JP2006048319A - Device, method, recording medium, and program for information processing - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To learn user's taste while taking temporal change of user's taste into consideration. <P>SOLUTION: When a content reproducing device 1 is operated by the user to reproduce contents, content information on the reproduced contents and operation information showing the reproducing operation are supplied to a predicting device 32. The predicting device 32 when supplied with the content information and operation information from the reproducing device 1 stores them in a user history DB as user history information showing a user's time-series viewing history of contents. The predicting device 32 generates prediction curves wherein temporary change of user's taste is predicted by the contents by using the user history information and supplies them to a recommendation part 33. The recommendation part 33 when supplied with a recommendation request from an operation part 2 supplies the contents having the highest taste among the prediction curves of the respective contents as recommended contents to a reproduction part 3. This invention is applicable to, for example, a learning device learning user's taste. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to an information processing apparatus and method, a recording medium, and a program, and more particularly, to an information processing apparatus and method, a recording medium, and a program that perform preference learning in consideration of temporal changes in user preferences.

  There is a content recommendation system that recommends content that matches a user's preference from among a large number of contents based on the user's own preference information.

  The user's own preference information used in the content recommendation system is, for example, the user's preference information is registered in advance, or the user's preference information is learned using an appreciation history that is a history of the user's appreciation of the content. Is generated.

  For example, a case where the content is music (music) will be described. In the method in which the user generates preference information by registering his / her preference information in advance, the user can select, for example, the genre of the song he / she likes (for example, rock, pops, classic, etc.), and the feature amount of the favorite song. The user's preference information is generated by registering the value (for example, tempo, mood, loudness, etc.) or the name of a favorite singer or band.

  In addition, in the method of generating preference information by learning the user's preference using the appreciation history, for example, in a content reproduction device that reproduces content, the user selects a song every time the user selects (listens to a song). Information on the music (the above-mentioned genre, each feature amount, the name of the singer, etc.) is analyzed or acquired and accumulated, whereby user preference information is generated. For example, in Patent Document 1, content is recommended based on information using a user's reference history.

  FIG. 1 shows an example of the configuration of a conventional content recommendation system.

  The content recommendation system in FIG. 1 includes a content reproduction device 1 and a content recommendation device 11.

  The content reproduction apparatus 1 includes an operation unit 2, a reproduction unit 3, and a content DB (Data Base) 4. The content recommendation device 11 includes a preference learning device 12 and a recommendation unit 13.

  When the user performs an operation of reproducing predetermined content (music) on the operation unit 2, the operation unit 2 specifies the content to the reproduction unit 3 and supplies a reproduction command. In addition, the operation unit 2 supplies, for example, operation details operated by the user, such as playback, stop, and fast forward, to the preference learning device 12 of the content recommendation device 11 as operation information.

  Further, when the user wants to listen to music by selecting (recommending) the optimum content for the user without listening to the specific content, the user requests the content recommendation in the operation unit 2. Perform the operation. The operation unit 2 outputs a recommendation request to the recommendation unit 13 of the content recommendation device 11 when an operation for requesting content recommendation is performed by the user.

  Based on the content playback command supplied from the operation unit 2, the playback unit 3 acquires content specified by the user from the content DB 4 and plays it back. In addition, the playback unit 3 includes information such as the title, genre (for example, rock, pops, and classic) of the content being played, feature amounts (for example, tempo, mood, and volume of sound), and singer name ( (Hereinafter referred to as content information as appropriate) is supplied to the preference learning device 12 of the content recommendation device 11.

  In addition, when the recommended genre of content is supplied from the recommendation unit 13 of the content recommendation device 11 as a response to the recommendation request from the operation unit 2, the playback unit 3 acquires content corresponding to the recommended genre from the content DB 4, Reproduce.

  The content DB 4 stores content (music) acquired from a predetermined recording medium or a content server (both not shown) that provides the content.

  The preference learning device 12 of the content recommendation device 11 is based on the operation information or the content information supplied from the operation unit 2 or the reproduction unit 3, respectively, and the user's preference level for each content genre (rock, pops, and classic) ( The preference learning result obtained as a result is supplied to the recommendation unit 13.

  When a recommendation request is supplied from the operation unit 2 of the content reproduction device 1, the recommendation unit 13 determines a genre optimal for the user based on the preference learning result supplied from the preference learning device 12. The selected genre is supplied to the playback unit 3 of the content playback apparatus 1 as a recommended genre.

  In the content recommendation system configured as described above, the content information of the reproduced content and the operation information corresponding to the user's operation are supplied from the content reproduction device 1 to the content recommendation device 11. The content recommendation device 11 learns the user's preference for each genre based on the content information and the operation information, and when there is a recommendation request for recommending content from the content reproduction device 1, the content recommendation device 11 asks the user based on the learning result. Recommend (supply) the optimal genre.

  FIG. 2 shows an example of learning by the preference learning device 12 based on content information and operation information. In FIG. 2, the user's preference for each genre is set in the range of 0 to 1, and the larger the value, the stronger the user's preference (the user likes it).

In the preference learning unit 12, the preference of the user for each genre, by learning, the transition of the state T ₁ to T ₆ as shown in FIG.

That is, in the initial state T ₁ , the preference degrees of the genres of rock, pop, and classic are all 0.5.

  The user voluntarily reproduces the music belonging to the genre of pops in the content reproduction apparatus 1 (hereinafter referred to as “pops reproduction, etc., as well as rock and classic”). In this case, the content information of the music being played and the operation information indicating that the playback operation has been performed are supplied from the content playback device 1 to the preference learning device 12 of the content recommendation device 11.

When the preference learning device 12 receives the content information and the operation information indicating that the pops are reproduced, the preference learning device 12 determines that the user likes the pops, and adds 0.1 to the pops preference level. Therefore, in the state T ₂ of the after the reproduction operation is performed, rock, pop, or preference of each classical genre, respectively, has become a 0.5,0.6,0.5.

