JP2008117222A - Information processing apparatus and method, and program - Google Patents

Abstract

User information related to a predetermined type of content obtained from one device can be used for recommendation of another type of content or recommendation on another device.
A client device generates at least one property information representing a user property of a movie from a history of user behavior related to movie playback or the like, and stores the property information in a property profile storage unit. On the other hand, the client device 1 generates one or more pieces of property information indicating the properties of the user regarding the music from the history of the user's behavior regarding the music reproduction and the like and stores the property information in the property profile storage unit 38. One or more property information about the music is stored in the property profile storage unit 138 via the communication units 39 and 139. The movie recommendation server 102 can make movie recommendations using not only movies but also music-related property information. The present invention is applicable to a content recommendation system.
[Selection] Figure 13

Description

  The present invention relates to an information processing apparatus, method, and program, and in particular, uses user information about a predetermined type of content obtained from one device for recommending another type of content or recommending on another device. The present invention relates to an information processing apparatus, method, and program that can be performed.

  2. Description of the Related Art Conventionally, an invention for searching for content such as a television program and music based on the user's preference and recommending it to the user, an invention for so-called content personalization has been proposed (see, for example, Patent Document 1). .

  For content personalization, a technique called collaborative filtering (hereinafter also referred to as CF technique) and a technique called content-based filtering (hereinafter also referred to as CBF technique) are widely used.

  The CF method manages the purchase history of each user's content as user preference information, detects another second user whose purchase history is similar to the first user who wants to recommend the content, This is a method of recommending content that is purchased by the second user and not purchased by the first user. For example, it is used in a mail order site on the Internet.

The CBF method is a method in which metadata previously given to the content by the distribution side or the sales side is directly used for preference extraction or content recommendation. That is, a feature vector obtained by vectorizing various types of metadata is used as user preference information, and the distance (cosine correlation) between a feature vector indicating user preference information (also referred to as a user preference vector) and a feature vector of each candidate content And the like, and recommending the content with the short distance calculated as content that matches the user's preference.
JP 2004-194107 A

  However, in both the CF method and the CBF method, user preference information such as a user purchase history and a user preference vector is extracted from one device used by the user. In other words, each conventional device independently extracts user preference information. Therefore, there is a problem that user preference information obtained with a certain device cannot be used with another device at all.

  To solve this problem, an approach that unifies the schema of preference expressions can be considered, but for example, the format of metadata differs for each type of content such as television programs, songs, movies, etc. However, this is not practical because the device-side formats are different each time.

  The present invention has been made in view of such a situation, and the user information about a predetermined type of content obtained from one device is recommended for another type of content or recommended by another device. It is intended to be used in the future.

  An information processing apparatus according to an aspect of the present invention is an information processing apparatus that executes processing for a predetermined type of content, and includes history information that indicates a history of user behavior related to the processing for the predetermined type of content. A property generation unit that generates one or more property information representing the property of the user for the predetermined type of content, and a communication unit that transmits the one or more property information generated by the property generation unit. Prepare.

  The property generation means generates at least one of orientation, breadth, depth, and maturity as the property information.

  The property generation means generates the orientation based on the property of one or more contents grasped from the history information.

  The property generation means, based on the number of values and the entropy for each of the plurality of content attributes when each of the contents grasped from the history information is classified into a plurality of content attributes, Generate at least one of them.

  The property generation means uses a multi-viewpoint cluster as the plurality of content attributes.

  The property generation means determines the maturity based on the number of values and entropy for each of the plurality of content attributes when each of the one or more contents grasped from the history information is classified into a plurality of content attributes. Generate.

  The property generation means further generates the maturity based on the property of one or more contents grasped from the history information.

  The property generation means divides the history information by a period, and generates the maturity based on the history information for a predetermined period among the divided periods.

  The property generation means normalizes all of the one or more property information.

  The communication means transmits information indicating the predetermined type of content in association with one or more of the property information.

  It further comprises holding means for holding the property information in one or more generated by the property generation means, the communication means further receives one or more of the property information transmitted from another information processing apparatus, The holding unit further holds the one or more pieces of property information received by the communication unit as property information about the information processing apparatus itself.

  The one or more pieces of property information received by the communication unit are property information about content of a type different from the predetermined type to be processed by the information processing apparatus.

  The communication means transmits one or more of the property information in association with a free keyword.

  An information processing method according to an aspect of the present invention is an information processing method of an information processing apparatus that performs processing on a predetermined type of content, and stores a history of user behavior related to the processing on the predetermined type of content. One or more property information representing the property of the user for the predetermined type of content is generated based on the history information indicated, and the one or more property information generated is transmitted from the information processing device to another information processing device. Including a step of performing control to transmit to.

  A program according to one aspect of the present invention is a program corresponding to the information processing method according to one aspect of the present invention described above.

  In the information processing apparatus, method, and program according to one aspect of the present invention, an information processing apparatus that performs processing on a predetermined type of content is targeted, and the following processing is performed. That is, one or more property information indicating the property of the user for the predetermined type of content is generated based on history information indicating a history of user behavior related to the processing for the predetermined type of content, and the information It is transmitted from the processing device to another information processing device.

  As described above, according to the present invention, it is possible to obtain property information representing the property of the user from the history information that is the source of the user preference information, not the user preference information itself. This makes it possible to use this property information obtained from one device to recommend another type of content or recommend it on another device.

  Embodiments of the present invention will be described below. Correspondences between the configuration requirements of the present invention and the embodiments described in the detailed description of the present invention are exemplified as follows. This description is to confirm that the embodiments supporting the present invention are described in the detailed description of the invention. Accordingly, although there are embodiments that are described in the detailed description of the invention but are not described here as embodiments corresponding to the constituent elements of the present invention, It does not mean that the embodiment does not correspond to the configuration requirements. Conversely, even if an embodiment is described here as corresponding to a configuration requirement, that means that the embodiment does not correspond to a configuration requirement other than the configuration requirement. It's not something to do.

  Further, this description does not mean that all the inventions corresponding to the specific examples described in the embodiments of the invention are described in the claims. In other words, this description is an invention corresponding to the specific example described in the embodiment of the invention, and the existence of an invention not described in the claims of this application, that is, in the future, a divisional application will be made. Nor does it deny the existence of an invention added by amendment.

An information processing apparatus according to one aspect of the present invention (for example, the client device 1 in FIG. 1 or the client device 101 in FIG. 13)
In an information processing apparatus that executes a process for a predetermined type of content (for example, a process of playing music for the client device 1 in FIG. 1 and a process of playing a movie for the client device 101 in FIG. 13).
Based on the history information indicating the history of the user's behavior related to the processing for the predetermined type of content (for example, history information stored in the history storage unit 34 in FIG. 1 or the history storage unit 134 in FIG. 13), the predetermined A property generation means (for example, the property extraction unit 37 in FIG. 1 or the property extraction unit 137 in FIG. 13) that generates one or more property information representing the property of the user for the type of content;
A communication unit (for example, the communication unit 39 in FIG. 1 or the communication unit 139 in FIG. 13) that transmits one or more pieces of the property information generated by the property generation unit.

  The property generation means generates at least one of orientation, breadth, depth, and maturity as the property information (for example, in the example of FIG. 12, orientation o, width w, depth d). To generate a property vector D = (o, w, d).

  The property generation means is a property of one or more contents grasped from the history information (for example, in the example of FIG. 10, the music m1 to m4 are the contents targeted for the history by the user's behavior (viewing or purchase), That is, the content is grasped from the history information, and the orientation is generated based on the number of downloads of the music pieces m1 to m4).

  The property generation means is configured so that values for each of the plurality of content attributes when the contents grasped from the history information are classified into a plurality of content attributes (for example, in the example of FIG. Content that is subject to history due to behavior (viewing or purchase), that is, content that is grasped from history information, and these music pieces m1 to m4 are classified into one or more of each cluster indicating each content attribute On the basis of the number and entropy of small circles indicating the results, that is, small circles present in each cluster of View1 to View3, at least one of the width and the depth is generated.

  The property generation means uses a multi-viewpoint cluster (for example, each cluster shown as an ellipse in each of the three Views 1 to 3 in FIG. 10) as the plurality of content attributes.

  The property generation means is a value for each of the plurality of content attributes when each of the one or more contents grasped from the history information is classified into a plurality of content attributes (for example, in the example of FIG. Content that is subject to history due to the behavior (viewing or purchase) of the content, that is, content grasped from the history information, and the result that this music mn is classified into one or more of each cluster indicating each content attribute The maturity is generated on the basis of the number and entropy of small circles indicating, that is, small circles present in each cluster of View1 and View2.

