button are accepted. (Sometimes) the program ID (original program ID) of the original program that was being viewed, and an index (original scene) that indicates the number of scenes in the original program that were being viewed (in the original program) index), the P feature keyword, the N feature keyword, the G feature keyword of the scene being viewed, and the recording reason information (in this example, the extended keyword, the evaluation value, the posting date and time) and recorded in association with each other The list is stored in the scene reservation list storage unit 21. Note that other information may be stored in the scene reservation list. For example, the recording reservation time or the time when the scene of the original program is selected may be included.

  In the case of the example of FIG. 15, when the scene reservation list of the present embodiment is integrated with the existing recording reservation list, the items after the original program ID are recorded for the programs reserved for recording by the conventional method. Should be left blank.

  As described above, a recording processing unit (not shown) records a program based on the information in the scene reservation list and stores it in the video storage unit 22.

  Now, let us return to the processing at the time of program recording.

  As already described, when a program is recorded, scene division and feature extraction (program feature extraction and scene feature extraction) are performed for the program, and information is recorded in each item of program information as shown in FIGS. Is done.

  By the way, in the examples of FIGS. 7 and 9, the item “Reason for recording” is blank, but the item “Reason for recording” is reserved for the program recommended and recorded from the original program as described above. Information is stored.

  At that time, the P / N / G feature keyword and recording reason information (in this example, extended keyword, evaluation value, and posting date / time) in the scene reservation list storage unit 21 are referred to for the program, and closed captions for each scene are used. It is checked whether the P feature keyword, the N feature keyword, the G feature keyword, and the extended keyword of the recording reason information are not included. If it is included, the P / N / G characteristic keyword and the recording reason information (in this example, an extended keyword, Store all or part of (evaluation value, posting date).

  FIG. 16 shows an example of video information after this processing (video information to which the item “recording reason” is added).

In FIG. 16, the symbol in “Reason for recording”
P: P feature keyword,
N: N feature keyword,
G: G feature keyword,
E: Extended keyword,
R: Evaluation value of extended keyword,
D: Date and time of extended keyword posting
Respectively.

  In FIG. 16, “recording reason” is directly stored in the video information storage unit 23, but “recording reason” is held in a separate table and a pointer to the “recording reason” is set as video information. You may make it store in the accumulation | storage part 23. FIG.

  By the way, the scene dividing unit 16 may change the behavior by using the above “recording reason”. For example, if character information is used when performing scene division, a change such as weighting the keyword weight in the “recording reason” at the time of vocabulary comparison can be considered.

  Hereinafter, presentation, selection, display, etc. of recorded programs will be described.

  The recorded program specifying unit 14 generates a screen (recorded program list screen) for the user to select a recorded program after recording the program.

  A list of recorded programs is obtained from the scene reservation list.

  FIG. 17 shows an example of a recorded program list screen (and an enlarged view of a scene list). In this example, the thumbnail, title, and broadcast date and time of each program are displayed.

  It is desirable to display the program for which the recording reason information is specified at the time of recording reservation so that the user can know that fact. For example, a “specific mark (for example, a star-shaped mark) (see 141 in FIG. 17)” may be added and presented on the thumbnail of the program. As a result, the user can immediately know that there is a reason for viewing the program and the program has been recorded.

  If the recorded program cannot be displayed on the list screen, it can be scrolled as in the prior art. Further, the user may be able to appropriately switch between the display in the thumbnail format and the display in the list format.

  Here, in the recorded program list screen of FIG. 17, among the recorded programs, “program for which recording reason information was specified at the time of recording reservation” is selected by the user or the cursor is put on the program. The recording reason and attribute information are displayed together with the scene list of the program.

  FIG. 18 shows a case where the scene list of the program “Polar bear artificial breeding” is displayed in the form of a timeline in a state where the cursor is placed on the recorded program “Polar bear artificial breeding” displayed with a star-shaped mark. (See 142 in FIG. 18). In this example, each square on the timeline represents the length of the scene, and a representative thumbnail of the scene and a label of the scene are displayed therein. At this time, the recording reason information designated by the user (in this case, the extended keyword “Tobe Zoo” and information on its attributes) is presented together. It should be noted that information such as the P feature keyword (in this example, “polar bear”) of the scene selected in the original program may be presented together. In addition, the program name of the original program that was played when the trigger by the “Search” button or the like that triggered the recording reason information was input, and the label information of the scene (in this case, the program name It may also be displayed together with “Animal Travel”, scene label “Polar Bear Raising”), or reservation date and time.

