Music recommendation model generation system and music recommendation system

The song recommendation model uses machine learning to recommend songs based on population and song usage data, addressing the challenge of user identification and historical data absence in karaoke systems, ensuring appropriate song suggestions.

JP7581200B2Active Publication Date: 2024-11-12NTT DOCOMO INC
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
JP2021530558
Authority / Receiving Office
JP · JP
Patent Type
Patents
Current Assignee / Owner
Priority Date
2019-07-05
Filing Date
2020-06-17
Publication Date
2024-11-12
Estimated Expiration
2040-06-17

AI Technical Summary

Technical Problem

Existing song recommendation systems in karaoke parlors struggle to identify individual users and recommend songs based on their singing history, especially when the user is not singing, as user identification is often not possible, and historical data is not available.

Method used

A song recommendation model generation system using machine learning that utilizes population information and song usage data from known karaoke locations to make recommendations, even without prior singing history, by incorporating learning data on user attributes, song information, and location-specific population data.

Benefits of technology

Enables appropriate song recommendations even when users have no singing history, leveraging location-specific population data and song usage patterns to provide tailored suggestions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 0007581200000001
    Figure 0007581200000001
  • Figure 0007581200000002
    Figure 0007581200000002
  • Figure 0007581200000003
    Figure 0007581200000003
Patent Text Reader

Abstract

The present invention makes a recommendation pertaining to a musical composition in a suitable manner. A model generation system 10 included in a recommendation server 1 is a musical composition recommendation model generation system that generates a musical composition recommendation model used in a recommendation pertaining to a musical composition, the model generation system 10 comprising: a learning data acquisition unit 11 that acquires for-learning population information indicating a population, for each attribute of people, in a location where a musical composition is used, and for-learning usage musical composition information pertaining to the musical composition, these items of information being learning data used in machine learning; and a model generation unit 12 that performs machine learning and generates a musical composition recommendation model, the model generation unit 12 using information based on the for-learning population information as an input to the musical composition recommendation model, and using information based on the for-learning usage musical composition information as an output of the musical composition recommendation model.
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] The present invention relates to a model generation system for music recommendation that generates a model for music recommendation used for recommending music, and a music recommendation system that determines information to be recommended using a model for music recommendation. [Background technology]

[0002] 2. Description of the Related Art Conventionally, it has been proposed to recommend songs to a user based on the user's singing history in karaoke (see, for example, Patent Document 1). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2012-78387 A Summary of the Invention [Problem to be solved by the invention]

[0004] In the above method, the singing history of the user to be recommended is necessary to make a recommendation. However, in a karaoke store, it is usually not possible to identify the user, and it is not possible to refer to the singing history of the user when visiting the store in the past. Therefore, it is not possible to recommend a song based on the singing history of the user when visiting the store in the past. In addition, if it is possible to refer to the singing history at the time of the visit to be recommended, it is not possible to recommend a song when the user has not sung and there is no singing history. In other words, it is not possible to recommend the first song. In addition, it is desirable to recommend a song more appropriately even when the user has sung several songs and there is a singing history.

[0005] An embodiment of the present invention has been made in consideration of the above, and aims to provide a model generation system for music recommendation and a music recommendation system that can appropriately recommend music. [Means for solving the problem]

[0006] In order to achieve the above object, a model generation system for music recommendation according to an embodiment of the present invention is a model generation system for music recommendation that generates a model for music recommendation to be used for music recommendation, and includes a learning data acquisition unit that acquires learning population information indicating the population for each person attribute in a place where a music is used and learning-use music information related to the music, which are learning data to be used for machine learning, and a model generation unit that performs machine learning to generate a model for music recommendation using information based on the learning population information acquired by the learning data acquisition unit as an input to the model for music recommendation and information based on the learning-use music information acquired by the learning data acquisition unit as an output of the model for music recommendation. The learning population information is information on locations where karaoke stores are known to exist. .

[0007] A music recommendation system according to an embodiment of the present invention is a music recommendation system that determines information to be recommended using a music recommendation model generated by a music recommendation model generation system, and includes a model usage data acquisition unit that acquires recommendation population information indicating the population of each person attribute in a place where the music is used, and a model usage unit that inputs information based on the recommendation population information acquired by the model usage data acquisition unit into the music recommendation model to determine information to be recommended. The recommendation population information is information on a location where it is known in advance that there is a karaoke store. .

