WO2007037139A1 - 情報処理装置、方法、およびプログラム - Google Patents
情報処理装置、方法、およびプログラム Download PDFInfo
- Publication number
- WO2007037139A1 WO2007037139A1 PCT/JP2006/318373 JP2006318373W WO2007037139A1 WO 2007037139 A1 WO2007037139 A1 WO 2007037139A1 JP 2006318373 W JP2006318373 W JP 2006318373W WO 2007037139 A1 WO2007037139 A1 WO 2007037139A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- user
- content
- cluster
- music
- information
- Prior art date
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y10—TECHNICAL SUBJECTS COVERED BY FORMER USPC
- Y10S—TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y10S707/00—Data processing: database and file management or data structures
- Y10S707/912—Applications of a database
- Y10S707/913—Multimedia
- Y10S707/916—Audio
Definitions
- the present invention relates to an information processing device, an information processing method, and a program, and in particular, classifies content into clusters, manages content features using the classified clusters, and searches and recommends content.
- the present invention relates to an information processing apparatus, an information processing method, and a program that are used in the future.
- CF collaborative filtering
- CBF content best filtering
- the CF method manages the purchase history of each user, detects other user X with a similar purchase history for user A who wants to recommend content, and detects the other user X recommends content purchased by user A but not purchased by user A. For example, it has been adopted by mail order sites on the Internet.
- Patent Document 1 Japanese Patent Application Laid-Open No. 2004-194107
- the present invention has been made in view of such a situation, and suppresses concentration of recommendations to some content in the CF method, and also for users with little history information.
- the content can be recommended.
- An information processing apparatus is an information processing apparatus that selects content that satisfies a predetermined condition from the power of a content group and presents the content to a user, and configures the content group
- Content classification means for classifying each content to be classified into one of the plurality of first clusters in each hierarchy according to the content metadata, and each content and each content is classified
- a holding unit that holds a database indicating a correspondence relationship with the first cluster in the hierarchy, a management unit that manages history information on the user's content, and a first first to be noted based on the history information
- Selecting means for identifying a cluster and selecting content classified into the identified first cluster; and presenting means for presenting the selected content.
- the selection means does not exist on the history information of the first user, the detection means for detecting the second user whose history information is similar to the first user, A specifying means for specifying the first cluster in which the content existing in the history information of the second user is classified, and an extracting means for extracting the content classified in the specified first cluster
- the presenting means can present the extracted content to the first user.
- An information processing apparatus includes user history information and the database. And generating means for generating preference information indicating the user's preferences in the first cluster unit, and grouping means for grouping users based on the preference information!
- the selection means includes a detection means for detecting a second user belonging to the same group as the first user, and is not present in the history information of the first user, and is not the second user's history information.
- the presenting means includes: identification means for identifying a first cluster in which content existing in history information is classified; and extraction means for extracting content classified in the identified first cluster. Can prompt the first user to present the extracted content.
- the information processing apparatus generates preference information indicating the user's preference in units of the first cluster based on the user's history information and the database.
- Generating means detecting means for detecting a second user having a preference similar to that of the first user and the preference information, and a preference of the first user;
- a specifying means for specifying the first cluster of interest based on the information and the preference information of the second user, and an extracting means for extracting the content classified into the specified first cluster.
- the presenting means can present the extracted content to the first user.
- the detecting means includes a normalizing means for normalizing user preference information and a weight calculation for calculating a weight for each layer for each user from the normalized user preference information. The first user among the users from the means, the weight for each layer and the preference information
- a similarity calculation means for calculating a similarity indicating the degree of preference similarity with other users among the users, and the preference is similar to that of the first user based on the calculated similarity.
- the second user can be detected.
- preference information indicating the user preference in units of the first cluster is generated based on the user history information and the database.
- generating means and grouping means for grouping users based on the preference information, and the selecting means detects a second user belonging to the same group as the first user.
- Detection means; identification means for identifying a first cluster of interest based on preference information of the first user and preference information of the second user; and the identified first Extraction means for extracting the content classified into the cluster is provided, and the presentation means can present the extracted content to the first user.
- a setting unit that sets a keyword for each of the first clusters into which the metadata is classified by the content classification unit, and a setting by the setting unit
- creating means for creating a reason sentence that represents the reason for presenting the content by using the keyword, and the presenting means can present the reason sentence.
- a metadata classification unit that classifies content metadata into one of a plurality of second clusters and assigns the hierarchy to the second cluster is further provided, and each content is allocated to the content classification unit It is possible to classify the plurality of first clusters into one or more!
- generation means for generating preference information indicating the user's preference in the first cluster unit, and the selection means includes all of the hierarchy levels Among all the first clusters, it is possible to select the first cluster indicated by the preference information and classified as the most first cluster.
- An information processing method is an information processing method for an information processing apparatus that selects content that satisfies a predetermined condition from the power of a content group and presents the content to a user.
- Each content constituting the content group is classified into one of a plurality of clusters in each hierarchy according to the content metadata, and each content and each content is classified in the hierarchy.
- Maintains a database indicating the correspondence with the cluster manages history information for the user's content, identifies a cluster of interest based on the history information, and is classified into the identified cluster Selecting a content and presenting the selected content.
- a program is a program for selecting content that satisfies a predetermined condition from the power of a content group and recommending it to a user. , Multiple in each hierarchy according to content metadata A database indicating the correspondence between each content and each cluster in which the content is classified, and managing history information on the content of the user, Based on the history information, the target cluster is specified, the content classified into the specified cluster is selected, and the computer is caused to execute a process including the step of presenting the selected content.
- each content constituting the content group is classified into one of a plurality of clusters in each hierarchy according to content metadata.
- a database indicating the correspondence between each content and the cluster in the hierarchy in which each content is classified is held.
- history information for the user's content is managed. Then, based on the history information, the cluster of interest is identified, the content classified into the identified cluster is selected, and the selected content is presented.
- FIG. 1 is a block diagram showing a configuration example of a recommendation system to which the present invention is applied.
- FIG. 2 is a diagram showing the concept of clusters and cluster layers for classifying music metadata.
- FIG. 3 is a diagram showing an example of a music cluster correspondence table.
- FIG. 4 is a diagram showing an example of a cluster music correspondence table.
- FIG. 5 is a diagram showing an example of a user preference vector.
- FIG. 6 is a flowchart for explaining preprocessing when offline.
- FIG. 7 is a flowchart for explaining a first recommendation process.
- FIG. 8 is a flowchart for explaining second and third recommendation processing.
- FIG. 9 is a flowchart illustrating a fourth recommendation process.
- FIG. 10 is a flowchart for explaining fifth and sixth recommendation processes.
- FIG. 11 is a flowchart for explaining a seventh recommendation process.
- FIG. 12 is a block diagram illustrating a configuration example of a general-purpose personal computer.
- FIG. 13 is a block diagram showing another example of the configuration of the recommendation system according to one embodiment of the present invention.
- FIG. 14 is a flowchart illustrating another example of pre-processing when offline.
- FIG. 15 is a diagram showing an example of metadata of each piece of music subjected to soft clustering.
- FIG. 16 is a diagram showing an example of metadata of each music piece.
- FIG. 17 is a diagram showing an example of metadata for each clustered music piece.
- FIG. 18 is a block diagram showing an example of the configuration of a similar user detection unit.
- FIG. 19 is a flowchart illustrating processing for detecting a user X with similar preferences.
- FIG. 20 is a diagram showing an example of a preference vector.
- FIG. 21 is a diagram showing an example of a regularized preference vector.
- FIG. 22 is a diagram showing an example of weights.
- FIG. 23 is a diagram showing an example of similarity calculated without weighting.
- FIG. 24 is a diagram showing an example of similarity calculated by weighting.
- FIG. 1 shows a configuration example of a recommendation system according to an embodiment of the present invention.
- This recommendation system 1 manages user history information (information on music data purchase, trial listening, search, possession, etc.) at a music data sales site established on the Internet, and recommends it using the CF method. The music to be selected is selected and presented to the user.
- the recommendation system 1 can also be applied to sales sites that sell contents other than music, such as television programs, movies, and books.
- the recommendation system 1 includes a music database (DB) 11 and a music database 11 in which metadata of a large number of music data (hereinafter simply referred to as music) to be recommended and sold to users is recorded.
- the clustering unit 12 Based on the metadata of each recorded song, the clustering unit 12 that clusters each song to generate cluster information for each song, and a keyword that indicates the characteristics of each cluster in each cluster layer and cluster layer. It consists of a keyword setting unit 13 to be set and a clustered database (DB) 14 that holds the clustering result of each music piece.