A holding unit (for example, the property profile storage unit 38 of FIG. 1 or the property profile storage unit 138 of FIG. 13) that holds one or more of the property information generated by the property generation unit;
The communication unit further receives one or more property information transmitted from another information processing apparatus (see the communication unit 139 in FIG. 13),
The holding unit further holds the one or more property information received by the communication unit as property information about the information processing apparatus itself.

An information processing method according to one aspect of the present invention includes:
An information processing method of an information processing apparatus (for example, the client device 1 in FIG. 1 or the client device 101 in FIG. 13) that executes processing for a predetermined type of content,
One or more pieces of property information representing the property of the user for the predetermined type of content are generated based on the history information indicating the history of the user's behavior regarding the processing for the predetermined type of content (for example, the property of FIG. 1). The processing of the extraction unit 37 and the property extraction unit 137 of FIG. 13),
The generated one or more property information is transmitted from the information processing apparatus to another information processing apparatus (for example, the processing of the communication unit 39 in FIG. 1 or the communication unit 139 in FIG. 13).
An information processing method including steps.

  The program according to one aspect of the present invention is a program corresponding to the above-described information processing method according to one aspect of the present invention, and is executed by, for example, the computer of FIG.

  Thus, the present invention can handle content. Here, content is widely generated by human creative activities. For example, movies, music, theatre, literary arts, photographs, comics, animations, computer games and other characters, figures, colors, sounds, actions or videos, or combinations thereof, or information related thereto are provided via an electronic computer. Is a content example. However, in this specification, so-called content data, that is, data generated by human creative activities, which can be processed by the device, for example, an electric signal, or fixed in a memory Are collectively referred to as contents without being particularly distinguished.

  Embodiments of the present invention will be described below with reference to the drawings.

  FIG. 1 shows a configuration example of an information processing system to which the present invention is applied.

  In the example of FIG. 1, the information processing system includes a client device 1, a music recommendation server 2, and a music distribution management server 3.

  The number of client devices constituting the information processing system is only one of the client devices 1 in the example of FIG. 1, but is not limited to the example of FIG. 1, and may be any number as long as it is one or more. . In reality, there are a large number of users, and since the types of contents handled are various, there will be a large number of client devices.

  Further, in the example of FIG. 1, the servers constituting the information processing system are only the music recommendation server 2 and the music distribution management server 3. However, the server is not limited to the example of FIG. A corresponding server may be employed.

  That is, the type of content handled by the information processing system in the example of FIG. 1 is music (musical piece). In other words, the information processing system in the example of FIG. 1 is an information processing system capable of performing preference extraction particularly in a music download service. Therefore, in the example of FIG. 1, the music recommendation server 2 and the music distribution management server 3 are provided.

  In the following description, for the sake of simplicity, it is assumed that the music downloaded to the client device 1 and played is handled as a favorite music (hereinafter referred to as a favorite music) for the user who uses the client device 1.

  The music recommendation server 2 is configured to include a music item database 11 and a music recommendation engine 12.

  The music item database 11 stores a plurality of downloadable music pieces in association with the metadata.

  In the present embodiment, since the music recommendation server 2 also performs the music download service itself, the music item database 11 also stores the actual data of the music to be downloaded. In other words, the music download service is performed by another server, and the music recommendation server 2 can store only the music metadata in the music item database 11 when only music recommendation is performed.

  In addition, hereinafter, for the sake of simplicity of explanation, unless it is particularly noted, when it is expressed that music is transmitted, it means that the metadata of the music is also transmitted together with the actual data of the music.

  The music recommendation engine 12 selects a music to be recommended to the user who uses the client device 1 (hereinafter referred to as a recommended music) from a plurality of music stored in the music item database 11, and the client device 1 provide. For this reason, the music recommendation engine 12 includes a similarity calculation unit 21 and a property matching filter unit 22.

  The similarity calculation unit 21 selects a song similar to a preference profile stored in a later-described preference profile storage unit 36 of the client device 1, that is, a song similar to a user's favorite song of the client device 1 as a recommended song candidate ( Hereinafter, this is selected as a recommendation candidate). Note that the similarity calculation method of the similarity calculation unit 21 is not particularly limited. For example, a method is used in which the metadata of each music is vectorized and the similarity is calculated based on the Cosine distance between the vectors. it can.

  The property adaptive filter unit 22 will be described later.

  The music distribution management server 3 manages various information related to each music distributed from the music recommendation server 2. Specifically, for example, the music distribution management server 3 can manage the total number of downloads for each music. That is, as described above, there are actually a large number of client devices as well as the client device 1, and one music piece is individually distributed from the music recommendation server 2 to each of the many client devices. . Therefore, the music distribution management server 3 can count, for example, the number of individual distributions from the music recommendation server 2 for each music and manage the count as the total number of downloads of the music. The total number of downloads for each piece of music is provided from the music distribution management server 3 to the client device 1 as appropriate.

  The client device 1 is configured to include a strategy determination unit 31 to a communication unit 39.

  The strategy determination unit 31 determines the recommendation strategy of the recommended music distributed from the music recommendation server 2 according to the property profile stored in the property profile storage unit 38 to be described later, and notifies the recommendation presentation unit 32 of it. A specific example of the recommendation strategy will be described later.

  The recommendation presentation unit 32 presents the recommended music distributed from the music recommendation server 2 to the user based on the recommendation strategy determined by the strategy determination unit 31. More precisely, the presentation of the recommended music does not mean the presentation of the recommended music as a whole. For example, the presentation of a part of the recommended music or the presentation of information generated from the metadata of the recommended music Specifically, it means presentation by display of image information, presentation by voice output of voice information, and the like.

  When the user accepts the recommendation and gives an instruction to view the recommended music to the client device 1 (the block receiving the instruction is not shown), the reproducing unit 33 reproduces the instructed music.

  As described above, the music reproduced by the reproducing unit 33, that is, the music that the user accepted and viewed is regarded as the user's favorite music in the present embodiment. Therefore, information related to the favorite music, for example, metadata in the present embodiment, is accumulated in the history accumulation unit 34 as history information.

  The preference extraction unit 35 extracts user preference information from the history information accumulated in the history accumulation unit 34 and causes the preference profile accumulation unit 36 to accumulate the user preference information. Note that the user preference information extraction method is not particularly limited, and for example, a CBF (Content Based Filtering) method or the like can be employed.

  On the other hand, the property extraction unit 37 extracts one or more pieces of information (hereinafter referred to as property information) representing the user property of the music from the history information stored in the history storage unit 34, and stores it in the property profile storage unit 38. Accumulate.

  More generally, the property information will be described. One or more property information representing the property of the user for the predetermined type of content is generated from the history of the user's behavior regarding the processing of one device for the predetermined type of content. become.

  As the user property information, for example, orientation, breadth, depth, and maturity level can be adopted, and at least one of these property information represents the user property for a predetermined type of content. .

  Orientation refers to information indicating the popularity of the content that the user likes, specifically, for example, information indicating whether the content is highly popular or niche.

  The breadth and depth indicate the breadth and depth of the user's content experience (experience in purchasing and viewing content) when viewed in cluster units that are determined in advance as clusters for classifying content, for example, genre units. It is. The area is information indicating how much the content experience is concentrated locally. By grasping the size, it is possible to understand the possibility of how much content recommendation is allowed by the user. The depth is information indicating how much the user has experienced the contents belonging to the cluster when viewed in a certain cluster unit. Here, the cluster is a broad concept that includes not only a general genre but also one created so as to better express individual preferences. A specific example of the cluster will be described later.

  Maturity is an indicator that combines both breadth and depth. For example, a product obtained by appropriately multiplying the width and depth can be adopted as the maturity. Furthermore, it can be adopted as a maturity that takes into account the niche that is intended. For example, the maturity may be obtained by appropriately multiplying the niche degree by one or both of width and depth. Furthermore, it is possible to adopt as the maturity what is obtained as described above within the range narrowed down by the period and narrowed down to the period. Specifically, for example, what is obtained as described above by focusing on the content experience of the last year can be adopted as the maturity.

  The above definition is an example of a definition that captures the user's content experience distribution, and is a definition for grasping the distribution as orientation, breadth, depth, or maturity. It is.