  This can be realized by, for example, processing as follows.

  The recorded program specifying unit 14 acquires information on the program from the video information storage unit 23 when the user moves the cursor over a thumbnail of a recorded program on the recorded program list screen, for example.

  On the recorded program list screen, a scene division result and a scene label are displayed as a scene list of the selected program. At that time, the scene in which the reason for recording is stored in the scene information is displayed with emphasis (in the example of FIG. 18, the scene is displayed in reverse video).

  Further, when a P / N / G feature keyword appears in the label text, the text is highlighted and displayed (in the example of FIG. 18, “North Pole” and “Sealing” are displayed in bold type. Represents the situation).

  Furthermore, information on the original program as the reserved date and time, information on the extended keyword, evaluation value, and posting date and time as recording reason information are presented at the bottom of the scene list.

  Since the program ID of the original program can be acquired from the scene reservation list, information on the program may be acquired from the video information storage unit 23 based on the program ID.

  In the example of FIG. 18, the P feature keyword “Polar bear” is extracted in a scene with the original program “Animal Travel”, and the recording reason information is generated based on the P feature keyword “Polar Bear”. , Posting date and time 2.5 months ago) ”is illustrated.

  Next, when the user moves the cursor (see 143 in FIG. 18) to a scene desired by the user in the scene list illustrated in FIG. 18, the recorded program designation unit 14 displays a scene menu.

  FIG. 19A shows an example of a scene menu presented in a state where one scene of a recorded program is selected. It should be noted that, in the scene menu of FIGS. 19B to 19D, a state where each item of “play from here”, “play from the beginning”, and “play related scenes of the recording source program” is selected is illustrated.

  If “Play from here” is selected, playback starts from the beginning of the scene where the cursor is.

  If “Play from the beginning” is selected, the program is played from the beginning.

  When “Play related scene of recording source program” is selected, the scene reservation list storage unit 21 is referred to and the scene of the original scene index is reproduced from the scene of the original program ID.

  As described above, according to the present embodiment, after the scene is divided, the program is recommended with the extended keyword based on the net moving image, so that it is possible to recommend a wide range of programs while being related to the scene currently being viewed. .

  Further, since an attribute such as an evaluation value is assigned to the extended keyword, the user can easily determine which extended keyword should be selected.

  In addition, the user can easily find and watch the portion of the recorded program that he / she would have wanted to see at the time of recording reservation.

Each of the above functions can be realized even if it is described as software and processed by a computer having an appropriate mechanism.
The present embodiment can also be implemented as a program for causing a computer to execute a predetermined procedure, causing a computer to function as a predetermined means, or causing a computer to realize a predetermined function. In addition, the present invention can be implemented as a computer-readable recording medium on which the program is recorded.

  Note that the present invention is not limited to the above-described embodiment as it is, and can be embodied by modifying the constituent elements without departing from the scope of the invention in the implementation stage. In addition, various inventions can be formed by appropriately combining a plurality of components disclosed in the embodiment. For example, some components may be deleted from all the components shown in the embodiment. Furthermore, constituent elements over different embodiments may be appropriately combined.

  DESCRIPTION OF SYMBOLS 11 ... Video reproduction / scene designation | designated part, 12 ... Recording reason information designation part, 13 ... Recording reservation program designation part, 14 ... Recorded program designation part, 15 ... Program feature extraction part, 16 ... Scene division part, 17 ... Scene feature Extraction unit, 18 ... viewing scene information acquisition unit, 19 ... recording reason information acquisition unit, 20 ... program search unit, 21 ... scene reservation list storage unit, 22 ... video storage unit, 23 ... video information storage unit

Claims (9)