- the clustered DB 14 retains, as clustering results, a cluster music correspondence table 15 indicating the music belonging to each cluster and a music cluster correspondence table 16 indicating the cluster to which each music belongs.
- the recommendation system 1 is a user history information database (DB) 17 that manages each user's history information, and selects a plurality of songs that are recommended candidates based on the user information.
- Selection unit 18 music selection unit 25 for selecting one song from a plurality of selected recommendation candidates, new for determining whether or not the selected music is novel for the recommended user From the sex determination unit 26, the selection reason generation unit 27 that generates a recommendation reason sentence when presenting the selected music to the user, and the presentation unit 28 that presents the selected music and the recommendation reason sentence to the user Composed.
- DB user history information database
- the recommendation candidate selection unit 18 includes a preference vector generation unit 19, a user grouping unit 20, a similar user detection unit 21, a difference detection unit 22, a recommended cluster determination unit 23, and an extraction unit 24.
- the music DB 11 is an Internet that supplies metadata of music recorded on a music CD. Like the above data servers such as CDDB (CD Data Base) and Music Navi, it holds metadata of songs that are recommended and sold.
- CDDB CD Data Base
- Music Navi it holds metadata of songs that are recommended and sold.
- the clustering unit 12 applies to each piece of music metadata items (artist name, genre, album, artist review, song review, title, tempo, beat, rhythm, etc.) Based on the combination (tempo, beat, rhythm, etc.), a cluster layer (1st to nth layers) as shown in Fig. 2 is created, and the music is one of a plurality of clusters provided in each cluster layer, or Classify into multiple groups (clustering).
- music metadata items artist name, genre, album, artist review, song review, title, tempo, beat, rhythm, etc.
- the power described in the example of music is also clustered in multiple layers for each artist and album using many metas.
- a multi-layer cluster for music a multi-layer cluster for artists, and a multi-layer cluster for album are used, respectively.
- any method may be used for clustering, but an optimum clustering method and distance measure are selected for each cluster layer.
- the metadata's actual information is a numeric attribute such as tempo
- a distance measure such as Euclidean distance by converting it into a numerical value using a quantification method such as principal component analysis if it is a nominal attribute such as title.
- Typical clustering methods include K-means method, hierarchical clustering method (group average method, farthest method, Ward method), and soft clustering method.
- clustering for example, constrained clustering
- a partial collection of correct answers (a set of real information 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.
- a clustering method that increases the independence of each formed cluster layer (that is, a clustering method with different characteristics).
- cluster information consisting of cluster IDs (eg, CL11 in Fig. 2) of clusters that classify actual information of each item of metadata is generated and output to the clustered DB14.
- each cluster is arbitrary and can include a plurality of actual information.
- a cluster that can classify only single real information may be provided.
- real information IDs (artist ID, album, title ID) that can only be classified into the cluster ID of the cluster may be used.
- the clustered DB 14 generates and holds a cluster-one song correspondence table 15 and a song-one cluster correspondence table 16 based on the cluster information of each song generated by the clustering unit 12.
- the clustered DB 14 also holds the keywords set for each cluster layer and each cluster set by the keyword setting unit 13.
- FIG. 3 shows an example of the music-cluster correspondence table 16.
- FIG. 4 shows an example of the cluster music correspondence table 15 corresponding to the music cluster correspondence table 16 shown in FIG.
- User history information DB17 holds history information that indicates the songs that each user has purchased, listened to, or searched for on the sales site, or the songs that have been purchased and declared as possessed by either user. Has been.
- the user history information DB17 contains The preference vector of each user generated by the preference vector generation unit 19 is held.
- the user history information DB 17 holds the result of grouping users by the user group section 20, that is, information indicating which user group each user belongs to.
- the preference vector generation unit 19 is based on the history information of each user stored in the user history information DB 17, and for each user, a multidimensional preference in which all clusters are one-dimensional.
- a vector is generated and output to the user history information DB17.
- the music cluster correspondence table 16 of the clustered DB 14 is referred to, and a predetermined value is added to the dimension of the preference vector corresponding to the cluster to which the music belongs.
- the generated user preferences are managed in the user history information DB17. If the user's history information is updated, such as by purchasing music, the preference vector will be updated.
- the cluster ID to which the first song belongs is CL11, CL22, CL32, CL43
- the dimension values corresponding to these 1 is added to each.
- the cluster ID to which the second music piece belongs is CL12, CL22, CL33, CL42, 1 is added to each of the dimension values corresponding thereto.
- the cluster ID to which the third song belongs is CL13, CL24, CL33, CL41, 1 is added to the dimension values corresponding to these.
- user X preference A solid (1, 1, 1, 0, 2, 0, 1, 0, 1, 2, 1, 1, 1) force is generated.
- each dimension may be changed according to the type of history information (that is, purchase, audition, search, or possession). For example, 1 may be added for purchase or possession, 0.5 may be added for trial listening, and 0.3 may be added for search.
- the user group section 20 groups all users based on the similarity of each user's preference stored in the user history information DB17.
- the value of each dimension of each user's preference vector is binarized. That is, when the value of each dimension is 1 or more, it is replaced with 1, and when it is 0, it remains 0. In this way, binarizing the values of each dimension of the preference vector makes it possible to reduce the amount of computation required to determine similarity compared to the case where binarization is not performed, and to easily perform grouping. be able to.
- This group result is managed by the user history information DB17.
- the similar user detection unit 21 compares the history information of each user held in the user history information DB 17 to make the history information similar to the user whose music is recommended. Detect users. In addition, the similar user detection unit 21 compares the preference vectors of each user stored in the user history information DB 17 to determine whether the preference vector is similar to the user whose music is recommended. Detect users. Based on the history information of each user stored in the user history information DB 17, the difference detection unit 22 determines whether the user whose music is recommended and the other users detected by the similar user detection unit 21. Detect differences in history information. The recommended cluster determination unit 23 determines a recommended cluster based on the difference in preference vectors between the user whose music is recommended and the other users detected by the similar user detection unit 21. Based on the processing results of the similar user detection unit 21 to the recommended cluster determination unit 23 and the clustered DB 14, the extraction unit 24 extracts songs that are recommended candidates.
- the music selection unit 25 selects one music from a plurality of extracted music according to a predetermined condition. For example, a song that belongs to a larger number of recommended clusters, a song that belongs to a recommended cluster in a cluster layer with a high priority set in advance, or one music at random is selected, and the selection result is a novelty determination unit. 26 and selection reason generation unit 27.
- the novelty determination unit 26 selects the music based on the preference vector of the user whose music is recommended. If the degree of overlap of the cluster to which the song belongs is greater than or equal to a predetermined percentage (for example, 30%), it is determined that there is no novelty, and if it is less than the prescribed percentage, it is novelty The judgment result is output to the selection reason generation unit 27.
- a predetermined percentage for example, 30%
- the selection reason generation unit 27 acquires a cluster layer to which the selected music belongs and a keyword corresponding to the cluster from the clustered DB 14, and uses the acquired keyword or the like to display a selection reason sentence indicating the reason for the selection. Generate. Also, based on the judgment result from the novelty judgment unit 26, for example, “unexpected” is for a novelty, “, one” is for a novelty, “familiar” The reason for selection is generated including the words such as. Then, the generated selection reason sentence is output to the presentation unit 28 together with the music ID of the selected music.
- the review text of the selected music piece may be directly quoted as the selection reason sentence, or the selection reason sentence may be generated using a word extracted from the review text of the selected music piece.
- Tf / idlf leaving can be applied to extract the word used in the selection reason sentence from the review text.
- the presentation unit 29 acquires information on the selected music piece from the music DB and presents it to the user side together with the generated selection reason sentence.
- step S1 the clustering unit 12 classifies all the music pieces in the music DB 11 into each item cluster layer (first to nth layers) of the music metadata, and stores the actual information of each item. Classification (clustering) is made into one of multiple clusters provided in the classified cluster layer. Then, the clustering unit 12 generates cluster information including the cluster ID power of the cluster that classifies the actual information of each item of the metadata as information indicating the characteristics of the music instead of the metadata, and outputs it to the clustered DB 14 To do. It should be noted that clustering may be omitted for music that has already been clustered, and clustering only for music that has not been clustered.
- the clustered DB 14 generates a cluster music correspondence table 15 and a music cluster correspondence table 16 based on the cluster information of each music generated by the clustering unit 12.