  In addition, the user performs an operation (hereinafter referred to as “dislike” operation) indicating that “the music is disliked” in the operation unit 2 while the content reproduction apparatus 1 is reproducing the classic. In this case, the content information of the music being played and the operation information indicating that the “dislike” operation has been performed are supplied from the content reproduction device 1 to the content recommendation device 11.

The preference learning device 12 subtracts 0.3 from the preference level of the classic based on the content information and the operation information indicating that the “dislike” operation has been performed on the classic. Therefore, in the state T ₃ after the "hate" operation has been performed, rock, pop or preference of each genre of classic,, respectively, and has a 0.5,0.6,0.2.

  At some point, while the content reproduction apparatus 1 is reproducing the lock, the operation unit 2 performs a fast-forward operation for fast-forwarding the reproduction of the music. In this case, content information of the music being played and operation information indicating that a fast-forward operation has been performed are supplied from the content reproduction device 1 to the content recommendation device 11.

The preference learning device 12 determines that the user does not like the lock based on the content information and the operation information indicating that the fast-forward operation has been performed on the lock, and subtracts 0.1 from the preference degree of the lock. Therefore, in the state T ₄ after the fast-forward operation is performed, rock, pop, or preference of each classical genre, respectively, has become a 0.4,0.6,0.2.

In this state T ₄ , when a recommendation request is supplied from the content reproduction device 1 to the content recommendation device 11, the content recommendation device 11 has the largest pop preference among the three genres. Therefore, pops are supplied to the content reproduction apparatus 1 as a recommended genre.

Further, as long as the preference level of each genre of the preference learning device 12 is in the state T ₄ , the content recommendation device 11 continues pops even if the recommendation request is supplied from the content reproduction device 1 to the content recommendation device 11. And recommend. As a result, the user is only listened to pop music, and gets tired of pop music. Finally, it is assumed that the user performs a “dislike” operation on the operation unit 2 while the pop is being played.

The preference learning device 12 subtracts 0.3 from the preference level of the pops based on the content information and the operation information indicating that the “dislike” operation has been performed on the pops. Accordingly, the preference level of each genre user stored in the preference learning device 12 is the state T ₅ , and the preference levels of rock, pop, or classic are 0.4, 0.3, and 0.2, respectively.

In the state T ₅ , the preference degree of the lock is the largest among the preference degrees of the three genres, so that the content recommendation device 11 supplies the lock to the content reproduction device 1 as a recommended genre this time.

  Since the pop is no longer recommended, the user reproduces the pop spontaneously in the content reproduction apparatus 1. In addition, the user performs an operation (hereinafter referred to as “like” operation) indicating that “I like this music” in the operation unit 2 while the content reproduction apparatus 1 is playing pops. In this case, content information of the music being played and operation information indicating that a playback operation and a “like” operation have been performed are supplied from the content playback device 1 to the content recommendation device 11.

The preference learning device 12 sets the pop preference degree to 0.4 (0.1 corresponding to the reproduction operation and “like” based on the content information and the operation information indicating that the reproduction operation and “like” operation of the pop are performed. "Total value of 0.3 corresponding to operation" is added. Accordingly, in the state T ₆ , the preference degrees of the rock, pop, and classic genres are 0.4, 0.7, and 0.2, respectively.

  In the preference learning of the preference learning device 12 described above with reference to FIG. 2, when the user is watching pops continuously, a phenomenon that the user eventually gets bored with the pops occurs. Until they receive some operation information indicating that they don't like pops, they recommend only pops. In addition, the operation that indicates that the user performed “I hate pops” (“I hate” operation) is “I hate” even though I did n’t really hate pops because I was bored temporarily. After the operation information indicating that the operation has been performed is supplied to the content recommendation device 11, the preference learning device 12 no longer recommends pops and recommends only locks.

  FIG. 3 shows another example of learning by the preference learning device 12 based on content information and operation information.

Initially, the preference learning unit 12, the preference of each genre of the user is in the state T _11. That is, the preference degrees of the genres of rock, pops, and classic are 0.3, 0.5, and 0.6, respectively. Thus, the content recommendation device 11, based on the preference of each genre of the state T _11, intensively recommended classic.

  By the way, users loved rock when they were young, and their preference for rock was strongest at that time. Since the user has not listened to the rock recently, he / she wants to listen to the rock that he had listened to for the first time in several years.

  However, in the preference learning by the preference learning device 12, since the user has a high degree of preference for classics, the lock is not recommended.

  Therefore, the user spontaneously reproduces the lock in the content reproduction apparatus 1. Further, the user performs a “like” operation on the operation unit 2 while the content reproduction apparatus 1 is reproducing the lock.

The preference learning device 12 sets the lock preference level to 0.4 (0.1 corresponding to the reproduction operation) based on the content information and the operation information indicating that the reproduction operation and the “like” operation are performed on the lock. The total value of 0.3 corresponding to the “like” operation is added. Therefore, in the state T _12, rock, pop, or preference of each classical genre, respectively, the 0.7,0.5,0.6.

According to the preference of the user for each genre of the state T _12, the content recommendation device 11, in turn, will recommend locking only. However, the user does not currently like the lock so much, and what the user currently likes is also classic, so the user is using the preference learning device 12 by stopping the lock being played back by the content playback device 1. Perform “dislike” operations to correct the degree.

The preference learning device 12 sets the lock preference level to 0.4 (0.1 corresponding to the stop operation) based on the content information and the operation information indicating that the stop operation and the “dislike” operation are performed on the lock. The total value of 0.3 corresponding to the “dislike” operation is subtracted. Therefore, in the state T _13, rock, pop, or preference of each classical genre, respectively, the 0.3,0.5,0.6.

  In the preference learning of the preference learning device 12 described above with reference to FIG. 3, although it does not apply to the user's everyday preference, a genre that he wants to listen to occasionally is not recommended. Then, the content recommendation device 11 regards the lock that the user wishes to occasionally listen to and plays as a change in the user's preference, and thereafter recommends only the lock.

  FIG. 4 shows still another example of learning by the preference learning device 12 based on content information and operation information.