For each reservation-recorded content, metadata including information related to the content, a first feature keyword that characterizes the content, information about a plurality of scenes that constitute the content, and each scene A first storage unit for storing content information including a second characteristic keyword to be characterized and a reason why the content is reserved for recording;
When a specific scene is selected from the scenes constituting the content being played back, the content information related to the content being played back stored in the first storage unit is changed to the specific scene. A first acquisition unit for acquiring related third feature keywords;
Based on information about a plurality of net moving images acquired using the third characteristic keyword, an extended keyword obtained by extending the third characteristic keyword is acquired, and at least the net moving image is used as an attribute of the extended keyword. A second acquisition unit for acquiring an evaluation value for
A first selection unit that presents a list of the acquired extended keywords and accepts a user selection for any of the presented extended keywords;
A second selection unit for presenting a list for recommending content that can be reserved for recording acquired using the user-selected extended keyword, and accepting user selection for any presented content;
A second storage unit for storing a reservation list in which information for recording reservation of the user-selected content is registered;
For the content that has been recorded based on the reservation list, the content information is used to present the scene of the content and the reason for the recording reservation including at least the extended keyword and its attribute. A content recommendation device comprising a presentation unit. A first extraction unit that extracts the first feature keyword from metadata about the content at the time of recording the content;
A division unit for dividing the content in order to generate data relating to the scene included in the content information at the time of recording the content;
2. The method according to claim 1, further comprising: a second extraction unit that extracts the second feature keyword for each scene of the content divided by the division unit from metadata related to the scene. The content recommendation device according to 1.   In the reservation list, for each of the contents reserved for recording, the feature keyword and the extended keyword and the attribute thereof related to the content being played back, which is the basis for recommending the content, are also registered. The reason for the recording reservation corresponding to the content included in the content information is that the feature keyword registered in the reservation list corresponding to the content and the extended keyword and its attribute are used. The content recommendation device according to claim 1, wherein the content recommendation device is stored at the time of recording. In the reservation list, for each of the contents reserved for recording, information related to the content and scene that was being reproduced and from which the content was recommended is also registered,
The said presentation part also presents the information regarding the content and the scene which were being reproduced and became the basis for recommending the content for the recorded content. Content recommendation device. The third characteristic keyword related to the specific scene of the content being reproduced is at least one or a plurality of second characteristic keywords related to the specific scene,
The content recommendation apparatus according to claim 1, wherein the net moving image is acquired using the second characteristic keyword as a query. The third feature keyword related to the specific scene of the content being played back further includes one or more first feature keywords related to the content and / or the specific scene of the content being played back Including one or more second feature keywords related to a predetermined scene other than
The net moving image is acquired as an AND query for the first feature keyword related to the content, and as a NOT query for the second feature keyword related to a predetermined scene other than the specific scene. The content recommendation device according to claim 1. The acquired information about the plurality of net videos includes an evaluation value for each net video,
The evaluation value as an attribute of the extended keyword is a predetermined value for an evaluation value included in the information about the plurality of net moving images from which the extended keyword is acquired, including the extended keyword. The content recommendation device according to claim 1, wherein the content recommendation device is a value obtained by performing statistical processing. The acquired information about the plurality of net videos includes the posting date and time for each net video,
The content recommendation apparatus according to claim 1, wherein the second acquisition unit also acquires a posting date and time for the net moving image as an attribute of the extended keyword. A content recommendation method for a content recommendation device,
A first storage unit included in the content recommendation device includes, for each reservation-recorded content, metadata including information relating to the content of the content, a first feature keyword that characterizes the content, and the content. Storing content information including information on a plurality of scenes constituting, a second feature keyword characterizing each scene, and a reason why the content is reserved for recording;
When the first acquisition unit included in the content recommendation device selects a specific scene from the scenes constituting the content being played back, the currently playing back stored in the first storage unit Obtaining a third feature keyword related to the specific scene from the content information related to the content;
A second acquisition unit included in the content recommendation device acquires an extended keyword obtained by extending the third feature keyword based on information about a plurality of net moving images acquired using the third feature keyword. And at least obtaining an evaluation value for the net video as an attribute of the extended keyword;
A first selection unit provided in the content recommendation device presents a list of the acquired extended keywords and accepts a user selection for any extended keyword presented;
The second selection unit included in the content recommendation device presents a list for recommending content that can be reserved for recording acquired using the extended keyword selected by the user, and a user for any presented content Accepting a selection;
Storing a reservation list in which information for recording reservation of the user-selected content is registered;
For the content that has been recorded based on the reservation list, the content information is used to present the scene of the content and the reason for the recording reservation including at least the extended keyword and its attribute. And a content recommendation method.

Images