- the preference vector generation unit 19 of the recommendation candidate selection unit 18 generates a preference vector for each user based on the history information of each user held in the user history information DB 17. And output to the user history information DB17.
- the user grouping section 20 groups all users based on the similarity of each user's preference vector held in the user history information DB17. However, in order to facilitate the process of determining the similarity of multidimensional preference vectors, the value of each dimension of each user's preference vector is binarized. The grouping result is output to the user history information DB17. This completes the offline preprocessing.
- the clustering of all the songs held in the music DB 11, the generation of each user's preference vector, and the user's grouping are performed as pre-processing, so that the first described later is performed.
- Thru 7 recommendation processes can be executed promptly. Note that some of the first to seventh recommendation processes do not use the user group information. Therefore, when only the recommendation process is executed without using the user group information, the process of step S3 is omitted. You can omit it.
- the first recommendation process will be described with reference to the flowchart of FIG.
- the user whose music is recommended is described as user A. This process is started, for example, when user A accesses the sales site.
- step S11 the similar user detection unit 21 compares the history information of the user A stored in the user history information DB 17 with the history information of the other users, thereby obtaining the user A. And detect other user X whose history information is most similar.
- step S12 the difference detection unit 22 has the user X (which has been purchased or held in the past) based on the user A and user history information stored in the user history information DB17. Yes) Detect music that user A does not have. If there are multiple songs that satisfy this condition, one of them is selected at random, for example. Let the detected music be song a.
- the recommended cluster determining unit 23 refers to the music-cluster correspondence table 16 of the clustered DB 14, and identifies a cluster of each cluster layer to which the music a belongs.
- the extractor 24 supports clustered DB14 cluster-single music. Referring to Table 15, the music that is classified in common for all the clusters specified in the process of step S13 is extracted. The music extracted here is set as a recommendation candidate. There can be multiple recommendation candidates. If there are no songs that are classified in common in all the clusters identified in step S13, the clusters identified in step S13 are classified into as many clusters as possible. The music that is present is extracted and set as a candidate for recommendation.
- step S15 the music selection unit 25 selects one song having the cluster information most similar to the music a detected in step S12 among the recommended candidate songs, and the selection result is the novelty determination unit 26. , And the selection reason generation unit 27.
- step S16 the novelty determination unit 26 determines the presence / absence of novelty based on the preference vector of user A and the cluster to which the selected music belongs, and outputs the determination result to the selection reason generation unit 27.
- the selection reason generation unit 27 obtains the cluster layer to which the selected music belongs and the keyword corresponding to the cluster from the clustered DB 14 and generates a selection reason sentence indicating the reason for the selection using the obtained keyword. To do.
- the selection reason sentence is also generated based on the determination result from the novelty determination unit 26.
- the generated selection reason sentence is output to the presentation unit 28 together with the music ID of the selected music.
- the presentation unit 29 acquires information about the selected music piece from the music DB and presents it to the user side along with the generated selection reason sentence. This completes the first recommendation process.
- This process is started when, for example, user A accesses the sales site.
- step S21 the similar user detection unit 21 compares user A's preference vector held in the user history information DB 17 with another user's preference vector and The other user X with the most similar preference vector is detected.
- the similarity between the preference vector of user A and the preference vector of another user is determined by, for example, calculating the cosine correlation value of both.
- step S22 the difference detection unit 22 has a value of 0 in the user A's preference vector and a dimension of the preference vector that has a value other than 0 in the user X's preference vector! And a cluster corresponding to the detected dimension is determined as a recommended cluster.
- the preference vector of user A is (1, 1, 0, 0, 1, 0, 1, 0, 1, 1, 1, 0, 1) shown in A of FIG.
- User X's preference power S In the case of (1, 1, 1, 0, 2, 0, 1, 0, 1, 2, 1, 1, 1) shown in Fig. 5B, Fig. 5 Clusters CL13 and CL42 are determined as recommended clusters as indicated by diagonal lines in B of FIG.
- step S23 the extraction unit 24 refers to the cluster music correspondence table 15 of the user history information DB 17 and the clustered DB 14, and among all the music categorized as the recommended cluster, the user X The candidate that user A does not have is extracted as a candidate for recommendation.
- step S24 the music selection unit 25 selects one music from among the recommended candidate music by one of the following three methods or a combination of the three methods, and determines the selection result as a novelty: Output to the unit 26 and the selection reason generation unit 27.
- the first method is to select songs that belong to more recommended clusters.
- the second method is a method in which priorities are given to the cluster layer in advance, and music pieces classified into recommended clusters belonging to the cluster layer with higher priorities are selected.
- the third method is a random selection method.
- step S25 the novelty determination unit 26 determines the presence / absence of novelty based on the preference vector of user A and the cluster to which the selected music belongs, and generates a determination result as a selection reason. Output to part 27.
- the selection reason generation unit 27 also acquires the keywords corresponding to the cluster layer cluster to which the selected music belongs, and also generates a selection reason sentence indicating the reason for selection using the acquired keywords and the like.
- the selection reason sentence is also generated based on the determination result from the novelty determination unit 26. Then, the generated selection reason sentence is output to the presentation unit 28 together with the music ID of the selected music.
- step S26 the presentation unit 29 acquires information on the selected music from the music DB and presents it to the user side along with the generated selection reason sentence. This completes the second recommendation process.
- the third recommendation process is a candidate for recommendation by extracting all the songs classified by the recommended cluster in step S23 in the second recommendation process described above, which user A does not have. And so on. Ie use The person X has and can also be a recommendation candidate. Since the other processes are the same as the second recommendation process, the description thereof is omitted.
- step S41 the similar user detecting unit 21 stores another user X belonging to the same group as the user A based on the group information of the user A held in the user history information DB 17. Determine at random.
- step S42 the difference detection unit 22 is a song held by the user X based on the history information of the user A and the user X held in the user history information DB 17, Detect music that A does not have. If there are multiple songs that satisfy this condition, one of them is selected at random, for example. The detected music is called music a.
- the recommended cluster determination unit 23 refers to the music-to-cluster correspondence table 16 in the clustered DB 14, and identifies a cluster in each cluster layer to which the music a belongs.
- step S44 the extraction unit 24 refers to the cluster-to-music-correspondence table 15 of the clustered DB 14, and extracts music that is classified in common for all the clusters specified in the process of step S43.
- the music extracted here is set as a recommendation candidate. There can be multiple recommendation candidates. If there are no songs that are classified in common in all clusters identified in step S43, the clusters identified in step S43 are classified in as many clusters as possible. Extract music and make it a recommended candidate.
- step S45 the music selection unit 25 selects one music whose cluster information is most similar to the music a detected in step S42 among the recommended candidates, and the selection result is the novelty determination unit 26. , And the selection reason generation unit 27.
- step S46 the novelty determination unit 26 determines the presence / absence of novelty based on the preference vector of the user A and the cluster to which the selected music belongs, and outputs the determination result to the selection reason generation unit 27.
- the selection reason generation unit 27 obtains the cluster layer to which the selected music belongs and the keyword corresponding to the cluster from the clustered DB 14 and generates a selection reason sentence indicating the reason for the selection using the obtained keyword. To do.
- the selection reason sentence is also generated based on the determination result from the novelty determination unit 26.
- step S47 the presentation unit 29 acquires information on the selected music piece from the music DB and presents it to the user side along with the generated selection reason sentence. This completes the fourth recommendation process.
- the group information of users who have been grouped by the offline pre-processing is used, and therefore user X similar to the history of user A is speeded up. It can be determined.
- step S51 the similar user detecting unit 21 stores another user X belonging to the same group as the user A based on the group information of the user A held in the user history information DB17. Determine at random.
- step S52 the difference detection unit 22 determines that the preference vector of user A has a value of 0 and the preference vector of user X has a value other than 0 /! And a cluster corresponding to the detected dimension is determined as a recommended cluster.
- step S53 the extraction unit 24 uses the user history information DB 17 and the clustered DB.
- step S54 the music selection unit 25 selects one song from among the recommended candidate songs by one of the following three methods or a combination of the three methods, and determines the selection result as a novelty: Output to the unit 26 and the selection reason generation unit 27.
- the first method is to select songs that belong to more recommended clusters.
- the second method is a method in which priorities are assigned in advance to the cluster layer, and music pieces classified into recommended clusters belonging to the cluster layer with higher priorities are selected.
- the third method is a random selection method.