In general, in preference learning, as the learning progresses, there is a tendency that the difference in preference between a user's favorite genre and a disliked genre increases. FIG state T ₂₁ 4 shows a state in which the difference in preference degree is wide open with the favorite genres and dislikes genre so user. That is, in the state T ₂₁ , the preference levels of the genres of rock, pops, and classic are 1.0, 0.4, and 0.0, respectively, and the lock with strong user preference has the highest preference level. 1.0, and the classic with low user preference has a minimum preference of 0.0.

In such a state T ₂₁ , even if the user next performs some operation indicating that he dislikes “lock” or some operation indicating that he likes “classic”, There is no change in the magnitude relationship between the preference levels of the genre, and no change occurs in the recommended genre.

For example, based on the preference level of each genre in the state T ₂₁ , the user performs a stop operation while the content reproduction device 1 is reproducing the lock recommended by the content recommendation device 11, and further spontaneously plays the classic. Assume that an operation for playback is performed. In this case, as shown in FIG. 4, the preference learning device 12 sets the lock preference degree to 0 based on the content information and the operation information indicating that the stop operation for the lock and the “play” operation for the classic are performed. .1 is subtracted (state T ₂₂ ), and 0.1 is added to the classic preference (state T ₂₃ ).

Next, it is assumed that the user performs an operation of spontaneously reproducing the classic again, and performs a fast-forward operation on the lock reproduced by the recommendation from the content recommendation device 11 after the reproduction of the classic. In this case, as shown in FIG. 4, the preference learning device 12 sets the classic preference level to 0.1 based on the content information and the operation information indicating that the playback operation for the classic and the fast-forward operation for the lock are performed. Is added (state T ₂₄ ), and 0.1 is subtracted from the lock preference degree (state T ₂₅ ). Therefore, in the state T ₂₅ , the preference degrees of the rock, pop, and classic genres are 0.8, 0.4, and 0.2, respectively.

Incidentally, in the state T ₂₁ to the state T ₂₅ of FIG. 4 (strong preference) largest among the preference three genres of rock, the preference of the classical is smallest among the three genres ( There is no change in the relationship between the degree of preference of each genre of rock, pop, and classic. Accordingly, in the state T ₂₁ to the state T _25, the content recommendation device 11 would recommend locking only.

  That is, in the preference learning of the preference learning device 12 described above with reference to FIG. 4, it falls into a local solution as the learning progresses, that is, with some content information and operation information, the magnitude relationship of the preference degree between each genre. As a result, only content of the same genre is recommended.

  In the state shown in FIG. 4, for example, the content reproduction apparatus 1 uses a genre of music recommended by the content recommendation apparatus 11 in FIG. When creating a playlist, the created playlist is a collection of songs of the same genre.

Japanese Patent Laid-Open No. 11-232278

  In the user preference learning in the above-described conventional content recommendation device, the example shown in FIG. 2 has a problem that it cannot cope with a short-term preference change (temporary boredom). Further, in the example shown in FIG. 3, there is a problem that it is impossible to cope with a long-term preference change. Further, in the example shown in FIG. 4, there is a problem that it falls into a local solution. These problems are considered to occur because preference learning does not consider temporal changes in user preferences.

  The present invention has been made in view of such circumstances, and makes it possible to perform preference learning in consideration of temporal changes in user preferences.

  The information processing apparatus according to the present invention uses a storage unit that stores user history information representing a time-series viewing history of a user's content, and a user preference level for each content using the user history information stored in the storage unit. And a predicting means for predicting a temporal change in the above.

  It is possible to further provide a recommendation unit that recommends the content having the highest user preference level at the present time among the temporal changes in the user preference level for each of the plurality of contents predicted by the prediction unit.

  The prediction means generates a prediction curve of the user's preference for the content based on a predetermined rule determined according to the pattern of the user's appreciation history, so that the temporal preference of the user's preference is determined for each content. Change can be predicted.

  In the prediction means, by analyzing the frequency of the waveform generated from the user's appreciation history, a characteristic waveform representing the appreciation characteristic for the content is obtained, and by generating a prediction curve of the degree of preference for the content from the characteristic waveform, For each content, a temporal change in the user's preference can be predicted.

  The prediction means can predict a temporal change in the user's preference degree for each feature amount of the content.

  The information processing method of the present invention uses a storage step for storing user history information representing a time-series viewing history of a user's content in a storage unit, and user history information stored in the storage unit, for each content. And a prediction step for predicting a temporal change in the degree of preference.

  The program recorded in the recording medium of the present invention uses a storage step for storing user history information representing a time-series viewing history of the user's content in the storage unit, and user history information stored in the storage unit, A prediction step of predicting temporal changes in the user's preference level for each content.

  The program of the present invention uses a storage step for storing user history information representing a time-series viewing history of a user's content in a storage unit, and a user's preference for each content using the user history information stored in the storage unit. The computer is caused to execute a process including a prediction step for predicting a temporal change in the degree.

  In the present invention, user history information representing a time-series viewing history of the user's content is stored in the storage means, and the user's preference time is determined for each content using the user history information stored in the storage means. Changes are expected.

  According to the present invention, preference learning can be performed in consideration of temporal changes in user preferences.

  Embodiments of the present invention will be described below. Correspondences between constituent elements described in the claims and specific examples in the embodiments of the present invention are exemplified as follows. This description is to confirm that specific examples supporting the invention described in the claims are described in the embodiments of the invention. Therefore, even though there are specific examples that are described in the embodiment of the invention but are not described here as corresponding to the configuration requirements, the specific examples are not included in the configuration. It does not mean that it does not correspond to a requirement. On the contrary, even if a specific example is described here as corresponding to a configuration requirement, this means that the specific example does not correspond to a configuration requirement other than the configuration requirement. not.

  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 claim 1 is provided.
In an information processing apparatus (for example, the content recommendation apparatus 31 in FIG. 5) that performs processing on a plurality of contents,
Storage means (for example, user history DB 51 in FIG. 6) for storing user history information representing a time-series viewing history of the content of the user;
Predicting means (for example, the predicting unit 53 in FIG. 6) for predicting temporal changes in the user's preference degree for each content using the user history information stored in the storage means. Features.