  Therefore, depending on how the property is understood, property information based on another definition can be adopted. Specifically, for example, the property can be grasped not only in the content unit but also in the artist unit, and the property information defined by paying attention to this artist can also be adopted.

  Here, before describing a specific generation method of each property information, a history information clustering method necessary for the generation method will be described with reference to FIGS. 2 to 9.

  The property extraction unit 37 stores each item (title, artist name, genre, review text, tempo, beat, rhythm, etc.) of the metadata of all the songs stored in the history storage unit 34 in a cluster as shown in FIG. The music is classified (clustered) into any one of a plurality of clusters provided in the cluster layer into which the real information of each item is classified.

  One piece of music may be classified into a plurality of clusters. The distance between clusters existing in the same cluster layer (indicating the degree of similarity) is assumed to be known. This clustering method will be described later. Then, the property extraction unit 37 generates cluster information including cluster IDs (eg, CL11 in FIG. 2) of clusters in which actual information of each item of metadata is classified as information indicating the characteristics of music instead of metadata. Or stored in the history storage unit 34 or the like.

  If there is no cluster suitable for classification, a new cluster may be newly established. The size of each cluster is arbitrary and can include a plurality of music pieces. In addition, you may provide the cluster which can classify | categorize only a single music. In this case, the ID (artist ID, album ID, title ID) of the actual information of the music that can only be classified into the cluster ID of the cluster may be used.

  FIG. 3 shows an example of cluster information. In the figure, for example, the cluster information of the song with the song ID = ABC123 is a cluster ID (CL12, CL21, CL35, CL47, CL52,..., CLn2). Further, for example, the cluster information of the song with the song ID = CTH863 indicates a cluster ID (CL11, CL25, CL31, CL42, CL53,..., CLn1).

  FIG. 4 shows an example of cluster-music ID information corresponding to the cluster information shown in FIG. In the figure, i indicates the i-th layer of the cluster layer. In the figure, for example, the song ID = CTH863 corresponds to the cluster ID = CL11. Further, for example, it is indicated that the song ID = ABC123 corresponds to the cluster ID = CL21.

Further, clustering of music (metadata) by the property extraction unit 37 will be described below.

  The clustering method may be any method, but an optimum clustering method and distance measure are selected for each cluster layer. For example, if the actual metadata information is a numerical value, in the case of a title or the like, it is converted into a numerical value using a quantification method such as principal component analysis, and a distance measure such as Euclidean distance is defined and clustered. . Typical clustering methods include K-means method and hierarchical clustering method.

  At this time, it is desirable to implement by clustering (for example, constrained clustering) reflecting the preference distance. For that purpose, a partial correct answer collection (a set of real information that is close to preference, a set of distant real information, etc.) is created by a preliminary survey, and numerical expressions, distances, and clustering methods that match it are used. Furthermore, it is desirable to select a clustering method (that is, a clustering method having different characteristics) that increases the independence of each cluster layer to be formed.

For example, a method of selecting two types of clustering methods having different characteristics from four types of clustering methods (hereinafter referred to as first to fourth methods) will be described with reference to FIGS.

  First, the property extraction unit 37 clusters artists A to J, which are actual metadata information, using the first to fourth methods. Assume that the results shown in FIG. 5 are obtained.

  That is, artist A through C are clustered in cluster CL1, artist D through G are clustered in cluster CL2 and artist H through J are clustered in cluster CL3 by the first method, and artists A and B are clustered in cluster CL1 by the second method. Artists C to F are clustered in cluster CL2 and artists G to J are clustered in cluster CL3. According to the third method, artists A, D, G and J are clustered in cluster CL1, artists B, E and H are clustered in cluster CL2. C, F, and I are clustered in cluster CL3. According to the fourth method, artists D, I, and J are clustered in cluster CL1, artists E to G are clustered in cluster CL2, and artists A to C and H are clustered in cluster CL3. To do.

  In this case, the overlapping rate (%) of the results obtained by the first to fourth methods is as shown in FIG. That is, the overlap rate of the first method and the second method is 0.8, the overlap rate of the first method and the third method is 0.3, the overlap rate of the first method and the fourth method is 0.4, and the second method The overlap rate of the third method is 0.3, the overlap rate of the second method and the fourth method is 0.3, and the overlap rate of the third method and the fourth method is 0.4.

  Since the characteristics of the two methods are considered to be different as the overlap rate shown in FIG. 6 is smaller, the combination in which the overlap rate is the minimum value of 0.3, that is, the combination of the first method and the third method, the second It is desirable to employ a combination of the method and the third method, or a combination of the second method and the fourth method.

  On the other hand, when the user himself / herself determines whether or not two of the artists A to J should be classified into the same cluster, it is assumed that a result as shown in FIG. 7 is obtained. However, in the figure, 1 means that they should be classified into the same cluster, and 0 means that they should be classified into different clusters. That is, in the figure, for example, it is shown that it is determined that artist A should be classified into the same cluster as artists B, C, F, H, and I, and artist B has artists C and D. , E, J should be classified into the same cluster.

  If it is assumed that the result shown in FIG. 7 is an ideal clustering result, the correct rate of the first to fourth methods described above is as shown in FIG. That is, the accuracy rate of the first method is 62.2%, the accuracy rate of the second method is 55.6%, the accuracy rate of the third method is 40.0%, and the accuracy rate of the fourth method is 66.7%. is there.

  Therefore, if importance is attached to the correct answer rate, it is desirable to adopt a combination of the first method and the fourth method having a high correct answer rate.

  Furthermore, in order to obtain a combination of clustering methods that consider the overlap rate and the correct answer rate, the correct answer overlap rates of the first to fourth methods are calculated as shown in FIG. A method with an extremely low accuracy rate is identified from the results of the accuracy rate of each method shown in FIG. 8, and the combination with the lowest accuracy rate overlapping rate among the combinations not including the identified method is shown in FIG. What is necessary is just to employ | adopt the specified combination specified from the result shown. That is, as a technique with an extremely low correct answer rate, the third technique with the correct answer rate of 40.0% in FIG. 8 is identified, and the correct answer duplication rate among the combinations not including the third technique is the lowest. The combination of the second method and the third method of 24.4% of 9 is selected.

It should be noted that an absolute threshold value may be specified for the duplication rate and the correct answer rate described above, and a method that cannot satisfy the threshold value may be excluded, or two indicators may be used to adopt a balanced method. For example, based on (overlapping rate and correct answer rate), for example, a comprehensive index such as the following two examples may be created, and a combination of clustering methods may be selected based on the comprehensive index.
Comprehensive index = accuracy rate x (1-overlap rate)
Comprehensive index = α × accuracy rate × β (1-duplication rate) (α and β are predetermined coefficients)

  Thus, by using the result of the clustering of music (metadata) by the property extraction unit 37, it is possible to grasp the content attribute by moving the viewpoint to each cluster layer. That is, it can be understood that each cluster belonging to each cluster layer indicates a plurality of content attributes corresponding to the cluster layer.

  In other words, it can be said that the clustering of music (metadata) by the property extraction unit 37 described above is a classification into a plurality of clusters from multiple viewpoints. Therefore, hereinafter, each cluster layer is appropriately referred to as “View”. However, the cluster of each view may be hereinafter referred to as a multi-viewpoint cluster as necessary, or may be referred to as a cluster as it is.

  When the property extraction unit 37 performs clustering of music (metadata) in this manner, the property extraction unit 37 generates each property information by using the result (hereinafter referred to as a multi-viewpoint clustering result).

  Therefore, specific examples of the method for generating each property information will be described below.

  However, for the sake of simplicity, the description will be made using the multi-viewpoint clustering result shown in FIG. That is, as shown in FIG. 10, it is assumed that four songs m1 to m4 indicated by small circles are clustered for each of View1 to View3 each having eight clusters (ellipses in the figure). I will explain. However, as shown in FIG. 10, in View1 and View3, each piece of music m1 to m4 is represented by two clusters, while in View2, each piece of music m1 to m4 is one cluster (hard clustering). It is expressed. Thus, since one music piece can be expressed by a plurality of clusters, in the following description, not the number of music pieces (4 in the example of FIG. 10) but the values (circles in the example of FIG. 10) included in each cluster. It is assumed that a number (hereinafter referred to as the number of music entries) is used.

  First, I will explain how to generate a preference (hereinafter also referred to as o).