- step S55 the novelty determination unit 26 determines the presence / absence of novelty based on the preference vector of user A and the cluster to which the selected music belongs, and the determination result is sent to the selection reason generation unit 27. Output.
- the selection reason generation unit 27 also acquires the keywords corresponding to the cluster layer cluster to which the selected music belongs, and also generates a selection reason sentence indicating the reason for selection using the acquired keywords.
- the selection reason sentence is also generated based on the determination result from the novelty determination unit 26. Then, the generated selection reason sentence is output to the presentation unit 28 together with the music ID of the selected music.
- the presentation unit 29 acquires information on the selected music from the music DB and presents it to the user side along with the generated selection reason sentence. This completes the fifth recommendation process.
- the sixth recommendation process is a candidate for recommendation by extracting all the music pieces classified into the recommended cluster by user A in step S53 in the fifth recommendation process described above and not having user A. And so on. That is, what the user X has can also be a recommendation candidate. Since the other processes are the same as the fifth recommendation process, the description thereof is omitted.
- the group information of the users who are grouped by the offline pre-processing is used, so that the user X similar to the history of the user A is promptly used. Can be determined.
- the seventh recommendation process will be described with reference to the flowchart of FIG. First, the seventh recommendation process will be described. This process is suitable when the history information of user A is extremely small or when there are few other users, for example, when user A accesses the sales site.
- step S61 the difference detection unit 22 detects a dimension whose value is equal to or greater than a predetermined value from the dimensions of the preference vector of the user A, and determines a cluster corresponding to the dimension as a recommended cluster. To do.
- step S62 the extraction unit 24 refers to the cluster music correspondence table 15 of the user history information DB 17 and the clustered DB 14, and among all the music classified as the recommended cluster, the user A Those not possessed are extracted and set as recommended candidates.
- step S63 the music selection unit 25 selects the most recommended music from among the recommended music candidates. One piece of music belonging to the cluster is selected, and the selection result is output to the novelty judgment unit 26 and the selection reason generation unit 27. If there are multiple songs belonging to the most recommended clusters, for example, one song is selected at random.
- step S64 the novelty determination unit 26 determines the presence / absence of novelty based on the preference vector of user A and the cluster to which the selected music belongs, and generates the determination result as a selection reason. Output to part 27.
- the selection reason generation unit 27 also acquires the keywords corresponding to the cluster layer cluster to which the selected music belongs, and also generates a selection reason sentence indicating the reason for selection using the acquired keywords and the like.
- the selection reason sentence is also generated based on the determination result from the novelty determination unit 26. Then, the generated selection reason sentence is output to the presentation unit 28 together with the music ID of the selected music.
- step S65 the presentation unit 29 acquires information on the selected music from the music DB and presents it to the user side along with the generated selection reason sentence. This completes the seventh recommendation process.
- the user's history information is replaced with a preference vector with each cluster as a one dimension, and the CF method is applied. Can be prevented from concentrating on a part of all songs existing in the song DB11.
- music can be recommended to users with little history information, and so-called cold start problems can be avoided.
- the reason why the recommended music is selected can be presented to User A, for example, User A knows whether the recommended music is novel to himself or herself. be able to.
- the present invention can be applied not only to recommending music, but also to sales sites that sell contents other than music, such as television programs, movies, books, and the like.
- the series of processes described above can be executed by a force software that can be executed by hardware.
- various functions can be executed by installing a computer built in dedicated hardware or various programs that make up the software.
- it is installed from a recording medium in a general-purpose personal computer configured as shown in FIG.
- This personal computer 100 incorporates a CPU (Central Processing Unit) 101.
- An input / output interface 105 is connected to the CPU 101 via the bus 104.
- a ROM (Read Only Memory) 102 and a RAM (Random Access Memory) 103 are connected to the node 104.
- the input / output interface 105 includes an input unit 106 including an input device such as a keyboard and a mouse for a user to input an operation command, a CRT (Cathode Ray Tube) or an LCD (Liquid Crystal Display) for displaying a screen. ) Etc., an output unit 107 consisting of a display, a storage unit 108 consisting of a hard disk drive etc. that stores programs and various data, a modem, a LAN (Local Area Network) adapter, etc., via a network typified by the Internet A communication unit 109 that executes communication processing is connected.
- an input unit 106 including an input device such as a keyboard and a mouse for a user to input an operation command, a CRT (Cathode Ray Tube) or an LCD (Liquid Crystal Display) for displaying a screen. ) Etc., an output unit 107 consisting of a display, a storage unit 108 consisting of a hard disk drive etc. that stores programs and various data,
- magnetic disks including flexible disks
- optical disks including CD-ROM (Compact Disc-Read Only Memory), DVD (Digital Versatile Disc)), magneto-optical disks (including MD (Mini Disc)
- MD Magneto-optical disks
- a drive 110 for reading / writing data from / to a recording medium 111 such as a semiconductor memory is connected.
- a program for causing the personal computer 100 to execute the above-described series of processing is supplied to the personal computer 100 in a state stored in the recording medium 111, read by the drive 110, and built in the storage unit 108. Installed in the drive.
- the program installed in the storage unit 108 is loaded from the storage unit 108 to the RAM 103 and executed by a command of the CPU 101 corresponding to a command from the user input to the input unit 106.
- FIG. 13 is a block diagram showing another example of the configuration of the recommendation system 1 according to the embodiment of the present invention.
- the same parts as those shown in FIG. 1 are denoted by the same reference numerals, and the description thereof is omitted.
- the recommendation system 1 shown in FIG. 13 includes a music DB 11, a keyword setting unit 13, a clustered DB 14, a user history information DB 17, a recommendation candidate selection unit 18, a music selection unit 25, a novelty determination unit 26,
- the selection reason generation unit 27, the presentation unit 28, the metadata clustering unit 201, and the music clustering unit 202 are configured.
- the metadata clustering unit 201 stores each piece of music recorded in the music database 11. Cluster metadata for. In other words, the metadata clustering unit 201 classifies the metadata of music pieces that are contents into a plurality of clusters, and assigns a hierarchy to the clusters.
- the metadata clustering unit 201 supplies the result of the clustering of the metadata of each music piece to the music clustering unit 202.
- the song clustering unit 202 clusters each song and generates cluster information of each song, similar to the clustering unit 12. To do. That is, the music clustering unit 202 generates cluster information corresponding to the result of clustering each music and outputs the cluster information to the clustered DB 14.
- the recommendation candidate selection unit 18 of the recommendation system 1 shown in FIG. 13 includes a preference vector generation unit 19, a user grouping unit 20, a difference detection unit 22, a recommended cluster determination unit 23, an extraction unit 24, and A similar user detection unit 203 is included.
- the similar user detection unit 203 compares each user's preference vector stored in the user history information DB 17 to determine whether the preference vector is similar to the user whose music is recommended. Detect users. More specifically, the similar user detection unit 203 normalizes a preference vector, which is an example of user preference information, and uses each normalized user preference vector for each user level. The weight is calculated, the similarity indicating the degree of preference similarity between users is calculated from the weight and preference vector for each hierarchy, and the preference is similar to that of the first user based on the calculated similarity. Detect a second user.
- step S201 the metadata clustering unit 201 acquires the metadata of the music from the music DB 11, and compresses the dimension of the acquired metadata. For example, in step S201, the metadata clustering unit 201 determines whether the metadata metadata obtained from the music DB 11 is the next of the metadata, LbA (the latent semantic analysis), PLbA (the probabilistic latent se mantic analysis), or quantification. Compress using methods such as Class III. [0104] Note that in step S201, the metadata clustering unit 201 may vectorize the song metadata.
- step S202 the metadata clustering unit 201 clusters the metadata of each music piece.
- the metadata clustering unit 201 performs soft clustering on the metadata of each music piece.
- the metadata clustering unit 201 sets each item so that the sum of the belonging weights to each cluster becomes 1 in each hierarchy. Soft clustering of song metadata.
- the attribution weights to the first cluster, the second cluster, the third cluster, and the fourth cluster in the first hierarchy of the metadata of the music specified by the music ID of ABC123 are respectively , 0.0, 0.8, 0.0, and 0.2.
- the attribution weights for the 5th, 6th, 7th, and 8th clusters in the 2nd hierarchy of the metadata of the music identified by the music ID of ABC123 are 0.4, 0.6, 0.0, and 0.0.
- the attribution weight to the ninth cluster, the tenth cluster, and the eleventh cluster in the third layer of the metadata of the music specified by the music ID of ABC123 is 0.0, 0.0, and 1. 0.