The information processing apparatus according to claim 2
Recommendation means for recommending the content having the highest user preference at the present time among the temporal changes in the user preference for each of the plurality of contents predicted by the prediction means (for example, FIG. 6 A recommendation unit 33) is further provided.

The information processing method according to claim 6 comprises:
In an information processing method for processing a plurality of contents,
A storage step (for example, step S1 in FIG. 20) of storing user history information representing a time-series viewing history of the content of the user in the storage unit;
A prediction step (for example, the process of step S2 in FIG. 20) of predicting a temporal change in the user's preference degree for each content using the user history information stored in the storage means. It is characterized by.

  The recording medium program according to claim 7 and the specific example of each step of the program according to claim 8 are the same as the specific example in the embodiment of the invention of each step of the information processing method according to claim 6. is there.

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

  FIG. 5 shows a configuration example of an embodiment of a content recommendation system to which the present invention is applied. In addition, the same code | symbol is attached | subjected about the part corresponding to FIG.

  That is, in the content recommendation system of FIG. 5, the content reproduction apparatus 1 is configured in the same manner as in FIG. 1, and the content reproduction apparatus 1 supplies content information, operation information, and a recommendation request to the content recommendation apparatus 31.

  However, in the content recommendation system of FIG. 1, the content playback apparatus 1 uses, as the operation information, operations such as playback, stop, and fast-forward of music, or “I hate this music” or “I like this music” as described above. However, in the content recommendation system shown in FIG. 5, only the operation information indicating the music playback operation is supplied from the content playback device 1 to the content recommendation device 31. In the embodiment, the content recommendation device 31 uses only the operation information for reproducing the music as the user history information representing the user history. As will be described later with reference to FIG. 21, the content recommendation device 31 uses sensor information as user history information in addition to operation information indicating a music playback operation (in addition to operation information). it can.

  The content recommendation device 31 includes a prediction device (learning device) 32 and a recommendation unit 33.

  The prediction device 32 is supplied with operation information from the operation unit 2 and content information from the reproduction unit 3. The prediction device 32 obtains a prediction curve for predicting a temporal change in the user's preference (degree) for each content based on the operation information and the content information, and supplies the prediction curve to the recommendation unit 33.

  A prediction curve of each content is supplied from the prediction device 32 to the recommendation unit 33. Further, when an operation for requesting content recommendation is performed in the operation unit 2 of the content reproduction apparatus 1, a recommendation request is supplied from the operation unit 2 to the recommendation unit 33.

  The recommendation unit 33 selects the content having the highest user preference (content optimal for the user) at the timing when the recommendation request is supplied from the operation unit 2 among the prediction curves of each content supplied from the prediction device 32. The content is determined and supplied to the playback unit 3 of the content playback apparatus 1 as recommended content.

  FIG. 6 shows a detailed configuration example of the prediction device 32.

  The prediction device 32 includes an acquisition unit 52 having a user history DB 51 and a prediction unit 53.

  Operation information and content information from the content reproduction apparatus 1 are supplied to the acquisition unit 52. The acquisition unit 52 has a timekeeping function, and stores the date and time when operation information indicating that a reproduction operation has been performed is supplied for each content in the user history DB 51.

  Therefore, whenever the user reproduces the content in the content reproduction apparatus 1, information on the reproduction date and time for each content is accumulated in the user history DB 51. That is, the information (user history information) stored in the user history DB 51 is information indicating when each content is reproduced, and is information indicating a time-series viewing history of the user's content.

  The acquisition unit 52 supplies the user history information stored in the user history DB 51 to the prediction unit 53.

  Using the user history information supplied from the acquisition unit 52, the prediction unit 53 obtains a prediction curve that predicts a temporal change in user preference (degree) for each content, and supplies the prediction curve to the recommendation unit 33.

  In the content recommendation system configured as shown in FIG. 5 and FIG. 6, every time a user reproduces content in the content reproduction device 1, the prediction device 32 of the content recommendation device 31 uses the user of the reproduced content. The history information is updated and stored in the user history DB 51.

  Based on the user history information supplied from the acquisition unit 52, the prediction unit 53 of the content recommendation device 31 obtains a prediction curve that predicts a temporal change in the user's preference for each content, and supplies the prediction curve to the recommendation unit 33. To do.

  Then, the recommendation unit 33 supplies the content reproduction device 1 with the content having the highest user preference at the timing when the recommendation request is supplied from the content reproduction device 1 to the content recommendation device 31 as the recommended content. Then, the content reproduction device 1 reproduces the content recommended (supplied) from the content recommendation device 31.

  Next, content recommendation processing in which the content recommendation device 31 recommends content will be described with reference to FIGS. 7 to 9 are diagrams for explaining the concept of the content recommendation process, and therefore the number of samples such as the number of times the content has been reproduced and the number of content is smaller than the actual number.

  In the content reproduction apparatus 1, the three contents (music pieces) of the contents A, B, and C have been reproduced in the past, and the contents A, B, and C are stored as user history information in the user history DB 51 of the prediction apparatus 32. For each, information of the date and time when each was reproduced in the past is stored.

  The user history information stored in the user history DB 51 is illustrated in FIG.

  In FIG. 7, for each content A, B, or C, a star is plotted corresponding to the time when the content is reproduced, with the horizontal axis being time t.

When the user history information shown in FIG. 7 is supplied to the prediction unit 53, the prediction unit 53 uses the user history information to change the user's preference over time for each content as shown in FIG. Prediction curves P _A (t) to P _C (t) are calculated.

In other words, in FIG. 8, the prediction curve P _A (t) for predicting the temporal change in the user's preference level of the content A and the temporal change in the user's preference level of the content B obtained by the prediction unit 53 are shown. _A prediction curve P _A (t) to be predicted and a prediction curve P _C (t) to predict a temporal change in the preference level of the user of the content C are shown. In addition, when it is not necessary to particularly distinguish each of the prediction curves P _A (t), P _B (t), or P _C (t), they are simply referred to as a prediction curve P (t).