  Here, the log of the number of downloads of the music at a certain point in time is normalized (0 to 1.0) and defined as the major orientation degree of the music. In the present embodiment, the number of downloads of music is provided from the music distribution management server 3 of FIG. 1 to the property extraction unit 37 via the communication unit 39 as described above.

  In this case, the property extraction unit 37 takes the user's major orientation as an average of the major orientations of the user's downloaded and played music (in the example of FIG. 1, the music in which the metadata is stored in the history storage unit 34). The major orientation of the user is determined as orientation o.

  It is assumed that the normalization term in this case is adjusted so that the major orientations of all users have a normal distribution. Here, “all users” means all users including a user who uses the client device 1 of FIG. 1 and a large number of users who respectively use a large number of client devices not shown in FIG. However, in practice, the normalization term is actually obtained by sampling. The reason for normalization is to match other property information, that is, scales such as width, depth, and maturity, which will be described later. That is, as will be described later, other property information is also normalized.

  Specifically, for example, in the example of FIG. 10, the number of downloads of the music m1 is 1024, the number of downloads of the music m2 is 8, the number of downloads of the music m3 is 64, and the download of the music m4 is downloaded. It is assumed that the number is 512 and the normalization term is 16 (the number of downloads of the most downloaded music among all the music was 65536).

  In this case, the orientation o is calculated as in the following equation (1). In addition, although the bottom of log described by each of the following formulas is 2, this bottom is omitted.

  Note that the above-described method for generating the orientation o is merely an example, and as the method for generating the orientation o, for example, as a major orientation degree of the song, the famous degree of famous music may be adopted, You may employ | adopt what normalized the ranking, such as sales of a music.

  More generally speaking, the major orientation of a song represents the nature of the song. In other words, the generation method of orientation o is not particularly limited as long as it is a method of generating based on the properties of each content.

  Next, a method for generating an area (hereinafter also referred to as w) will be described.

The property extraction unit 37 obtains entropy for each view of the multi-viewpoint clustering result. The number of song entries S _vi belonging to the cluster vi (i represents the cluster number) of the view v (v represents the layer number, and in the example of FIG. 10 is any one of 1 to 3). Is divided by the total number of music entries S = ΣS _vi to be P _vi , the entropy E _{v of} View _v is as shown in the following equation (2). In the example of FIG. 10, in View v, the number of the cluster at the upper left corner is set to 1, and numbers 2 to 4 are respectively assigned to the clusters arranged sequentially in the right direction. , And 6 to 8 are assigned to each cluster sequentially arranged in the right direction.

However, if the number of song entries is 0 (P _vi = 0), a certain minute value (for example, the square of n) corresponding to the number of cluster types n is entered, and P _vi and E _v are expressed by the following formulas (3 ) To formula (5).

The nature extraction unit 37, a normalized value of minimum entropy E _v-min in entropy E _v, obtained as breadth w. It is assumed that the normalization term is adjusted so that the maximum entropy E _max determined by the number of clusters is 1.0.

Specifically, for example, each of entropies E _{1 to} E ₃ of View _{1 to View 3} in FIG. 10 is obtained as shown in Equations (6) to (8). Therefore, the minimum entropy E _v-min is as shown in Expression (9), and the maximum entropy E _max is as shown in Expression (10). As a result, the width w is as shown in Expression (11).

  Next, a method for generating a depth (hereinafter also referred to as d) will be described.

In this case, the property extraction unit 37 specifies the view having the minimum entropy E _v-min used for calculating the width w, and among the clusters in the view, the song entry of the cluster in which P _vi has the maximum value. A value obtained by normalizing the number S _vi-max is calculated as the depth d.

Specifically, for example, the maximum track number of entries S _vi-max for all users is 100, since View3 in the example of FIG. 10 has a minimum entropy E _v-min, tracks the number of entries S _vi-max the following formula As shown in (12), as a result, the depth d is as shown in the following equation (13).

  Note that the above-described generation method of the width w and the depth d is merely an example, and for example, the following generation method of the width w and the depth d can be employed.

  That is, in the above-described example, the object mapped to the multi-viewpoint cluster is the music stored in the history storage unit 34 (hereinafter also referred to as user-owned music), but the music of the music recommendation server 2 in FIG. The above-described width w and depth d may be calculated using the result of mapping all music pieces in the item database 11 to the multi-viewpoint cluster.

However, when all the music pieces are mapped to the multi-viewpoint cluster, the number of affiliations to each multi-viewpoint cluster has a bias, and the degree of the bias differs depending on the view. Therefore, the property extraction unit 37 may perform normalization between views based on the cluster affiliation ratio when all music pieces are mapped, and obtain the entropy E _v using this normalized value. Specifically, for example, when the number of affiliations of cluster vi in a certain view v is n _i and the total number of all affiliations is N, when P _vi × (N / n _i ) is normalized, P ' _vi and Therefore, the property extraction unit 37 may obtain the entropy E _v using this value P ′ _vi . That is, the nature extraction unit 37 obtains a P _vi using music entry number S _vi user possession music, may determine the P _'vi using the P _vi.

That is, the property extraction unit 37 uses the value P ′ _vi calculated by the following equation (14) instead of the above-described value P _vi as the value for obtaining the entropy E _{v of} View v, thereby entropy. E _v can also be obtained.

  Specifically, for example, it is assumed that the result of mapping all musical pieces to the multi-viewpoint cluster is as shown in FIG. That is, as shown in FIG. 11, it is assumed that two or more music pieces including the user-owned music piece mn are clustered with respect to two views 1 and 2 each having two clusters (ellipses in the figure). However, in the following description, it is assumed that the number of music entries, that is, the number of small circles in FIG. 11 is used.

In this case, the entropy E ₁ of View1 would be computed as shown in equation (15) to (17).

On the other hand, the entropy E ₂ of View2 would be computed as shown in equation (18) to (20).

Therefore, E ₁ <E _{2 in} the example of FIG.

  The above-described method for generating the width w and the depth d is also merely an example. That is, more generally speaking, the generation method of the width w and the depth d is a plurality of contents when each of the contents grasped from the user's history is classified into a plurality of content attributes (multi-viewpoint clusters in the above example). A method of generating at least one of the width w and the depth d based on the number of values for each content attribute and entropy is applicable, and some specific examples of such a method are described above. It will be done.

  Next, a method for generating maturity (hereinafter also referred to as m) will be described.

That is, the entropy E _v is calculated by the normalization calculation between the views using the multi-view clustering result (for example, the result shown in FIG. 11) for all the music described above, and as a result, the minimum entropy E _v-min is calculated. Suppose that the View you have becomes View vo.

In this case, in this view vo, when the total number of affiliations in the cluster vo-i is na _i , the number of affiliations of user-owned music is nu _i , and the total number of clusters is nc, the property extraction unit 37 The maturity m can be calculated according to the equation (21).

  Specifically, for example, in the example of FIG. 11, the maturity m is calculated as shown in the following equation (22).

  m = {(1/3) + (1/1)} / 2 = 2/3 (22)

  If all the music pieces are user-owned music pieces, m = 1.0.

  The above-described method for generating maturity m is also merely an example. That is, as described above, the maturity m is an index having both the width w and the depth d. Therefore, as described above, as the generation method of the width w and the depth d, a plurality of contents when each content grasped from the user history is classified into a plurality of content attributes (multi-viewpoint clusters in the above-described example). A method of generating at least one of the width w and the depth d based on the number of values for each content attribute and the entropy is applicable. Therefore, for example, the maturity m is also generated based on the number of values and entropy for each of the plurality of content attributes when each of the contents grasped from the user's history is classified into a plurality of content attributes. It is also possible to adopt a method such as.

  Furthermore, as described above, the maturity m may be one that has both the width w and the depth d, and the niche that is the orientation o is added. In this case, for example, in consideration of the fact that, as described above, the method for generating orientation o based on the properties of each content can be adopted as the method for generating orientation o, the method described above is used as the method for generating maturity m. In addition to the above, it is also possible to adopt a method in which a method of generating the maturity m based on the property of one or more contents grasped from the history information is added.

  Alternatively, for example, as described above, the maturity m can be obtained within a range narrowed to a predetermined time. That is, for example, when a user's history is divided by a period, and each of the contents grasped from the user's history for a predetermined period among the divided periods is classified into a plurality of content attributes, a value for each of the plurality of content attributes A method of generating the maturity m based on the number and entropy can also be adopted.