- the attribution weight to each of the four clusters in the nth layer of the metadata of the music specified by the music ID of ABC123 is 1.0, 0.0, 0.0, and 0.0, respectively. is there.
- the attribution weights to the first cluster, the second cluster, the third cluster, and the fourth cluster in the first hierarchy of the metadata of the music specified by the music ID of CTH863 are respectively , 1.0, 0.0, 0.0, and 0.0.
- the attribution weights for the fifth cluster, the sixth cluster, the seventh cluster, and the eighth cluster in the second layer of the metadata of the music specified by the music ID of CTH863 are 0.0, 0.5, 0.5, and 0.0.
- the attribution weights for the ninth, tenth, and eleventh clusters in the third layer of the metadata of the song identified by the song ID that is CTH863 are 0.7, 0.3, and 0.0.
- the attribution weight to each of the four clusters in the nth layer of the metadata of the music specified by the music ID of CTH863 is 0.0, 0.8, 0.2, and 0.0, respectively. is there. [0109]
- the attribution weights to the first cluster, the second cluster, the third cluster, and the fourth cluster in the first layer of the metadata of the music specified by the music ID of XYZ567 are respectively , 0.0, 0.4, 0.6, and 0.0.
- the attribution weights for the fifth cluster, the sixth cluster, the seventh cluster, and the eighth cluster in the second layer of the metadata of the music specified by the music ID of XYZ567 are 0.0, 0.0, 0.0, and 1.0.
- the attribution weight to the ninth cluster, the tenth cluster, and the eleventh cluster in the third layer of the metadata of the music specified by the music ID of XYZ567 is 0.9, 0.0, and 0.
- the attribution weight to each of the four clusters in the nth layer of the metadata of the music specified by the music ID of XYZ567 is 0.3, 0.0, 0.0, and 0.7, respectively. is there.
- the soft clustering of the metadata of each music piece is not limited to one in which the sum of the weights belonging to each cluster of items, ie, music pieces, is 1 in each hierarchy. Each item may not belong to any cluster in each hierarchy.
- step S203 the metadata clustering unit 201 assigns a cluster layer.
- FIG. 16 is a diagram illustrating an example of metadata.
- the metadata shown in Fig. 16 is categorical data with a value of either 0 or 1 for simplicity.
- Metadata 1, metadata 2, and metadata 3 belong to the higher-level classification, and metadata 4, metadata 5, and metadata 3 belong to the higher-order classification.
- metadata 6 belongs.
- metadata relating to an artist belongs to the meta group 1
- the meta data 1 indicates the appearance of the artist
- the meta data 2 indicates a group.
- metadata related to a genre belongs to the meta group 2
- the meta data 4 indicates pops
- the meta data 5 indicates locks.
- the metadata 1 to sol metadata 6 of the song identified by the song ID ABC123 are 1, 1, 1, 1, 1, 1 and CTH863, respectively.
- the metadata 1 to metadata 6 of the music specified by the music ID are 0, 1, 0, 0, 1, 1, and XY
- the metadata 1 to metadata 6 of the music specified by the music ID of Z567 are 1, 1, 1, 1, 1, 1, respectively.
- the metadata 1 to metadata 6 of the music specified by the music ID of EKF534 are 1, 0, 1, 0, 0, 1 respectively, and the music specified by the music ID of OPQ385 is specified.
- Metadata 1 to metadata 6 are 1, 0, 1, 1, 0, and 0, respectively.
- metadata 1 regarding the music specified by the music ID of ABC123 or the music specified by the music ID of OPQ385 is regarded as a vector.
- each of the metadata 2 to metadata 6 regarding the music specified by the music ID of ABC123 or the music specified by the music ID of OPQ385 is regarded as a vector.
- one metadata value for multiple songs is regarded as a vector.
- metadata 1, metadata 3, and metadata 4 regarded as vectors are in clusters within a Manhattan distance of 1, and metadata 2, metadata 5, and metadata. 6 is lumped into other clusters within a Manhattan distance of one.
- FIG. 17 shows an example of metadata that is clustered and assigned layers in this way.
- metadata 1, metadata 3, and metadata 4 belong to the first layer
- metadata 2, metadata 5, and metadata 6 belong to the second layer.
- each layer is formed by a collection of highly correlated metadata, and content clustering is performed in it, so in normal hierarchies where the genre, artist, etc. are hierarchized as they are, Differences between subtle contents that cannot be expressed can be reflected in the cluster.
- step S204 the music clustering unit 202 clusters music for each layer. That is, the music clustering unit 202 classifies each content as one of a plurality of clusters in each of the assigned hierarchies.
- Step S205 and step S206 are the same as step S2 and step S3 in Fig. 6, respectively, and thus description thereof is omitted.
- the level of detail of the content expression by metadata (the level of detail of the expression)
- the content can be clustered by reducing the amount of data and the amount of calculation while maintaining the degree.
- FIG. 18 is a block diagram showing an example of the configuration of the similar user detection unit 203.
- the similar user detection unit 203 includes a normalization unit 231, a weight calculation unit 232, and a similarity calculation unit 233.
- the normalizing unit 231 normalizes a preference vector that is an example of user preference information.
- the weight calculation unit 232 calculates a weight for each layer for each user from the preference vector of each user that has been normalized.
- the similarity calculation unit 233 calculates a similarity indicating the degree of preference similarity between the user who recommends music and other users from the weights and preference titles for each layer.
- step S231 the normalization unit 231 normalizes each user's preference vector.
- FIG. 20 is a diagram showing an example of each user's preference vector generated by the preference vector generation unit 19 and held in the user history information DB 17. That is, FIG. 20 shows an example of the preference vector before being normalized.
- the first four elements belong to the first layer, the next four elements belong to the second layer, and the next three elements belong to the third layer. The last four elements belong to the fourth layer.
- the user preference vector specified by the user ID U001 is (0. 0, 2. 8, 0. 0, 2. 2, 0. 4, 0. 6 , 0. 8, 0. 0, 0. 5, 0. 4, 0. 4, 0. 0, 0. 5, 0. 4, 0. 0).
- the first four elements which are 0.0, 2. 8, 0. 0, 2.2 respectively, belong to the first layer, and are 0.4, 0. 6, 0. 8, 0.
- the next four elements that are 0 are the second
- the last four elements belong to the fourth layer.
- the user preference vector specified by the user ID U002 is (0. 2, 0. 8, 0. 5, 0. 6, 0. 0, 0.5. , 0. 5, 0. 0, 0. 7, 0. 3, 0. 6, 0. 0, 0. 6, 0. 2, 0. 0).
- the first four elements which are 0.2, 0. 8, 0. 5, 0.6, respectively, belong to the first layer and are 0.0, 0, 5, 0.5, 0.5.
- the next four elements that are 0 belong to the second layer, and the next three elements that are 0.7, 0. 3, and 0.6 respectively belong to the third layer and are 0.0, 0, 0, respectively.
- the last four elements, 6, 0. 2, 0. 0, belong to the fourth layer.
- the user preference vector specified by the user ID U003 is (0. 0, 2. 2, 0. 1, 1. 6, 0. 0, 1.0. , 2. 0, 1. 4, 0. 0, 1. 2, 0. 1, 0. 3, 0. 4, 0. 6, 0. 7).
- the first four elements which are 0.0, 2. 2, 0. 1, 1. 6, respectively, belong to the first layer, and are 0.0, 1. 0, 2. 0, 1.
- the next four elements that are 4 belong to the second layer, and the next three elements that are 0.0, 1. 2, and 0.1, respectively, belong to the third layer, and are 0.3, 0, respectively.
- the last four elements, 4, 0. 6, 0. 7, belong to the fourth layer.
- step S231 the normal part 231 normalizes each preference vector so that the norm in each layer is 1.
- FIG. 21 is a diagram showing an example of a preference vector obtained by regularizing the preference vector of FIG. 20 so that the norm in each layer is 1.
- FIG. 21 is a diagram showing an example of a preference vector obtained by regularizing the preference vector of FIG. 20 so that the norm in each layer is 1.
- the normalized preference vector of the user identified by the user ID U001 is (0. 0, 0. 8, 0. 0, 0. 6, 0.4. , 0. 6, 0. 7, 0. 0, 0. 7, 0. 5, 0. 5, 0. 0, 0. 5, 0. 4, 0. 0).
- the first four elements, which are 0. 0, 0. 8, 0. 0, 0. 6 respectively belong to the first layer and are 0. 4, 0. 6, 0. 7, 0. 0, respectively.