[0008] According to one embodiment of the present invention, a song recommendation can be made based on a song recommendation model generated by machine learning, using population information of a place where the song is used. Even if a user to whom a recommendation is made does not sing and has no singing history, an appropriate recommendation can be made, and even if a user has a singing history, an appropriate recommendation can be made according to the population of each person's attribute in the place. Therefore, according to one embodiment of the present invention, a song recommendation can be made appropriately. Effect of the Invention

[0009] According to one embodiment of the present invention, it is possible to appropriately recommend music. [Brief description of the drawings]

[0010] [Figure 1] 1 is a diagram showing a configuration of a recommendation server which is a model generation system for music recommendation and a music recommendation system according to an embodiment of the present invention. [Diagram 2] 11 is a table showing an example of information related to music and the like stored in a recommendation server. [Diagram 3] 13 is a table showing examples of feature amounts converted from a singer ID. [Figure 4] 11 is a table showing an example of the importance of words contained in lyrics of a song. [Diagram 5] FIG. 11 is a diagram showing an example of a spectrogram converted from audio data of a piece of music. [Figure 6] 11 is a table showing an example of time-series information indicating songs sung in the past. [Figure 7] 1 is a table showing an example of learning data for generating a model for music recommendation. [Figure 8] FIG. 11 is a diagram illustrating an example of a correspondence relationship of learning data for generating a model for music recommendation. [Figure 9] FIG. 13 is a diagram illustrating an example of learning data for generating a type estimation model. [Figure 10]FIG. 13 is a diagram illustrating an example of clustering performed to generate a user type. [Figure 11] 11 is a table showing an example of identification numbers associated with each session. [Figure 12] FIG. 13 is a diagram illustrating an example of a correspondence relationship of learning data for generating a type estimation model. [Figure 13] FIG. 10 is a diagram illustrating a schematic diagram of how a music recommendation model and a type estimation model are used. [Figure 14] 1 is a flowchart showing processing executed when generating a music recommendation model and a type estimation model in a music recommendation model generation system and a recommendation server which is the music recommendation system according to an embodiment of the present invention. [Figure 15] 1 is a flowchart showing a process executed when a recommendation server, which is a model generation system for music recommendation and a music recommendation system according to an embodiment of the present invention, recommends music to a user. [Figure 16] 1 is a diagram showing a hardware configuration of a recommendation server which is a model generation system for music recommendation and a music recommendation system according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0011] Hereinafter, embodiments of a music recommendation model generation system and a music recommendation system according to the present invention will be described in detail with reference to the drawings. In the description of the drawings, the same elements are given the same reference numerals and duplicated descriptions will be omitted.

[0012] FIG. 1 shows a recommendation server 1, which is a model generation system for music recommendation and a music recommendation system according to this embodiment. The recommendation server 1 is a system that recommends information about music to a user. Specifically, the recommendation server 1 recommends a song or a singer to be sung when the user sings karaoke. The recommendation server 1 makes a recommendation regarding the first song to be sung. In addition, the recommendation server 1 makes a recommendation regarding the next song to be sung after a song has been sung. The recommendation server 1 may make a recommendation regarding the next song to be sung every time a song is sung.

[0013] In this embodiment, the users to be recommended may include a group of multiple users who sing karaoke at the same time, for example, a group of users who take turns singing. In the following description, even when the word "user" is used simply, it includes a group of multiple users. Note that the recommendation server 1 may recommend information about music used by users for purposes other than karaoke. For example, the recommendation server 1 may recommend music that users listen to.

[0014] The recommendation server 1 performs machine learning to generate a model for music recommendation (trained model), and uses the generated model for music recommendation to make a recommendation. The recommendation server 1 also performs type estimation to determine which of a plurality of types the user belongs to, and uses the result of type estimation to recommend to the user. The user type is a classification according to the music sung by the user. The recommendation server 1 performs machine learning to generate a model for type estimation (trained model), and uses the generated model for type estimation to make a type estimation. The recommendation server 1 includes a model generation system 10 as a configuration for generating a model for music recommendation and a model for type estimation, and a model utilization system 20 as a configuration for recommendation and type estimation. The model generation system 10 is a model generation system for music recommendation according to this embodiment. The model utilization system 20 is a music recommendation system according to this embodiment.

[0015] The recommendation server 1 is configured by a server device having a communication function. The recommendation server 1 may be realized by a plurality of server devices, i.e., a computer system. The recommendation server 1 can transmit and receive information between the terminal 30 and the data server 40 via a network such as the Internet or a dedicated line. When transmitting and receiving information, the information may be relayed by a relay server provided between those devices.