  In this way, the property extraction unit 37 can calculate each property information. Therefore, the property extracting unit 37 further includes a vector (hereinafter referred to as a property vector) having one or more of the respective property information calculated in this way, for example, orientation o, width w, depth d. The property vector D = (o, w, d) is generated, and the property vector D can be stored in the property profile storage unit 38 as the user property profile. For example, in the example of FIG. 10 described above, the property vector D = (o, w, d) = (0.4375, 0.37, 0.04) is calculated and stored in the property profile storage unit 38.

  The dimension of the property vector and each element are not particularly limited to the property vector D. For example, a property vector having maturity m as one element may be employed instead of the width w and the depth d.

  Further, in the above-described example, the property vector D is generated by handling all music equally, but it is not particularly necessary to treat all music equally. For example, the property extraction unit 37 generates the property vector for each genre by performing the above-described series of processing for each genre of music, and accumulates them in the property profile storage unit 38 as the property profile of the user. You can also.

  In the following description, for simplicity of explanation, the property extraction unit 37 does not use values of the orientation o, the width w, and the depth d itself, but binarizes those values with an appropriate threshold value. It is assumed that a property vector D having each value after binarization as each element is generated and stored in the property profile storage unit 38.

  In this case, for the orientation o, 1 is a value indicating “major orientation”, and 0 is a value indicating “niche orientation”. Regarding the width w, 1 is a value indicating “wide”, and 0 is a value indicating “narrow”. Regarding the depth d, 1 is a value indicating “deep”, and 0 is a value indicating “shallow”.

  For example, in the example of FIG. 10 described above, as each property information, o = 0.4375, w = 0.37, d = 0.04 is calculated. Therefore, when the threshold is 0.5, the property vector D = (0,0,0) = (Niche orientation, narrow, shallow) is obtained and stored in the property profile storage unit 38.

  Here, it is assumed that the property matching filter unit 22 of the music recommendation engine 12 in FIG. 1 has, for example, information shown in FIG. 12 as information for interpreting the meaning of the property vector D. In this case, the property matching filter unit 22 can easily interpret the meaning of the property vector D stored in the property profile storage unit 38 using the information of FIG.

  As a result, the property matching filter unit 22 selects (filters) a song that matches the meaning of the property vector D from the one or more recommended candidates determined by the similarity calculation of the similarity calculation unit 21 as a recommended song. The selected recommended music can be provided to the client device 1.

  Alternatively, depending on the meaning of the property vector D, the property matching filter unit 22 can separately determine a song that is not a recommendation candidate as a recommended song. For example, if o = major-oriented, even if the music is not similar to the user preference information (music that has not been selected as a recommendation candidate), the property-adapting filter unit 22 has a high-majority-oriented music such as the current hit. The music in the middle can be determined as the recommended music.

Specifically, for example, in the example of FIG. 10, the property vector D = (niche orientation, narrow, shallow) is obtained and stored in the property profile storage unit 38. Therefore, the property adaptation filter unit 22 can determine that the “user property of the music”, which is meant by the property vector D, is basically “niche” from the information in FIG. Interpretation that it may be the result of "coincidence?" In this case, for example, the property matching filter unit 22 is based on this interpretation, and is one or more of one or more recommendation candidates determined by the similarity calculation of the similarity calculation unit 21, that is, one or more similar to the user preference information. Of these music pieces, music pieces with a low major orientation can be provided to the client device 1 as recommended music pieces. However, in consideration of the fact that d = shallow, for example, the property adaptive filter unit 22 basically performs selection (filtering) within a range that does not remove a view cluster having the minimum entropy E _v-min .

  For example, when the property vector D = (niche-oriented, wide, deep) is accumulated in the property profile accumulation unit 38, the property adaptation filter unit 22 can execute the following processing. That is, the property matching filter unit 22 means that the property vector D means “user property about music” is basically “like niche” from the information of FIG. It is interpreted as “broad range of defense” and “greedy” for watching music. In this case, for example, the property matching filter unit 22 is based on this interpretation, and is one or more of one or more recommendation candidates determined by the similarity calculation of the similarity calculation unit 21, that is, one or more similar to the user preference information. Of these music pieces, music pieces with a low major orientation can be provided to the client device 1 as recommended music pieces. However, in this case, in consideration of the fact that w = wide and d = deep, for example, the property-adapting filter unit 22 basically performs selection (filtering) so as to skip between clusters having a history appropriately. .

  In this way, when the recommended music based on the interpretation of the property vector D = (niche-oriented, wide, deep) has been provided to the client device 1, the strategy determination unit 31 described later is suitable for the recommended music, for example. It is possible to determine a recommendation strategy to present with storage information. Further, when the user accepts the recommendation, that is, when the recommended music is reproduced and stored in the history storage unit 34, the property matching filter unit 22 further selects a music with a low major orientation in the cluster. By selecting as a recommended song, the user's tendency to “be deep” can be satisfied.

  For example, when the property vector D = (major-oriented, wide, shallow) is stored in the property profile storage unit 38, the property matching filter unit 22 can execute the following processing. In other words, the property adaptation filter unit 22 means that the property vector D means “user property of music” is basically a major orientation, that is, “Meher” from the information of FIG. Interprets as “selling OK”. In this case, for example, the property matching filter unit 22 is based on this interpretation regardless of the recommendation candidate determined by the similarity calculation of the similarity calculation unit 21, that is, regardless of the music similar to the user preference information. It is possible to provide the client device 1 with a major-oriented music such as a hit music in addition to the recommended music. In addition, considering the fact that w = wide and d = shallow, for example, the number of recommended songs selected from recommended candidates is limited to about 20% of the recommended music. . In other words, the remaining 80% of recommended songs are highly-oriented music such as hit songs.

  Thus, when the recommended music corresponding to the user's property vector D is provided from the music recommendation server 2 to the strategy determining unit 31 of the client device 1, the strategy determining unit 31 selects the user's property vector D, that is, Using the property vector D stored in the property profile storage unit 38, the recommended music presentation strategy can be determined.

  The presentation method is not particularly limited. For example, if the information shown in FIG. 12 can be used in the same manner as the property-adapting filter unit 22, the strategy determining unit 31 determines the property vector D from the information shown in FIG. The method of interpreting the meaning of and determining the strategy according to the interpretation can be adopted. Specifically, for example, in the case of the property vector D = (o, narrow, deep), the strategy determination unit 31 may determine the strategy of continuously presenting the same kind of music in a concentrated, narrow and deep manner. it can. On the other hand, for example, in the case of the property vector D = (o, wide, shallow), the strategy determination unit 31 can determine the strategy of continuously presenting various musical pieces as wide and shallow as possible.

  The property information described above, that is, the orientation o, the width w, the depth d, and the maturity m can be transferred from the client device 1 to another device. The transfer protocol in this case is not particularly limited. For example, in general, SOAP (Simple Object Access Protocol) communication, delivery by Memory Stick (trademark), and the like can be considered, but a standard defined between devices may be used. Furthermore, when transferring to other devices, a simple property vector is not transferred, but in addition to the property information, a vector (hereinafter referred to as transfer) having information encoded as a content type, such as music and movies. It is preferable to transfer a property vector SD).

  Therefore, hereinafter, information in which the content type is coded is described as kind, and as a transfer property vector SD = (o, w, d, kind), from the client device 1 which is a music playback device to another device, for example, An example of transferring to a movie playback device will be described.

  In this case, it is assumed that the movie playback device is configured as, for example, the client device 101 in FIG. The client device 101 is configured to include a strategy determination unit 131 through a communication unit 139. Each of the strategy determination unit 131 to the communication unit 139 has basically the same function and configuration as each of the strategy determination unit 31 to the communication unit 39 (FIG. 1) of the client device 1.

  However, the type of content to be processed, that is, the type of content to be recommended or reproduced, is music (music) in the client device 1, but is different in that it is a movie in the client device 101.

  Accordingly, the client device 101 receives a recommended movie from the movie recommendation server 102. The movie recommendation server 102 includes a movie item database 111 and a movie recommendation engine 112 including a similarity calculation unit 121 and a property matching filter unit 122. Each of the movie item database 111 and the movie recommendation engine 112 including the similarity calculation unit 121 and the property adaptation filter unit 122 includes the song item database 11 of the song recommendation server 2, the similarity calculation unit 21, and the property adaptation filter unit 22. It has basically the same function and configuration as each of the music recommendation engines 12 (FIG. 1). However, the type of content to be processed, that is, the type of content to be recommended is music (music) in the music recommendation server 2, whereas the movie recommendation server 102 is a movie.