  In FIG. 8, the horizontal axis of each content represents time t as in FIG. 7, and the vertical axis represents the user's preference level (preference strength) for the content. The user's preference for content is set in the range of +1 to -1, for example.

  Further, a time t ′ indicated by a dotted line in FIG. 8 indicates the current time. The prediction curve P (t) generates a prediction curve P (t) that matches the change in the user's preference for the past (time shown on the left side of the time t ′ in the figure) (the timing at which the content is played back). The change in the preference, that is, the prediction curve P (t) is predicted to continue in the future (the time shown on the right side of the time t ′ in the figure), thereby predicting the change in the preference of the user in the future. Predicted with curve P (t).

  The prediction unit 53 supplies the prediction curve P (t) of each content thus obtained to the recommendation unit 33. When the content playback apparatus 1 newly plays back the content A, B, or C and the operation information and the content information are supplied to the prediction device 32, the prediction device 32 updates the user history information. Is used to recalculate the prediction curve P (t) and supply it to the recommendation unit 33 Accordingly, the prediction curve P (t) is updated as needed and supplied to the recommendation unit 33.

  As shown in FIG. 8, the user preference of each content represented by the prediction curve P (t) changes with time, and the content with the highest preference among the contents A, B, and C also changes with time. is doing.

After the prediction curves P (t) of the contents A, B, and C shown in FIG. 8 are supplied to the recommendation unit 33, the content recommendation is performed from the operation unit 2 of the content reproduction apparatus 1 at time t ″ shown in FIG. It is assumed that a recommendation request is supplied to the recommendation unit 33 of the device 31. At time t ″, the preference levels P _A (t ″), P _B (t ″), and P _C ( Among t ″), P _B (t ″) is the largest, P _A (t ″) is the second largest, and P _C (t ″) is the third largest (smallest). (Ie, P _B (t ″)> P _A (t ″)> P _C (t ″)).

  Therefore, the recommendation unit 33 supplies the content B, which is the content with the highest user preference at the present time, to the reproduction unit 3 as the recommended content in response to the recommendation request from the operation unit 2. Note that the recommendation unit 33 does not supply only one piece of content with the highest user preference to the playback unit 3 so that the user preference level is known (for example, the user preference is high). All the content may be supplied in order).

  As described above, in response to the recommendation request from the content reproduction device 1, the content recommendation device 31 determines the user's preference level of each content from time-series information (user history information) that the user has reproduced the content in the past. And the content having the highest user preference level at the present time is supplied to the content reproduction apparatus 1 as the content that matches the user preference (optimum for the user).

  By the way, in the content recommendation processing of the content recommendation device 31 described above, it is important how the prediction unit 53 of the prediction device 32 generates the prediction curve P (t).

  Therefore, next, two generation methods in which the prediction unit 53 generates a prediction curve P (t) for predicting a temporal change in the user's preference for each content will be described. The first generation method is a method of generating a prediction curve P (t) based on a predetermined rule determined according to the user's appreciation history pattern, and the second generation method is based on the user's appreciation history. This is a method of generating a prediction curve from the characteristic waveform obtained by analyzing the frequency of the generated waveform to obtain a characteristic waveform representing the characteristic of appreciation for the content. The first and second generation methods are examples of a generation method for generating the prediction curve P (t), and the prediction curve may be generated by a method other than the first and second generation methods.

  First, the first generation method for generating the prediction curve P (t) will be described.

The prediction unit 53 classifies each content to be recommended from the user history information supplied from the acquisition unit 52 into, for example, the following five patterns (1) to (5), and according to the classified patterns: The prediction curves E ₁ (t) to E ₅ (t) are applied respectively.

(1) Content that has recently been newly renewed and added to the recommendation target (2) Content that has recently been replayed frequently (3) Content that has been replayed at a certain period (4) Opportunities for replaying gradually (5) Content that has been played frequently before but has not been played recently

In other words, the prediction unit 53 follows the rule that the prediction curves E ₁ (t) to E ₅ (t) are applied to the five patterns (1) to (5), respectively. ₁ (t) to E ₅ (t) are generated. It is assumed that the user's preference degree for each content at the present time (current) of each of the prediction curves E ₁ (t) to E ₅ (t) is 0 (neutral state that they do not like or dislike).

It can be considered that the user's preference (interest) is strong with respect to the content classified into the pattern (1), that is, the content newly reproduced recently. In addition, the user is likely to be addicted to this content in the future. Therefore, the prediction unit 53 applies the prediction curve E ₁ (t) shown in FIG. 10 to the content newly reproduced and added to the recommendation target (content classified into the pattern (1)). To do.

As shown in FIG. 10, the prediction curve E ₁ (t) is a curve in which the user's preference for content is the greatest as it is closer to the current time, and the preference is attenuated to 0 as time passes.

For content classified into the pattern (2), that is, content that has been played frequently recently, it is expected that the more often it is played, the later it will get bored temporarily. It is conceivable that the user's preference level gradually decreases. Therefore, the prediction unit 53 applies the prediction curve E ₂ (t) shown in FIG. 11 to content that has been frequently played back recently (content classified as pattern (2)).

As shown in FIG. 11, the prediction curve E ₂ (t) indicates that the user's preference for the content gradually decreases from 0 as time passes, and the user's preference for the content increases after a certain period of time. It is also a curve that returns to 0 (rises).

For content classified as pattern (3), that is, content that is played back at a certain cycle, it is expected that the user will be interested in the same cycle in the future. It is considered that the user's preference changes with the period. Therefore, the prediction unit 53 applies the prediction curve E ₃ (t) shown in FIG. 12 to the content (content classified into the pattern (3)) reproduced at a certain period.

As shown in FIG. 12, the prediction curve E ₃ (t) is a curve in which the preference degree changes in the same cycle as the cycle T of the user's preference change with respect to past contents.