- the next four elements that belong to the second layer are 0.7, 0.5, 0.5, respectively, and the next three elements that belong to the third layer belong to the third layer, which are 0.0, 0, respectively.
- the last four elements, 5, 0. 4, 0. 0, belong to the fourth layer.
- the normalized preference vector of the user specified by the user ID U002 is (0. 2, 0. 7, 0. 4, 0. 5, 0. 0. , 0. 7, 0. 7, 0. 0, 0. 7, 0. 3, 0. 6, 0. 0, 0. 8, 0. 3, 0. 0).
- 0, 2, 0, 7, 0. 4, 0.5 respectively The first four elements belong to the first layer, and the next four elements, which are 0.0, 0. 7, 0. 7, and 0. 0, respectively, belong to the second layer and are each 0.7.
- the next three elements, 0. 3, 0. 6 belong to the third layer, and the last four elements, which are 0. 0, 0. 8, 0. 3, 0. 0 respectively, Belongs to a layer.
- the normalized preference vector of the user specified by the user ID U003 is (0. 0, 0. 8, 0. 0, 0. 6, 0. 0 , 0. 4, 0. 8, 0. 5, 0. 0, 1. 0, 0. 1, 0. 3, 0. 2, 0. 2, 0. 3).
- the first four elements, which are 0. 0, 0. 8, 0. 0, 0. 6 respectively belong to the first layer and are 0. 0, 0. 4, 0. 8, 0. 5 respectively.
- the next four elements belonging to the second layer belong to the second layer, and the next three elements that are 0.0, 1. 0, and 0.1 respectively belong to the third layer and are each 0.3, 0.
- step S232 the weight calculation unit 232 calculates the weight for each of the user's preference vector hierarchies. For example, in step S232, the weight calculation unit 232 calculates a weight that is a variance of elements belonging to one layer for each layer.
- FIG. 22 is a diagram illustrating an example of weights that are variances of elements belonging to each hierarchy, calculated for each hierarchy for each user.
- the weight of the first layer, the weight of the second layer, the weight of the third layer, and the weight of the fourth layer for the user specified by the user ID U001 are 0, respectively. 17, 0. 10, 0. 01, and 0.06.
- the weight of the first layer, the weight of the second layer, the weight of the third layer, and the weight of the fourth layer for the user specified by the user ID U002 are 0.05, 0. 17, 0.05, and 0.16.
- the weight of the first layer, the weight of the second layer, the weight of the third layer, and the weight of the fourth layer for the user specified by the user ID U003 are 0.16, 0. 10, 0.31, and 0.00.
- step S233 the similarity calculation unit 233 calculates the weighted preference similarity for each user.
- step S234 the similar user detecting unit 203 detects the user X having the highest preference similarity from the users, and the process ends.
- L is a value indicating the number of preference vector hierarchies
- 1 is a value specifying the preference vector hierarchies
- C (l) represents the entire cluster of preference vectors
- c is a value that identifies the cluster
- h indicates a value of a normalized preference vector element.
- the first layer element and the user preference vector elements specified by the user ID U002 When the elements in the first layer are multiplied by the corresponding elements, and the multiplied results are added up, the value placed in the first layer of the user ID U002 in FIG. 23 is obtained as 0.88.
- elements of the user preference vector specified by the user ID U001 and the user preference vector specified by the user ID U002 are multiplied by the corresponding elements, and the result of multiplication is added to each of the second, third, and fourth layers of the user ID that is U00 2 in FIG. The values 0.99, 0.97, and 0.50 are obtained.
- the preference similarity between the user specified by the user ID U001 and the user specified by the user ID U002 is the first layer, the second layer, and the third layer.
- the values obtained by caloring 0.88, 0.92, 0.97, and 0.50 obtained in this way are 3. 27.
- the value 1.00 which is the value placed in the first layer of the user ID U003 in FIG. 23, is obtained. It is done.
- the user preference vector element specified by the user ID U001 and the user preference specified by the user ID U003 are used. Tuttle elements are multiplied by the corresponding elements, and when the multiplied results are accumulated, they are placed in the second, third, and fourth layers of user ID U003 in FIG. The values 0.77, 0.57, 0.15 are calculated.
- the preference similarity between the user specified by the user ID U001 and the user specified by the user ID U003 is the first layer, the second layer, the third layer , And the 4th layer, respectively, are calculated as 2.50, which is the value obtained by caloring 1.00, 0.77, 0.57, and 0.15.
- the preference similarity U001 between the user identified by the user ID U001 and the user identified by the user ID U002 is U001.
- the user ID specified by the user ID and the user ID specified by the user ID U003 is greater than the preference similarity, so the user ID U002 is the user ID having the maximum preference similarity.
- the user specified by is detected.
- step S233 the similarity calculation unit 233 calculates the weighted similarity sim (u, v) of the user u and the user V according to the equation (2).
- L is a value indicating the number of preference vector hierarchies
- 1 is a value specifying the preference vector hierarchies.
- C (l) represents the entire cluster of preference vectors
- c is a value that identifies the cluster.
- h indicates a value of a normalized preference vector element.
- b indicates the weight of each layer! /.
- Figure 24 shows the case where user X is a user identified by a user ID U001 and a user identified by a user ID U002 and a user identified by a user ID U003. It is a figure which shows the example of the similarity of weighted preference.
- the values shown in FIG. 24 are values obtained by multiplying the similarity sim (u, v) calculated by equation (2) by 100.
- each of the elements of the first layer has the weight of the first layer of the user specified by the user ID U001.
- the first layer Each of the elements is multiplied by the weight of the first layer of the user specified by the user ID U002, multiplied by the corresponding elements, and the multiplied results are added up to calculate U 002 in FIG.
- the value 0.72 that is the value placed in the first layer of the user ID is obtained.
- each of the elements of the user preference vector specified by the user ID U001 is the interest specified by the user ID U001.
- the weight of the first layer of the user is multiplied, and the weight of the first layer of the user specified by the user ID of U002 is added to each element of the user preference vector specified by the user ID of U002.
- the weighted preference similarity between the user specified by the user ID U001 and the user specified by the user ID U002 is the first layer, the second layer, It is assumed that 2.76, 1.54, 0.03, and 0.48 obtained by calculating each of the third and fourth layers are 2.76.
- each of the elements of the first layer of the user preference vector elements of the user ID U001 is the first layer of the user specified by the user ID U001.
- the first layer of the user specified by the user ID U003 is added to each of the first layer elements of the user preference vector elements specified by the user ID U003 multiplied by the weight. If the corresponding weights are multiplied, the corresponding elements are multiplied, and the multiplied results are integrated, the value 2.74, which is the value placed in the first layer of the user ID U003 in FIG. 24, is obtained.
- the user's preference vector identified by the user ID U001 is added to each of the elements of the user preference vector identified by the user ID U001.
- Each of the user's preference vector elements specified by the user ID U003 is multiplied by the weight of the first layer, and the first layer weight of the user specified by the user ID U003 is multiplied to correspond.
- the multiplied results are multiplied, and the multiplied results are accumulated, the values are arranged in the second layer, the third layer, and the fourth layer of the user ID U 003 in FIG. , 0.10, 1.00 force is required.
- the user identified by the user ID U001 and the user ID U003 are special.
- the similarity of the weighted preference with the defined users is the following: 2.74, 0.79, obtained for each of the first, second, third, and fourth layers. It is 3.64, which is the value obtained by caloring 0.10 and 0.00.
- the value of each element of the user's preference vector specified by the user ID U001 is the first value compared to the second to fourth layers. As compared to the second to fourth layers, the value of each element in the first layer is predicted to be related to the user's preference specified by the user ID U001.
- the user preference vector specified by the user ID U002 and the value of each element of the first layer of the user preference vector specified by the user ID U003 are U003.
- the value of each element of the first layer of the user's preference vector specified by the user ID is U001 by the value of each element of the first layer of the user's preference vector specified by the user ID U002. It approximates the value of each element of the first layer of the user's preference vector specified by a user ID. Therefore, compared to the user specified by the user ID U002, the user preference specified by the user ID U003 is similar to the user preference specified by the user ID U001. It is predicted that
- the value changes more greatly depending on the value predicted to be related to the user's preference compared to the value expected to be less related to the user's preference. Since the similarity of preferences can be obtained, users with similar preferences can be detected more accurately.
- step S232 the weight calculation unit 232 has been described as calculating weights that are the variances of elements belonging to each layer, for example.