  In this case, the client device 1 transfers the property information stored in the property profile storage unit 38, that is, the transfer property vector SD = having the orientation o, the width w, the depth d, and kind = music. (o, w, d, kind) can be generated and transferred to the client device 101 via the communication unit 39.

  The communication unit 139 on the client device 101 side receives this transfer property vector SD and stores it in the property profile storage unit 138. Thereby, the client device 101 which is a movie playback device uses the property information obtained on the client device 1 side which is the music playback device, that is, the property information indicating the property of the user about the music piece, for movie recommendation or the like. become able to.

  Hereinafter, as a specific example, three elements of property information (hereinafter referred to as property tmp) of the transfer property vector SD = (o, w, d, kind = music) transferred from the client device 1 to the client device 101 will be described. ) Is an example where o = niche orientation, w = narrow, and d = deep. Also, it is assumed that a major degree of orientation is given to a movie, as is the case with music (music).

  In this case, the movie recommendation server 102 recommends, for example, the following recommendation according to the amount of history information stored in the client device 101, that is, the number of movies viewed by the user using the client device 101. It can be carried out.

  First, an example of recommending a movie in a state in which the amount of history information stored in the client device 101, that is, the history information stored in the history storage unit 134 is small will be described.

  The property extraction unit 137 of the client device 101 can also extract history information about a movie from the history storage unit 134 and use the history information to generate (calculate) various property information about the movie. Hereinafter, the property information generated on the client device 101 side in this way, that is, the property represented by the orientation o, the width d, the depth d, etc. about the movie is referred to as a property B. This property B is also stored in the property profile storage unit 138.

  As described above, since the property B is generated based on the history information of the client device 101, the property B generated when the accumulated amount of the history information is small can sufficiently reflect the tendency of the user to watch the movie. It can be judged that it is not.

  Therefore, in such a case, the movie recommendation engine 112 of the movie recommendation server 102 uses the property tmp stored in the property profile storage unit 138 of the client device 101, that is, the property B indicating the tendency of the user to watch the movie. The recommended movie can be determined from the recommended candidates or other than the recommended candidates by using the property tmp indicating the propensity to view music instead of.

  The recommendation candidate here refers to the similarity of the movie recommendation engine 112 using the preference profile stored in the preference profile storage unit 136 of the client device 101, that is, user preference information about the movie (hereinafter referred to as preference B). A movie determined by the degree calculator 121. That is, the recommendation candidate here refers to a movie similar to the preference B.

  More specifically, for example, when one favorite movie is selected by the client device 101, that is, a user's favorite movie playback instruction is given, and as a result, the movie is played by the playback unit 133. When the meta information of the movie is accumulated in the history accumulation unit 134, the preference extraction unit 135 generates a preference B based on the selected movie and accumulates the preference B in the preference profile accumulation unit 136.

  Then, the similarity calculation unit 121 of the movie recommendation engine 112 selects a movie similar to the preference B from the movies stored in the movie item database 111, that is, one preference first selected by the client device 101. A movie similar to the movie is determined as a recommended candidate.

  Then, the property matching filter unit 122 of the movie recommendation engine 112 uses the property tmp stored in the property profile storage unit 138 of the client device 101, that is, music instead of the property B indicating the tendency of the user to watch the movie. Using the property tmp indicating the viewing tendency, it is possible to determine a recommended movie from among the recommended candidates or from other candidates.

  Here, the property tmp is o = niche oriented, w = narrow, d = deep (hereinafter, such an expression is described as a property tmp = niche oriented & narrow & deep). The unit 122 interprets that the user is likely to be niche-oriented and narrow and deep in movie viewing as well as music viewing, and based on this interpretation, for example, the client device 101 first selects from among the recommended candidates It is possible to determine a movie of the same genre as the selected favorite movie as a recommended movie, or to determine a movie with the same performer and a low major orientation as a recommended movie.

  In this way, when the recommendation by the movie recommendation engine 112 is performed using the preference B and the property tmp, such recommendation is hereinafter referred to as a recommendation based on the preference B and the property tmp.

  When the user accepts the recommendation, that is, the recommended movie determined by the recommendation is provided to the client device 101 and presented to the user by the recommendation presenting unit 132 with the recommendation strategy determined by the strategy determining unit 131. When a reproduction instruction is given by the user who received the presentation and the reproduction unit 133 reproduces the content, it can be determined that the user's propensity to view music and the propriety to view a movie are similar.

  Therefore, every time a recommendation is accepted by the user, that is, every time another favorite movie is selected on the client device 101, the movie recommendation engine 112 makes a recommendation based on the preference B and the property tmp, that is, the attribute of the property tmp. You can repeat the recommendation of movies similar to. By repeating such a recommendation, the recommendation tmp = niche orientation & narrow & deep, that is, the deep recommendation in a narrow range is realized.

  On the other hand, when the user does not accept the recommendation, that is, the recommended movie determined by the recommendation is provided to the client device 101, and the recommendation presenting unit 132 with the recommendation strategy determined by the strategy determining unit 131 is provided. However, if the playback instruction is not given by the user and, as a result, playback by the playback unit 133 is not performed, there is a possibility that the user's propensity for viewing music and the propensity for viewing movies are dissimilar. I can judge that there is.

  In other words, if the user has not accepted the recommendation several times, for example, if the number of times the recommendation is not accepted is described as miss, and the miss is greater than or equal to a predetermined number (5 times, etc.), the user It can be determined that the propensity to watch music and the propensity to watch movies are dissimilar.

  Therefore, in such a case, the movie recommendation engine 112 performs a recommendation using only the preference B (hereinafter referred to as a recommendation based on the preference B) without using the property tmp.

  Regardless of the recommendation based on the preference B and the property tmp and the recommendation based on the preference B, every time the recommendation is accepted by the user, that is, another favorite movie is selected on the client device 101. Each time, the amount of history information stored in the client device 101 increases, and the property B is updated accordingly. That is, the property B is information that gradually reflects the user's propensity to watch movies.

  Accordingly, when the amount of history information stored in the client device 101 exceeds a certain value, for example, when the number of history information (hereinafter referred to as history number) exceeds a certain number (for example, 20), the movie recommendation server 102 The movie recommendation engine 112 uses the property B stored in the property profile storage unit 138 of the client device 101, that is, the property B indicating the tendency of the user to watch the movie, from among the recommended candidates or from other than that, A recommended movie can be determined.

  The recommendation candidate itself in this case is also determined by the similarity calculation unit 121 of the movie recommendation engine 112 using the preference B stored in the preference profile storage unit 136 of the client device 101.

  More specifically, for example, the similarity calculation unit 121 of the movie recommendation engine 112 determines a movie similar to the preference B among the movies stored in the movie item database 111 as a recommendation candidate.

  Then, the property matching filter unit 122 of the movie recommendation engine 112 uses the property B stored in the property profile storage unit 138 of the client device 101, that is, using the property B indicating the tendency of the user to watch the movie, A recommended movie is determined from among the recommended candidates or from other candidates.

  In this case, for example, when the property B and the property tmp are the same, that is, when the tendency of the user to watch music and the tendency to watch the movie are substantially the same, in this example, the property B = niche orientation & narrow & In the deep case, the property matching filter unit 122 can determine the recommended movie as follows. That is, the property matching filter unit 122 interprets that the user has a tendency of being niche-oriented and narrow and deep in movie viewing as well as music viewing, and based on the interpretation, for example, the major orientation degree is selected from the recommended candidates. A low movie can be determined as a recommended movie.

  On the other hand, when the property B is different from the property tmp, that is, when the tendency of the user to watch music and the tendency to watch the movie are dissimilar, specifically, for example, the property B = major orientation & wide & shallow In this case, the user can interpret it as having a tendency to have a particular interest in watching music but not to watch a movie. Therefore, based on this interpretation, the property matching filter unit 122 includes, for example, general hits in the recommended movie separately from the recommended candidates, and recommends preference candidates (movies similar to the preference B) to the total recommended number. For example, it is possible to make a recommendation to stop at about 20%.

  In this way, when the recommendation by the movie recommendation engine 112 is performed using the preference B and the property B, such recommendation is hereinafter referred to as a recommendation based on the preference B and the property B.

  A detailed example of a series of processing (hereinafter referred to as movie recommendation processing) by the movie recommendation server 102 and the client device 101 is shown in the flowchart of FIG.

  In step S1, the client device 101 determines whether or not a favorite movie has been selected.