For content classified as pattern (4), that is, content whose playback opportunities are gradually decreasing, it is expected that the user is gradually getting tired of this content. It seems that the user's interest in this content will gradually decline. Therefore, the prediction unit 53 applies the prediction curve E ₄ (t) shown in FIG. 13 to the content (content classified into the pattern (4)) whose playback opportunities are gradually decreasing.

As shown in FIG. 13, the prediction curve E ₄ (t) is a curve (straight line) in which the user's preference for content gradually decreases from zero.

For content classified as pattern (5), that is, content that has been frequently played back before and has not been played back recently, it is expected that the user will want to play this content again. Is done. Therefore, the prediction unit 53 is a prediction curve E shown in FIG. 14 for content that has been frequently played back before and that has not been played back recently (content classified as pattern (5)). ₅ Apply (t).

As shown in FIG. 14, the prediction curve E ₅ (t) shows that the user's preference for content gradually increases from 0 as time passes, and the user's preference for content increases after a certain period of time. The curve returns to zero.

As described above, according to the first generation method, the prediction unit 53 classifies each content to be recommended into the patterns (1) to (5) described above, and is determined according to the pattern. The prediction curves E ₁ (t) to E ₅ (t) can be used as the prediction curves P (t) representing changes in the user's temporal preference of the content.

When the content corresponds to a plurality of patterns among the patterns (1) to (5), a curve obtained by synthesizing the prediction curves E ₁ (t) to E ₅ (t) of the corresponding patterns is obtained. The prediction curve P (t) of the content is used. For example, when a certain content is recently renewed and replayed more frequently, the content corresponds to both patterns (1) and (2). In this case, a curve obtained by synthesizing the prediction curve E ₁ (t) and the prediction curve E ₂ (t) is a prediction curve P (t) (= E ₁ (t) + E ₂ (t)) of the content.

  Next, a second generation method for generating the prediction curve P (t) will be described.

  First, the prediction unit 53 generates a waveform (hereinafter, appropriately referred to as a history waveform) that matches a change in the user's preference for past content from the user history information supplied from the acquisition unit 52.

  In order to obtain a history waveform, a user's past appreciation history is analyzed and created by various rules, or a waveform is created from a change in user's preference obtained by the conventional method as described in FIG. May be.

  In the present embodiment, in order to simplify the description, for example, the prediction unit 53 says that when the user explicitly reproduces the content, the user's preference for the content is strong (the preference degree is large). Based on the rule, a waveform (history waveform) representing a change in the user's past preference is generated.

  For example, a certain content has an appreciation history as shown in FIG. 15 in the past. That is, the user is viewing the content by explicit playback a total of 10 times in a predetermined period until now.

  For a user's appreciation history shown in FIG. 15 for a certain content, the prediction unit 53 sets the user's preference at the time of reproduction (date and time) to 1 and the user's preference at other times as shown in FIG. A waveform with a degree of 0 is generated.

  Next, the prediction unit 53 cuts the high frequency component of the waveform shown in FIG. 16, generates a waveform from which the DC component is removed, and uses it as a history waveform.

  FIG. 17 shows a history waveform obtained from the waveform shown in FIG.

  The prediction unit 53 performs orthogonal transform (frequency analysis) on the history waveform shown in FIG. 17 and can continuously connect to the history waveform from the past to the present among the waveforms having the frequency characteristics (fluctuation) obtained thereby. A phase waveform is selected (generated) and set as a prediction curve P (t).

  That is, FIG. 18 shows the frequency characteristics (frequency analysis result) of the history waveform obtained by orthogonal transformation of the history waveform shown in FIG. 17, where the horizontal axis represents the degree of preference change and the vertical axis represents the preference change. Represents size.

  Further, FIG. 19 has a frequency characteristic shown in FIG. 18 and selects a waveform having a phase that can be continuously connected to a history waveform from the past to the present, and a prediction curve P (t ).

  As described above, according to the second generation method, when the prediction unit 53 generates a history waveform from the user history information and connects successive waveforms after the history waveform, the same frequency characteristics (fluctuation) as before. A prediction curve P (t) such that

  Next, content recommendation processing of the content recommendation device 31 in FIG. 5 will be described with reference to the flowchart in FIG.

  When a user operates the operation unit 2 and a predetermined content is reproduced by the content reproduction device 1, operation information indicating that a reproduction operation has been performed is supplied from the operation unit 2 to the acquisition unit 52 of the prediction device 32. In addition, content information of the reproduced content is supplied from the reproduction unit 3 to the acquisition unit 52.

  When the operation information or the content information is supplied from the operation unit 2 or the playback unit 3 to the acquisition unit 52, the acquisition unit 52 stores the operation information and content information in the user history DB 51 in step S1. As a result, the user history information stored in the user history DB 51 is updated. In step S1, the acquisition unit 52 supplies the user history information stored in the user history DB 51 to the prediction unit 53, and proceeds to step S2.

  In step S2, the prediction unit 53 uses the user history information supplied from the acquisition unit 52 to obtain a prediction curve P (t) for predicting a temporal change in the user's preference level for each content. The process proceeds to step S3. Here, the prediction curve P (t) is generated by the first or second generation method described above, for example.

  In step S3, the recommendation unit 33 determines whether or not a recommendation request has been received from the content reproduction device 1 (the operation unit 2 thereof).

  In step S3, the process is repeated until it is determined that there is a recommendation request from the content reproduction apparatus 1. When it is determined that there is a recommendation request, the process proceeds to step S4, and the recommendation unit 33 is supplied from the prediction unit 53. Among the prediction curves P (t) of each content, the content having the highest user preference at the present time is determined, and the content is supplied to the playback unit 3 of the content playback device 1 as the recommended content, and the process ends. .

  Through the content recommendation process described above, the recommended content is reproduced (presented to the user) in the content reproduction apparatus 1 to which the recommended content has been supplied.