- the present invention is not limited to this.
- the entropy H is calculated by the equation (3) and the entropy H is subtracted from 1. Try to calculate a certain weight.
- the program may be processed by a single computer, or may be distributedly processed by a plurality of computers. Furthermore, the program may be transferred to a remote computer and executed.
- the system represents the entire apparatus constituted by a plurality of apparatuses.
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
Description
Claims
Priority Applications (3)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
EP06798036A EP1835419A4 (en) | 2005-09-28 | 2006-09-15 | DEVICE, METHOD AND INFORMATION PROGRAM FOR INFORMATION PROCESSING |
CN200680001299A CN100594496C (zh) | 2005-09-28 | 2006-09-15 | 信息处理装置和方法 |
US11/791,361 US8117211B2 (en) | 2005-09-28 | 2006-09-15 | Information processing device and method, and program |
Applications Claiming Priority (4)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2005-281357 | 2005-09-28 | ||
JP2005281357 | 2005-09-28 | ||
JP2006-151012 | 2006-05-31 | ||
JP2006151012A JP4378646B2 (ja) | 2005-09-28 | 2006-05-31 | 情報処理装置、情報処理方法、およびプログラム |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2007037139A1 true WO2007037139A1 (ja) | 2007-04-05 |
Family
ID=37899569
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2006/318373 WO2007037139A1 (ja) | 2005-09-28 | 2006-09-15 | 情報処理装置、方法、およびプログラム |
Country Status (6)
Country | Link |
---|---|
US (1) | US8117211B2 (ja) |
EP (1) | EP1835419A4 (ja) |
JP (1) | JP4378646B2 (ja) |
KR (1) | KR20080045659A (ja) |
CN (1) | CN100594496C (ja) |
WO (1) | WO2007037139A1 (ja) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108920630A (zh) * | 2018-06-29 | 2018-11-30 | 联想(北京)有限公司 | 一种数据处理方法及电子设备 |
Families Citing this family (48)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US9065727B1 (en) | 2012-08-31 | 2015-06-23 | Google Inc. | Device identifier similarity models derived from online event signals |
US8244673B2 (en) * | 2006-08-30 | 2012-08-14 | Panasonic Corporation | Information presenting device, information presenting method, information presenting program, and integrated circuit |
JP5392808B2 (ja) * | 2007-11-27 | 2014-01-22 | 日本電気株式会社 | コンテクスト対応情報推薦装置、コンテクスト対応情報推薦方法およびコンテクスト対応情報推薦プログラム |
JP4538758B2 (ja) * | 2007-12-04 | 2010-09-08 | ソニー株式会社 | 情報処理装置および方法、プログラム、並びに記録媒体 |
EP2068276A1 (en) | 2007-12-04 | 2009-06-10 | Sony Corporation | Information processing device and method, program, and recording medium |
JP4433326B2 (ja) | 2007-12-04 | 2010-03-17 | ソニー株式会社 | 情報処理装置および方法、並びにプログラム |
JP4538760B2 (ja) * | 2007-12-04 | 2010-09-08 | ソニー株式会社 | 情報処理装置および方法、プログラム、並びに記録媒体 |
US20090198666A1 (en) * | 2008-02-01 | 2009-08-06 | Winston Ronald H | Affinity matching system and method |
CN101547167A (zh) * | 2008-03-25 | 2009-09-30 | 华为技术有限公司 | 内容分类方法、装置及系统 |
JP5166949B2 (ja) * | 2008-04-10 | 2013-03-21 | 株式会社エヌ・ティ・ティ・ドコモ | レコメンド情報生成装置およびレコメンド情報生成方法 |
JP4790753B2 (ja) * | 2008-04-28 | 2011-10-12 | 日本電信電話株式会社 | 情報推薦装置、情報推薦方法、プログラム及び記録媒体 |
US8386486B2 (en) * | 2008-07-02 | 2013-02-26 | Palo Alto Research Center Incorporated | Method for facilitating social networking based on fashion-related information |
CN101661483B (zh) * | 2008-08-29 | 2012-10-03 | 株式会社理光 | 一种推荐系统及推荐方法 |
JP4650552B2 (ja) * | 2008-10-14 | 2011-03-16 | ソニー株式会社 | 電子機器、コンテンツ推薦方法及びプログラム |
KR101111121B1 (ko) * | 2009-06-29 | 2012-02-13 | 주식회사 모임 | 인공지능에 기반한 개인별 맞춤형 컨텐츠 제공 장치 및 방법, 그 기록 매체 |
US8635255B2 (en) * | 2009-06-30 | 2014-01-21 | Verizon Patent And Licensing Inc. | Methods and systems for automatically customizing an interaction experience of a user with a media content application |
EP2481018A4 (en) * | 2009-09-21 | 2013-06-12 | Ericsson Telefon Ab L M | METHOD AND DEVICE FOR PERFORMING A RECOMMENDATION |
JP5609056B2 (ja) * | 2009-10-14 | 2014-10-22 | ソニー株式会社 | コンテンツ関係可視化装置、表示制御装置、コンテンツ関係可視化方法およびプログラム |
JP2011158980A (ja) * | 2010-01-29 | 2011-08-18 | Brother Industries Ltd | 消費者情報処理装置 |
CN102870114B (zh) * | 2010-04-28 | 2016-05-25 | Jvc建伍株式会社 | 项目选择装置和项目选择方法 |
CN102486926B (zh) * | 2010-12-03 | 2015-04-08 | 上海博泰悦臻电子设备制造有限公司 | 个性化音乐媒体信息获取方法及系统 |
US8719248B2 (en) * | 2011-05-26 | 2014-05-06 | Verizon Patent And Licensing Inc. | Semantic-based search engine for content |
US8990208B2 (en) * | 2011-09-22 | 2015-03-24 | Fujitsu Limited | Information management and networking |
JP5727920B2 (ja) * | 2011-11-21 | 2015-06-03 | 株式会社Nttドコモ | レコメンドコンテンツ出力装置及びレコメンドコンテンツ出力方法 |
JP5696025B2 (ja) * | 2011-11-22 | 2015-04-08 | 日本電信電話株式会社 | 商品情報推薦装置及び方法及びプログラム |
JP5785864B2 (ja) * | 2011-12-06 | 2015-09-30 | 株式会社日立製作所 | 製造ラインシミュレーション装置、データ入力支援装置、およびデータ入力支援方法 |
US9053185B1 (en) | 2012-04-30 | 2015-06-09 | Google Inc. | Generating a representative model for a plurality of models identified by similar feature data |
US8914500B1 (en) | 2012-05-21 | 2014-12-16 | Google Inc. | Creating a classifier model to determine whether a network user should be added to a list |
JP5846442B2 (ja) * | 2012-05-28 | 2016-01-20 | ソニー株式会社 | 情報処理装置、情報処理方法、および、プログラム |
JP6127388B2 (ja) * | 2012-06-13 | 2017-05-17 | ソニー株式会社 | 情報処理装置、情報処理方法、およびプログラム |
US8886575B1 (en) | 2012-06-27 | 2014-11-11 | Google Inc. | Selecting an algorithm for identifying similar user identifiers based on predicted click-through-rate |
US8874589B1 (en) | 2012-07-16 | 2014-10-28 | Google Inc. | Adjust similar users identification based on performance feedback |
US8782197B1 (en) | 2012-07-17 | 2014-07-15 | Google, Inc. | Determining a model refresh rate |
US8886799B1 (en) | 2012-08-29 | 2014-11-11 | Google Inc. | Identifying a similar user identifier |
JP5235251B1 (ja) * | 2012-09-28 | 2013-07-10 | 楽天株式会社 | 情報処理装置、情報処理方法及び情報処理プログラム |
KR101464044B1 (ko) * | 2012-09-28 | 2014-11-20 | 주식회사 엘지유플러스 | 관심 키워드 제공 장치 및 방법 |
CN102917269B (zh) * | 2012-09-29 | 2016-06-29 | 青岛海信电器股份有限公司 | 一种电视节目推荐系统及方法 |
JP6014515B2 (ja) * | 2013-02-22 | 2016-10-25 | 株式会社エヌ・ティ・ティ・データ | レコメンド情報提供システム、レコメンド情報生成装置、レコメンド情報提供方法、およびプログラム |
KR101581917B1 (ko) * | 2014-08-12 | 2016-01-04 | 네이버 주식회사 | 컨텐츠 표시 제어 장치, 컨텐츠 표시 제어 방법 및 이를 컴퓨터에서 실행하기 위한 컴퓨터 프로그램 |
CN104731954B (zh) * | 2015-04-01 | 2018-01-26 | 天翼爱音乐文化科技有限公司 | 基于群透视音乐推荐方法与系统 |
WO2017133007A1 (en) * | 2016-02-05 | 2017-08-10 | Hewlett Packard Enterprise Development Lp | User interest and relationship determination |
CN108681919A (zh) * | 2018-05-10 | 2018-10-19 | 苏州跃盟信息科技有限公司 | 一种内容推送方法及装置 |