  If it is determined that the favorite movie has not been selected, the determination process of step S1 is executed again. That is, until the favorite movie is selected on the client device 101 side, the determination process in step S1 is repeated. After that, if the favorite movie is selected on the client device 101 side, YES in step S1. If it is determined that there is, the process proceeds to step S2.

  In this case, a favorite movie is reproduced by the reproduction unit 133, and the meta information is accumulated in the history accumulation unit 134 as history information. Therefore, in step S <b> 2, the preference extraction unit 135 of the client device 101 calculates the preference B using the history information and accumulates the preference B in the preference profile accumulation unit 136.

  In step S3, the movie recommendation engine 112 of the movie recommendation server 102 sets the property tmp. As described above, the property tmp is a property generated by the client device 1 that is a music playback device, that is, a property of the user regarding the music, and is transferred from the client device 1 and is a property on the client device 101 side. It is stored in the profile storage unit 138. Here, setting the property tmp means that the movie recommendation engine 112 reads the property tmp from the property profile storage unit 138, transfers it to itself, and holds it.

  In step S4, the movie recommendation engine 112 determines whether or not the history number of the client device 101 is less than 20. The management of the number of histories may be performed on the client device 101 side or on the movie recommendation server 102 side. Here, it is assumed that the history number is managed by the playback unit 133 of the client device 101 and is notified to the movie recommendation engine 112 by communication or the like (not shown in FIG. 13).

  Note that the threshold 20 serving as the determination criterion in step S4 is merely an example, and another threshold may be employed.

  When the number of histories does not exceed 20, it is determined as YES in Step S4, and the process proceeds to Step S6.

  In step S6, the movie recommendation engine 112 determines whether or not miss (a value indicating the number of times the user has not accepted a recommendation) is less than five. This miss management may be performed on the client device 101 side or on the movie recommendation server 102 side. Here, miss is managed by the playback unit 133 of the client device 101 together with the number of histories described above, and is notified to the movie recommendation engine 112 by communication or the like (not shown in FIG. 13).

  Note that the threshold 5 serving as the determination criterion in step S6 is merely an example, and another threshold may be employed.

  If the number of misses is less than 5, it is determined as YES in step S6, and the process proceeds to step S8. In step S8, the movie recommendation engine 112 makes a recommendation based on the preference B and the property tmp described above.

  On the other hand, when the number of misses is 5 or more, it is determined as NO in step S6, and the process proceeds to step S7. In step S7, the movie recommendation engine 112 makes a recommendation based on the preference B described above.

  If the number of histories is 20 or more, it is determined as NO in step S4, and the process proceeds to step S5. In step S5, the movie recommendation engine 112 makes a recommendation based on the preference B and the property B described above.

  When the recommended movie determined by the recommendation process in any one of steps S5, S7, and S8 is provided to the client device 101 and presented by the recommendation presenting unit 132 with the recommendation strategy determined by the strategy determining unit 131, The process proceeds to step S9.

  In step S9, the playback unit 133 of the client device 101 determines whether the recommendation has been accepted.

  If the user has not instructed to play the recommended movie, the playback unit 133 determines in step S9 that the recommendation has not been accepted, and increments miss by 1 in step S10. Thereafter, the process returns to step S4, and the subsequent processes are repeated.

  On the other hand, when the user gives a playback instruction for the recommended movie, the playback unit 133 determines in step S9 that the recommendation has been accepted, plays the recommended movie, and stores the meta information of the recommended movie as a history. Stored in the unit 134. Then, the process proceeds to step S11.

  In step S11, the preference extraction unit 135 calculates the preference B based on the history information of the history storage unit 134 including the meta information of the recommended movie whose recommendation has been accepted in the process of the previous step S9, and stores the preference profile. Stored in the unit 136. That is, the content of preference B is updated.

  In step S12, the property extraction unit 137 calculates the property B based on the history information of the history storage unit 134 including the meta information of the recommended movie whose recommendation has been accepted in the process of the previous step S9. The profile is accumulated in the profile accumulation unit 138. That is, the content of property B is updated.

  In step S13, the reproducing unit 133 increments the history number by one. Thereafter, the process returns to step S4, and the subsequent processes are repeated.

  FIG. 15 is a flowchart for explaining a processing example when the property tmp = niche orientation & narrow & deep as a specific example of the recommendation processing based on the preference B and property B in step S5.

  In step S21, the movie recommendation engine 112 compares the property B stored in the property profile storage unit 138 of the client device 101 with the property tmp, and determines whether or not property B = property tmp.

  If property B = niche orientation & narrow & deep, it is determined in step S21 that property B = property tmp, and the process proceeds to step S22. In step S22, the movie recommendation engine 112 makes a niche recommendation with the preference B. Thereby, the process of step S5 in FIG. 14 ends, and the process proceeds to step S9.

  Note that “niche recommendation by preference B” in step S22 means that, for example, a recommendation candidate having a low major orientation is selected from the recommendation candidates determined by preference B as a recommended movie.

  If the property B is other than (niche oriented & narrow & deep), it is determined in step S21 that the property B = the property tmp is not satisfied, and the process proceeds to step S23. In step S23, the movie recommendation engine 112 determines whether or not the property B = major orientation & wide & shallow.

  If the property B = major orientation & wide & shallow, it is determined YES in step S23, and the process proceeds to step S24. In step S24, the movie recommendation engine 112 makes a recommendation that the hit B recommendation is 20% and the preference B recommendation is 20%. Thereby, the process of step S5 in FIG. 14 ends, and the process proceeds to step S9.

  The “recommendation with 80% hit work and 20% preference B recommendation” in step S24 means that, for example, a movie with a high degree of major orientation means “hit work”, and the recommendation candidate determined by preference B is The meaning of “preference B recommendation” means that the recommended movie is determined so that the hit movies out of the recommended movies occupy 80% and the preference B recommendation occupies 20%.

  If the property B is other than (niche oriented & narrow & deep) and (major oriented & wide & shallow), it is determined as NO in step S23, and the process proceeds to step S25. In step S25, the movie recommendation engine 112 makes a recommendation according to the preference B and the property B. Thereby, the process of step S5 in FIG. 14 ends, and the process proceeds to step S9.

  The movie recommendation process in the examples of FIGS. 14 and 15 described above is a process example assuming that the content type is a movie, but in reality, the content type is not particularly limited. That is, by setting the property for the first type of content as the property tmp, the property for the second type of content as the property B, and the preference information for the second type of content as the preference B, As a process for recommending the second type of content, the movie recommendation process of the example of FIG. 14 or 15 can be applied as it is.

  Further, the system capable of executing the movie recommendation process of FIGS. 14 and 15 is not particularly limited to the information processing system of FIG. Here, the system represents the entire apparatus composed of a plurality of apparatuses and processing units.

  That is, the movie recommendation server 102 and the client device 101 can be regarded as one device.

  In other words, the movie item database 111 and the movie recommendation engine 112, which are components of the movie recommendation server 102, need not be mounted on the movie recommendation server 102, and may be mounted on the client device 101 side. May be mounted on another device (not shown). In any case, it is obvious that the movie recommendation process of FIG. 14 or 15 can be executed. The above relationship between the movie recommendation server 102 and the client device 101 applies to the relationship between the music recommendation server 2 and the client device 1 in the same manner, and a recommendation server for a predetermined type of content not shown in FIG. The same applies to the relationship with the client device for playback.

  Although not shown, the recommendation server, such as the music recommendation server 2 or the movie recommendation server 102, has another second function having a property generation function for a first client device that does not have a property generation function. Recommendations can be made by using the properties obtained by the client device. That is, the recommendation server can also make a recommendation using the preference obtained by the first client device and the property obtained by the second client device.

  Furthermore, the usage of the property information is not limited to the above-described usage, and can be various usages in which the user can use the propensity of behavior with respect to the content.

  Specifically, for example, the property information can be used as a free keyword as follows.

  That is, the first client device generates a vector (hereinafter, referred to as a free keyword property vector FD) having free keywords relating to the type of content and the preference as well as various property information. Can be generated as a free keyword property vector FD = (o, w, d, kind, kw1, kw2,...) And transmitted to the second client device.

  In this case, the second client device can search each element of the free keyword property vector FD for each attribute and register it as the value of the attribute if a match is found. It can be used to help create initial user preference information. For example, the keyword part of the free keyword property vector FD is (kw1, kw2, kw3, kw4) = (music, samba, frog △△△△△ × zo, romantic), and the second client device If the corresponding recommendation server is a music recommendation server, the genre- {samba}, name- {kae ○ × △ ν △ × zo} are easily linked, and the mood- {romantic} also has a mood attribute. Depending on the situation.