  FIG. 21 shows a configuration example of another embodiment of the content recommendation device 31. Note that portions corresponding to those in FIG. 6 are denoted by the same reference numerals, and description thereof is omitted.

  That is, in the content recommendation device 31 of FIG. 21, a prediction device 71 is provided instead of the prediction device 32 (FIG. 6). The prediction device 71 is configured in the same manner as the prediction device 32 of FIG. 6 except that a sensor 81 is newly provided.

  The sensor 81 includes, for example, various sensors that acquire various types of information such as video and audio information, user position information, environmental information such as temperature, humidity, and smell, and biological information such as heart rate and blood pressure. The acquired information is supplied to the acquisition unit 52 as sensor information.

  In addition to the information on the date and time when the content is played back, the acquisition unit 52 receives the sensor information supplied from the sensor 81 when the operation information and the content information are supplied, as well as the user history DB 51. Remember me.

  As a result, in the user history information, information about the state (feeling, facial expression, etc.) of the user at that time is added to the date and time information of the reproduced content, and the information is supplied from the acquisition unit 52 to the prediction unit 53. The The prediction unit 53 generates a prediction curve P (t) that predicts a temporal change in the user's preference for each content after analyzing the user's state at that time together with the date and time when the content was played back. It becomes possible to do.

  By the way, in the above-described example, the prediction device 32 obtains the prediction curve P (t) for predicting the temporal change of the user's preference degree for each content. It is possible to classify and predict temporal changes in the user's preference degree for each feature amount of the classified content.

  For example, with respect to the feature amount of tempo content, the content to be recommended can be classified into three feature amounts: a slow tempo song, a normal tempo song, or a fast tempo song.

Then, as shown in FIG. 22, the prediction device 32 plots the time when the content is reproduced for each of a slow tempo music, a normal tempo music, or a fast tempo music. Then, as in the case of obtaining the prediction curve P (t) for each content in FIGS. 7 to 9, prediction curves P _D1 (t) and P _D2 representing changes in the user's preference for the plotted content. (T), and P _D3 (t) are generated.

After that, when a recommendation request is supplied from the operation unit 2 of the content reproduction device 1 to the recommendation unit 33 of the content recommendation device 31 at time t ″, the music with slow tempo and normal tempo at time t ″ , and a fast tempo of each user preference P _D1 (t ") of the song, P _D2 (t '), and _{P D3 (t") (P} D2 t) preference is the largest among the "are The tempo of the recommended music is supplied from the recommendation unit 33 to the playback unit 3.

  It is also possible to predict a temporal change in the user's preference degree for each field that covers a wider range than the content unit. Here, the field including contents includes, for example, content categories such as work, travel, and economy.

  Furthermore, a prediction curve P (t) that predicts temporal changes in user preference is obtained for each content, feature quantity, content category, other metadata, etc. Based on the curve P ′ (t), the content having the highest degree of preference at the present time may be determined as the recommended content.

For example, if the content is a rock song, the prediction curve P ₁ (t) of the content itself, a prediction curve P ₂ (t) for “rock” for each genre that is more detailed than the content, and “music” that is wider than the content Whether or not to recommend rock music based on the prediction curve P ′ (t) obtained by combining the respective prediction curves P ₃ (t) for “” and which content is recommended content) Is determined.

Also, for example, if the content is a TV program for hot spring travel, the prediction curve P ₁ (t) for the TV program, the prediction curve P ₂ (t) for travel, and the hot spring, mountain, sea, cherry blossom viewing, etc. A prediction curve P ₃ (t) for a hot spring in a detailed genre is obtained, and based on the prediction curve P ′ (t) obtained by combining them, whether to recommend a hot spring trip TV program (which content Whether or not to be recommended content).

  Note that the resolution of the time axis when obtaining the above-described prediction curve P (t) can be appropriately changed depending on the type of content for which a change in the user's preference level is predicted. For example, in the recommendation of content for creating a playlist of music, the prediction curve P (t) is obtained with a time resolution in units of minutes, but content including a plurality of genres such as music-movie-game is included. In the recommendation, the prediction curve P (t) can be obtained with a time resolution in units of months.

  Moreover, the user's preference degree of each content calculated | required by the content recommendation apparatus 31 and the user's preference degree of each content calculated | required by the other preference learning apparatus can also be combined.

For example, at the time t ″ described above with reference to FIG. 9, when the recommendation request is supplied from the content reproduction apparatus 1 to the content recommendation apparatus 31, the preference levels P _A (t ″) and P _B ( t ″) and P _C (t ″) are 0.5, 0.7, and 0.2, respectively, and the preference levels Q _A (t of contents A, B, and C obtained by other preference learning devices ”), Q _B (t ″), and Q _C (t ″) are 0.5, 0.6, and 0.6, respectively, the final preference of each content is as follows: Can be sought.

Content A preference = P _A (t ″) + Q _A (t ″) = 0.5 + 0.5 = 1.0
Content B preference = P _B (t ″) + Q _B (t ″) = 0.7 + 0.6 = 1.3
Content C preference = P _C (t ″) + Q _C (t ″) = 0.2 + 0.6 = 0.8

  That is, the final preference degree of each content can be obtained by the sum of the preference degree obtained by the content recommendation device 31 and the preference degree obtained by another preference learning device.

  Moreover, you may obtain | require the final preference of each content as follows.

Content A preference = P _A (t ″) × Q _A (t ″) = 0.5 × 0.5 = 0.25
Content B preference = P _B (t ″) × Q _B (t ″) = 0.7 × 0.6 = 0.42
Content C preference = P _C (t ″) × Q _C (t ″) = 0.2 × 0.6 = 0.12.

  That is, the final preference degree of each content may be obtained as a product of the preference degree obtained by the content recommendation device 31 and the preference degree obtained by another preference learning device.

  As described above, according to the content recommendation device 31 of FIG. 5, it is possible to recommend content in consideration of a temporal change (in time series) of the user's preference, such as getting into the content or getting bored.

  In addition, if not all of the recommended content is recommended but only all the recommended content is ordered and supplied (recommended) to the content reproduction apparatus 1, time-series preferences are important. It can be applied to creating playlists.