CN109086813B (zh) * | 2018-07-23 | 2021-05-14 | 广州虎牙信息科技有限公司 | 主播相似度的确定方法、装置、设备及存储介质 |
KR101922725B1 (ko) * | 2018-07-25 | 2018-11-27 | (주)엔터랩 | 인공지능 과제매칭 방법, 장치 및 프로그램 |
JP7003088B2 (ja) * | 2019-05-27 | 2022-01-20 | 楽天グループ株式会社 | レコメンド装置、レコメンド方法、及びレコメンドプログラム |
KR102284393B1 (ko) * | 2020-08-26 | 2021-08-04 | (주)인더스트리미디어 | 학습 콘텐츠 추천 장치 및 방법 |
CN113949931A (zh) * | 2021-10-13 | 2022-01-18 | 深圳市视壮科技有限公司 | 一种iptv节目推荐方法及系统 |
WO2024084691A1 (ja) * | 2022-10-21 | 2024-04-25 | ヤマハ発動機株式会社 | 対話型提案出力システム |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2003216636A (ja) * | 2002-01-23 | 2003-07-31 | Hitachi Ltd | 最新記事推薦方法およびシステム |
JP2004178521A (ja) * | 2002-11-29 | 2004-06-24 | Superstation Inc | 推薦システム |
JP2004355340A (ja) * | 2003-05-29 | 2004-12-16 | Sony Corp | 情報処理装置および方法、プログラム、並びに記録媒体 |
Family Cites Families (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7617127B2 (en) * | 2000-04-28 | 2009-11-10 | Netflix, Inc. | Approach for estimating user ratings of items |
JP3672023B2 (ja) * | 2001-04-23 | 2005-07-13 | 日本電気株式会社 | 番組推薦システムおよび番組推薦方法 |
US20030033370A1 (en) * | 2001-08-07 | 2003-02-13 | Nicholas Trotta | Media-related content personalization |
US9374451B2 (en) * | 2002-02-04 | 2016-06-21 | Nokia Technologies Oy | System and method for multimodal short-cuts to digital services |
JP4161701B2 (ja) | 2002-12-12 | 2008-10-08 | ソニー株式会社 | 情報処理装置および情報処理方法、記録媒体、並びにプログラム |
JP2004194108A (ja) | 2002-12-12 | 2004-07-08 | Sony Corp | 情報処理装置および情報処理方法、記録媒体、並びにプログラム |
EP1571835A4 (en) | 2002-12-12 | 2010-10-20 | Sony Corp | DEVICE, METHOD AND SYSTEM FOR PROCESSING DATA, RECORDING MEDIUM, AND PROGRAM |
JP4547596B2 (ja) | 2002-12-12 | 2010-09-22 | ソニー株式会社 | 情報処理装置および情報処理方法、並びにプログラム |
JP4003127B2 (ja) | 2002-12-12 | 2007-11-07 | ソニー株式会社 | 情報処理装置および情報処理方法、情報処理システム、記録媒体、並びにプログラム |
JPWO2004055302A1 (ja) | 2002-12-13 | 2006-04-20 | 松下電器産業株式会社 | 車載機器制御システム、車載機器制御装置、及び車載機器制御方法 |
-
2006
- 2006-05-31 JP JP2006151012A patent/JP4378646B2/ja not_active Expired - Fee Related
- 2006-09-15 KR KR1020077011814A patent/KR20080045659A/ko active IP Right Grant
- 2006-09-15 US US11/791,361 patent/US8117211B2/en not_active Expired - Fee Related
- 2006-09-15 CN CN200680001299A patent/CN100594496C/zh not_active Expired - Fee Related
- 2006-09-15 WO PCT/JP2006/318373 patent/WO2007037139A1/ja active Application Filing
- 2006-09-15 EP EP06798036A patent/EP1835419A4/en not_active Withdrawn
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2003216636A (ja) * | 2002-01-23 | 2003-07-31 | Hitachi Ltd | 最新記事推薦方法およびシステム |
JP2004178521A (ja) * | 2002-11-29 | 2004-06-24 | Superstation Inc | 推薦システム |
JP2004355340A (ja) * | 2003-05-29 | 2004-12-16 | Sony Corp | 情報処理装置および方法、プログラム、並びに記録媒体 |
Non-Patent Citations (1)
Title |
---|
See also references of EP1835419A4 * |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108920630A (zh) * | 2018-06-29 | 2018-11-30 | 联想(北京)有限公司 | 一种数据处理方法及电子设备 |
Also Published As
Publication number | Publication date |
---|---|
CN101069184A (zh) | 2007-11-07 |
JP4378646B2 (ja) | 2009-12-09 |
US8117211B2 (en) | 2012-02-14 |
EP1835419A4 (en) | 2009-09-16 |
EP1835419A1 (en) | 2007-09-19 |
JP2007122683A (ja) | 2007-05-17 |
US20090077132A1 (en) | 2009-03-19 |
CN100594496C (zh) | 2010-03-17 |
KR20080045659A (ko) | 2008-05-23 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
JP4378646B2 (ja) | 情報処理装置、情報処理方法、およびプログラム | |
US7953735B2 (en) | Information processing apparatus, method and program | |
Liang et al. | Connecting users and items with weighted tags for personalized item recommendations | |
US7743059B2 (en) | Cluster-based management of collections of items | |
US8560545B2 (en) | Item recommendation system which considers user ratings of item clusters | |
US7966225B2 (en) | Method, system, and medium for cluster-based categorization and presentation of item recommendations | |
US20140040280A1 (en) | System and method for identifying similar media objects | |
US20100325126A1 (en) | Recommendation based on low-rank approximation | |
US20090063568A1 (en) | Method and apparatus for constructing user profile using content tag, and method for content recommendation using the constructed user profile | |
JP2011175362A (ja) | 情報処理装置、重要度算出方法及びプログラム | |
US20080243816A1 (en) | Processes for calculating item distances and performing item clustering | |
WO2017013667A1 (en) | Method for product search using the user-weighted, attribute-based, sort-ordering and system thereof | |
US20090055376A1 (en) | System and method for identifying similar media objects | |
WO2007063328A2 (en) | Information retrieval system and method using a bayesian algorithm based on probabilistic similarity scores | |
US20060085405A1 (en) | Method for analyzing and classifying electronic document | |
Chang et al. | Music recommender using deep embedding-based features and behavior-based reinforcement learning | |
WO2008121872A1 (en) | Cluster-based assessment of user interests | |
KR102249466B1 (ko) | 인공지능 추천 모델을 사용하여 추천 정보를 제공하는 데이터 카탈로그 제공 방법 및 시스템 | |
Donaldson | A hybrid social-acoustic recommendation system for popular music | |
Gaikwad et al. | E-commerce recommendation system using improved probabilistic model | |
JP2007115222A (ja) | 情報処理装置および方法、並びにプログラム | |
Lin et al. | Association rule mining with a correlation-based interestingness measure for video semantic concept detection | |
CN111708945A (zh) | 产品的推荐方法及装置、电子设备、计算机存储介质 | |
Hsu et al. | Designing a graph-based framework to support a multi-modal approach for music information retrieval | |
JP2007183927A (ja) | 情報処理装置および方法、並びにプログラム |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
WWE | Wipo information: entry into national phase |
Ref document number: 2006798036 Country of ref document: EP |
|
WWE | Wipo information: entry into national phase |
Ref document number: 3809/DELNP/2007 Country of ref document: IN |
|
WWE | Wipo information: entry into national phase |
Ref document number: 1020077011814 Country of ref document: KR |
|
WWE | Wipo information: entry into national phase |
Ref document number: 200680001299.6 Country of ref document: CN |
|
121 | Ep: the epo has been informed by wipo that ep was designated in this application | ||
WWP | Wipo information: published in national office |
Ref document number: 2006798036 Country of ref document: EP |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
WWE | Wipo information: entry into national phase |
Ref document number: 11791361 Country of ref document: US |