  As described above, in the present invention, the detailed user preference information itself is not shared among a plurality of devices, but the user property (1) obtained by the statistical processing from the history information that is the source of the user preference information is generated. The above property information group) can be shared among a plurality of devices. This makes it possible to independently make content recommendations not only on one device but also on other devices by utilizing the properties of the user obtained on one device, and also independently recommend other types of content. It becomes like this.

  In other words, for example, the propensity of the user's behavior with respect to content obtained by one device can be utilized by other devices. Further, it is possible to make a recommendation (for example, from music to a movie) across the content type area. Also, any of these can be implemented without strict standards and thus allowing for unique interpretation and use of the device. That is, for example, even if the property information itself is the same information, the first device can perform the first interpretation and execute the process using the first interpretation, while obtaining the property information from the first device. In addition, the second device can perform the second interpretation, and can independently execute the process using the second interpretation.

  By the way, the series of processes described above can be executed by hardware, but can also be executed by software.

  In this case, for example, a personal computer shown in FIG. 16 may be employed as at least a part of the above-described client device or server.

  In FIG. 16, a CPU (Central Processing Unit) 201 executes various processes according to a program recorded in a ROM (Read Only Memory) 202 or a program loaded from a storage unit 208 to a RAM (Random Access Memory) 203. To do. The RAM 203 also appropriately stores data necessary for the CPU 201 to execute various processes.

  The CPU 201, the ROM 202, and the RAM 203 are connected to each other via the bus 204. An input / output interface 205 is also connected to the bus 204.

  The input / output interface 205 includes an input unit 206 such as a keyboard and a mouse, an output unit 207 including a display, a storage unit 208 including a hard disk, and a communication unit 209 including a modem and a terminal adapter. It is connected. The communication unit 209 controls communication performed with other devices (not shown) via a network including the Internet.

  A drive 210 is also connected to the input / output interface 205 as necessary, and a removable medium 211 such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory is appropriately installed, and a computer program read from them is loaded. Installed in the storage unit 208 as necessary.

  When a series of processing is executed by software, a program constituting the software executes various functions by installing a computer incorporated in dedicated hardware or various programs. For example, a general-purpose personal computer is installed from a network or a recording medium.

  As shown in FIG. 16, the recording medium containing such a program is distributed to provide a program to the user separately from the apparatus main body, and a magnetic disk (including a floppy disk) on which the program is recorded. , Removable media (package media) consisting of optical disks (including compact disk-read only memory (CD-ROM), DVD (digital versatile disk)), magneto-optical disks (including MD (mini-disk)), or semiconductor memory ) 211, but also includes a ROM 202 in which a program is recorded and a hard disk included in the storage unit 208 provided to the user in a state of being incorporated in the apparatus main body in advance.

  In the present specification, the step of describing the program recorded on the recording medium is not limited to the processing performed in chronological order according to the order, but is not necessarily performed in chronological order, either in parallel or individually. The process to be executed is also included.

  Further, in the present specification, the system represents the entire apparatus including a plurality of apparatuses and processing units.

It is a figure which shows the structural example of the information processing system to which this invention is applied. It is a figure which shows the concept of the cluster and cluster layer which classify | categorize metadata, ie, the concept of a multiview cluster. It is a figure which shows an example of cluster information. It is a figure which shows an example of cluster-music ID information. It is a figure for demonstrating the method of selecting two types of methods from the clustering 1st thru | or 4th method. It is a figure for demonstrating the method of selecting two types of methods from the clustering 1st thru | or 4th method. It is a figure for demonstrating the method of selecting two types of methods from the clustering 1st thru | or 4th method. It is a figure for demonstrating the method of selecting two types of methods from the clustering 1st thru | or 4th method. It is a figure for demonstrating the method of selecting two types of methods from the clustering 1st thru | or 4th method. It is a figure utilized in order to demonstrate the production | generation method of property information, Comprising: It is a figure which shows the specific example of the result of a multiview cluster. It is a figure utilized in order to demonstrate the production | generation method of property information, Comprising: It is a figure which shows the specific example of the result of a multiview cluster. It is a figure which shows the specific example of the content selection using property information. It is a figure which shows another structural example of the information processing system to which this invention is applied. It is a flowchart explaining the example of a movie recommendation process which the information processing system of FIG. 13 performs. It is a flowchart explaining the specific example of the recommendation process based on the preference B and property B corresponding to step S5 of FIG. It is a block diagram which shows the structural example of the personal computer as an information processing apparatus with which this invention is applied.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1 Client apparatus, 2 Music recommendation server, 3 Music distribution management server, 11 Music item database, 12 Music recommendation engine, 21 Similarity calculation part, 22 Property adaptation filter part, 31 Strategy determination part, 32 Recommendation presentation part, 33 Playback part , 34 history storage unit, 35 preference extraction unit, 36 preference profile storage unit, 37 property extraction unit, 38 property profile storage unit, 39 communication unit, 101 client device, 102 movie recommendation server, 111 song item database, 112 song recommendation engine , 121 similarity calculation unit, 122 property adaptation filter unit, 131 strategy determination unit, 132 recommendation presentation unit, 133 playback unit, 134 history storage unit, 135 preference extraction unit, 136 preference profile storage unit, 137 property extraction Department, 138 properties profile storage section, 139 communication unit, 201 CPU, 202 ROM, 203 RAM, 208 storage unit, 211 a removable media

Claims (16)

In an information processing apparatus that executes processing for a predetermined type of content,
Property generation means for generating one or more property information representing the property of the user for the predetermined type of content based on history information indicating a history of user behavior regarding the processing for the predetermined type of content;
An information processing apparatus comprising: communication means for transmitting one or more pieces of the property information generated by the property generation means. The information processing apparatus according to claim 1, wherein the property generation unit generates at least one of orientation, breadth, depth, and maturity as the property information. The information processing apparatus according to claim 2, wherein the property generation unit generates the orientation based on a property of one or more contents grasped from the history information. The property generation means, based on the number of values and entropy for each of the plurality of content attributes when each of the one or more contents grasped from the history information is classified into a plurality of content attributes, The information processing apparatus according to claim 2, wherein at least one of the depth and the depth is generated. The information processing apparatus according to claim 4, wherein the property generation unit uses a multi-viewpoint cluster as the plurality of content attributes. The property generation means determines the maturity based on the number of values and entropy for each of the plurality of content attributes when each of the one or more contents grasped from the history information is classified into a plurality of content attributes. The information processing apparatus according to claim 2. The information processing apparatus according to claim 6, wherein the property generation unit uses a multi-viewpoint cluster as the plurality of content attributes. The information processing apparatus according to claim 6, wherein the property generation unit further generates the maturity based on a property of one or more contents grasped from the history information. The information processing apparatus according to claim 6, wherein the property generation unit generates the maturity based on history information for a predetermined period of the divided periods, the history information being divided by periods. The information processing apparatus according to claim 1, wherein the property generation unit normalizes all of the one or more property information. The information processing apparatus according to claim 1, wherein the communication unit transmits information indicating the predetermined type of content in association with one or more pieces of the property information. A holding unit that holds one or more pieces of the property information generated by the property generation unit;
The communication means further receives one or more of the property information transmitted from another information processing apparatus,
The information processing apparatus according to claim 1, wherein the holding unit further holds one or more of the property information received by the communication unit as property information about the information processing device itself. The information processing apparatus according to claim 12, wherein the one or more pieces of property information received by the communication unit are property information regarding content of a type different from the predetermined type to be processed by the information processing device. The information processing apparatus according to claim 1, wherein the communication unit transmits a free keyword in association with one or more of the property information. In an information processing method of an information processing apparatus that executes processing for a predetermined type of content,
Generating one or more property information indicating the property of the user for the predetermined type of content based on history information indicating a history of user behavior regarding the processing for the predetermined type of content;
An information processing method comprising a step of transmitting the generated one or more property information from the information processing apparatus to another information processing apparatus. A program to be executed by a computer that controls a device that executes processing for a predetermined type of content,
Generating one or more property information indicating the property of the user for the predetermined type of content based on history information indicating a history of user behavior regarding the processing for the predetermined type of content;
A program including a step of performing control to transmit one or more generated property information from the device to another device.

Images