  Furthermore, there is no problem of falling into a local solution as occurs in a conventional content recommendation device.

  In addition, the information necessary for the content recommendation device 31 to predict the temporal change of the user's preference is only the operation information indicating the reproduction operation and the content information. It is not necessary to perform operations such as "." Further, it is not necessary to perform the “dislike” operation for correcting the user's preference degree of the preference learning device 12 as described above in the content recommendation device 11 of FIG.

  Furthermore, the content recommendation device 31 of FIG. 5 can be used in combination with other conventional content recommendation devices. That is, as described above, the user preference level of each content obtained by the content recommendation device 31 is combined with the user preference level of each content obtained by another preference learning device, and the result is finally obtained. User preference.

  A series of processes such as the content recommendation process described above can be executed by dedicated hardware or can be executed by software. When a series of processing is performed by software, for example, the series of processing can be realized by causing a (personal) computer as shown in FIG. 23 to execute a program.

  23, a CPU (Central Processing Unit) 101 executes various processes according to a program stored in a ROM (Read Only Memory) 102 or a program loaded from a storage unit 108 to a RAM (Random Access Memory) 103. To do. The RAM 103 also appropriately stores data necessary for the CPU 101 to execute various processes.

  The CPU 101, ROM 102, and RAM 103 are connected to each other via a bus 104. An input / output interface 105 is also connected to the bus 104.

  The input / output interface 105 includes an input unit 106 including a keyboard and a mouse, a display including a CRT (Cathode Ray Tube) and an LCD (Liquid Crystal display), an output unit 107 including a speaker, and a hard disk. A communication unit 109 including a storage unit 108, a modem, a terminal adapter, and the like is connected. The communication unit 109 performs communication processing via a network such as the Internet.

  A drive 110 is connected to the input / output interface 105 as necessary, and a magnetic disk 121, an optical disk 122, a magneto-optical disk 123, a semiconductor memory 124, or the like is appropriately mounted, and a computer program read from them is It is installed in the storage unit 108 as necessary.

  In the present specification, the steps described in the flowcharts are executed in parallel or individually even if they are not necessarily processed in time series, as well as processes performed in time series in the described order. It also includes processing.

  Further, in this specification, the system represents the entire apparatus constituted by a plurality of apparatuses.

It is a block diagram which shows an example of a structure of the conventional content recommendation system. It is a figure explaining learning of the preference learning apparatus. It is a figure explaining learning of the preference learning apparatus. It is a figure explaining learning of the preference learning apparatus. It is a block diagram which shows the structural example of one Embodiment of the content recommendation system to which this invention is applied. 3 is a block diagram illustrating a detailed configuration example of a prediction device 32. FIG. It is a figure explaining a content recommendation process. It is a figure explaining a content recommendation process. It is a figure explaining a content recommendation process. It is a figure explaining the 1st production | generation method which produces | generates the prediction curve P (t). It is a figure explaining the 1st production | generation method which produces | generates the prediction curve P (t). It is a figure explaining the 1st production | generation method which produces | generates the prediction curve P (t). It is a figure explaining the 1st production | generation method which produces | generates the prediction curve P (t). It is a figure explaining the 1st production | generation method which produces | generates the prediction curve P (t). It is a figure explaining the 2nd production | generation method which produces | generates the prediction curve P (t). It is a figure explaining the 2nd production | generation method which produces | generates the prediction curve P (t). It is a figure explaining the 2nd production | generation method which produces | generates the prediction curve P (t). It is a figure explaining the 2nd production | generation method which produces | generates the prediction curve P (t). It is a figure explaining the 2nd production | generation method which produces | generates the prediction curve P (t). It is a flowchart explaining a content recommendation process. It is a block diagram which shows the structural example of other embodiment of the content recommendation apparatus 31. FIG. It is a figure explaining the other example of the prediction curve P (t). It is a block diagram which shows the structural example of one Embodiment of the computer to which this invention is applied.

Explanation of symbols

  31 content recommendation device, 32 prediction device (learning device), 33 recommendation unit, 51 user history DB, 52 acquisition unit, 53 prediction unit

Claims (8)

In an information processing apparatus that processes a plurality of contents,
Storage means for storing user history information representing a time-series viewing history of the content of the user;
An information processing apparatus comprising: prediction means for predicting a temporal change in the user's preference degree for each content using the user history information stored in the storage means. It further comprises recommendation means for recommending the content having the highest user preference at the present time among temporal changes in the user preference for each of the plurality of contents predicted by the prediction means. The information processing apparatus according to claim 1. The prediction means generates a prediction curve of the user's preference degree for the content based on a predetermined rule determined according to a pattern of the user's viewing history, so that the user's preference for each content. The information processing apparatus according to claim 1, wherein a temporal change in the degree is predicted. The prediction means performs frequency analysis on a waveform generated from the viewing history of the user, thereby obtaining a characteristic waveform representing a viewing characteristic for the content, and generates a prediction curve for a preference degree for the content from the characteristic waveform. The information processing apparatus according to claim 1, wherein a temporal change in the user preference is predicted for each content. The information processing apparatus according to claim 1, wherein the prediction unit predicts a temporal change in the user's preference degree for each feature amount of the content. In an information processing method for processing a plurality of contents,
A storage step of storing in the storage means user history information representing a time-series viewing history of the user's content;
An information processing method comprising: a prediction step of predicting a temporal change in the user's preference level for each content using the user history information stored in the storage unit. A program for processing a plurality of contents,
A storage step of storing in the storage means user history information representing a time-series viewing history of the user's content;
A computer-readable program comprising: a prediction step of predicting temporal changes in the user's preference level for each content using the user history information stored in the storage means. Recorded recording medium. In a program that processes multiple contents,
A storage step of storing in the storage means user history information representing a time-series viewing history of the user's content;
A program for causing a computer to execute a process including a prediction step of predicting a temporal change in the user's preference degree for each content by using the user history information stored in the storage unit. .

Images