[0016] The terminal 30 is a terminal used by a user in a store such as a karaoke store. The terminal 30 is used, for example, in a store such as a karaoke store to search for and input songs. The terminal 30 is, for example, an information terminal provided in the karaoke store. Alternatively, the terminal 30 may be an information terminal such as a smartphone owned by an individual user. In this case, an application for a karaoke store that realizes the function of the terminal 30 is installed in the information terminal and used. The terminal 30 makes a request for recommended content to the recommendation server 1, receives information related to the recommendation from the recommendation server 1, and outputs, for example, display. The user can refer to the output to decide which song to sing.

[0017] The data server 40 is a device that stores data used for recommendations by the recommendation server 1. Specific information stored therein will be described later. A plurality of data servers 40 may be provided for each type of data.

[0018] Next, a description will be given of the functions of the model generation system 10 and the model utilization system 20 included in the recommendation server 1 according to this embodiment. As shown in FIG. 1, the model generation system 10 includes a learning data acquisition unit 11 and a model generation unit 12.

[0019] The learning data acquisition unit 11 is a functional unit that acquires learning data used in machine learning to create a model for recommending music. The learning data acquisition unit 11 acquires, as learning data, learning population information indicating the population for each person attribute (characteristic) in a place where a music piece is used, and learning-use music information related to the music piece. The learning data acquisition unit acquires learning population information indicating the person attributes and the population for each time in a place where a music piece is used. The person attributes include at least one of the person's occupation and preferences.

[0020] Further, the learning data acquisition unit 11 acquires, as the learning data, learning time series information, which is time series information regarding the multiple songs used, and learning use song information regarding the songs used after the multiple songs. The learning data acquisition unit 11 acquires learning time series information including information regarding the singer of the used song. The learning data acquisition unit 11 acquires learning time series information including information regarding the genre of the used song. The learning data acquisition unit 11 acquires learning time series information including information regarding words included in the lyrics of the used song. The learning data acquisition unit 11 acquires learning time series information including information regarding the audio data of the used song. The learning data acquisition unit 11 acquires information that is time series information regarding the multiple songs used and includes information indicating the time when the song was used, and generates learning time series information and learning use song information based on the time. The learning data acquisition unit 11 acquires learning type information indicating the type to which the user who used the multiple songs belongs. Specifically, the learning data acquisition unit 11 acquires the learning data as follows.

[0021] The learning data acquisition unit 11 stores information about songs in advance. The songs are songs that can be provided by karaoke, that is, songs that can be used by users. The learning data acquisition unit 11 stores a song table shown in FIG. 2(a). The song table stores song IDs, singer names, genre names, etc. in association with each other. The song ID is information (identifier) ​​that identifies each song that is set in advance. The singer name is the name of the singer who sings the song. The genre name is information that indicates the genre of the song. As shown in FIG. 2(a), genre names include "J-POP" and "ballad". The information that is stored in advance in the recommendation server 1 is acquired from the data server 40, etc., unless otherwise described.

[0022] The learning data acquisition unit 11 stores a singer ID table shown in FIG. 2(b). The song table stores singer names and singer IDs in association with each other. The singer ID is information (identifier) ​​that identifies each singer that has been set in advance. The singer ID is a numerical value, for example, an integer as shown in FIG. 2(b), for use in machine learning. The learning data acquisition unit 11 converts the singer names in the song table shown in FIG. 2(a) into singer IDs based on the associations shown in the singer ID table shown in FIG. 2(b).

[0023] The learning data acquisition unit 11 also stores an ID table for genre names in the same way as for singer names, and converts the genre names in the song table into genre IDs based on the ID table. Like singer IDs, genre IDs are also made into numerical values, for example integers, for use in machine learning. In this way, the learning data acquisition unit 11 generates a song table with IDs (converted into IDs) as shown in FIG. 2(c). Note that the ID table may not be acquired from the data server 40, but may be generated by the recommendation server 1.

[0024] As will be described later, singer IDs are converted into features in the model for song recommendation. An example of a feature is shown in FIG. 3(a). A feature is a numerical value with a preset number of dimensions. In the example shown in FIG. 3(a), it is a two-dimensional feature. The number of dimensions of a feature is usually a numerical value with several dimensions to several tens of dimensions. The feature is updated during machine learning and is replaced with a better numerical value. The genre ID is also converted into a feature in the same way as the singer ID. FIG. 3(b) shows a song table that has been converted into features (converted into features).

[0025] The learning data acquisition unit 11 acquires information indicating the lyrics of each song from the data server 40. The learning data acquisition unit 11 performs morphological analysis on the lyrics and divides them into words. For each song, the learning data acquisition unit 11 calculates the importance of each word in consideration of the frequency of occurrence of each word in the lyrics. The words for which the importance is calculated may be set in advance. For example, the importance of words that are verbs, adjectives, and adjectival verbs may be calculated. The learning data acquisition unit 11 calculates the importance of a song (lyrics) d by the following formula: j The word t in i Importance of tfidf(t i ,d j ) is calculated. idf(t i )=log(total number of songs / words t i (Number of songs in which t appears) = log(N / df(t i )) tf(t i ,d j )=Song d j The word in i Number of occurrences / song d j The sum of the occurrences of all words in = f(t i ,d j ) / Σ tk∈dj f(t k ,d j ) tfidf(t i ,d j )=tf(t i ,d j )·idf(t i ) The importance of words for a certain song is shown in Fig. 4. Note that, as information based on the lyrics of a song, any information that has been made into a feature value other than the importance of words may be used.

[0026] The learning data acquisition unit 11 acquires the audio data of each piece of music from the data server 40. The audio data is, for example, data (sound source) used to play karaoke songs, and is audio waveform data that is a numerical value of signal strength for each time. MP3 (MPEG-1 Audio Layer-3) may be used as the audio data. The learning data acquisition unit 11 converts the audio data of each piece of music into a spectrogram (mel spectrogram). The learning data acquisition unit 11 converts audio data of a time period that is commonly set in advance for each piece of music into a spectrogram. For example, audio data for one minute from the beginning of a piece of music is converted into a spectrogram. This is because pieces of music with different time lengths are treated as data of the same size. The conversion of audio data into a spectrogram can be performed by a conventional method. The converted spectrogram becomes data of numerical values ​​(e.g., dB) of signal strength for each time (e.g., seconds elapsed from the beginning of the piece of music) and frequency (e.g., Hz). An example of a spectrogram for a piece of music is shown in FIG. 5. As shown in Fig. 5, the spectrogram can be treated as an image. Any information that is characterized other than the spectrogram may be used as the information based on the audio data. Furthermore, the above-mentioned waveform data itself may be used as the information based on the audio data.

[0027] The learning data acquisition unit 11 stores information indicating the population for each location and person attribute in advance. For example, the learning data acquisition unit 11 stores information on the population for each hour for each preset 500m square mesh. The person attributes are attributes that can affect the selection of songs in karaoke, and are the person's occupation and taste. The types of occupations include "office worker," "self-employed," "housewife / husband," "student," and "unemployed." The types of tastes include hobbies that users like, such as "shopping" and "anime." The above information is information indicating the population for each mesh and time, which is the number of occupations x the number of tastes. Note that only one of the occupations and the tastes may be used as the attributes. In addition, other attributes that can affect the selection of songs in karaoke, such as gender and age, may be used. The above population information for each attribute can be generated as spatial statistical information, for example, from information indicating the location of a mobile phone and information registered about a mobile phone subscriber.

[0028] The learning data acquisition unit 11 acquires time-series information on a plurality of songs used in the past. Specifically, the learning data acquisition unit 11 acquires a log, which is time-series information indicating a plurality of songs sung in the past at a store such as a karaoke store. The learning data acquisition unit 11 acquires the information from the data server 40. FIG. 6 shows an example of the information. The information is a correspondence between a terminal ID, a singing time, a song, and a song ID. The terminal ID is information (identifier) ​​that identifies the terminal 30 used to use a song at a store such as a karaoke store. A terminal ID is set in advance for each terminal 30. The singing time is information indicating the time of use (sung) (for example, year, month, day, hour, and minute as shown in FIG. 6). The song is the name of the song used (sung). The song ID is the song ID of the song used (sung).

[0029] In a store such as a karaoke shop, a user rarely logs in when using the terminal 30. Therefore, unlike the information shown in this embodiment, the singing history in a store such as a karaoke shop may not include information indicating which user sang. For machine learning, it is necessary to treat the history information related to the use of music as a user unit. The learning data acquisition unit 11 generates learning time-series information and learning use music information, which are information used for machine learning, based on the singing time included in the acquired information. In this embodiment, a user's session is from the start to the end of karaoke, for example, from entering a store such as a karaoke shop to leaving the store. The learning data acquisition unit 11 generates the above information in units of sessions.

[0030] The learning data acquisition unit 11 refers to the singing time of the information of the song sung immediately before with the same terminal ID for each piece of acquired information. The learning data acquisition unit 11 judges whether a preset fixed time has passed since the singing of the immediately previous song. If the fixed time has not passed, it is assumed that the song is sung by the same user, and the learning data acquisition unit 11 regards the information as information constituting the same session as the information of the immediately previous song. If the fixed time has passed, it is assumed that the song is sung by a different user, and the learning data acquisition unit 11 regards the information as information constituting a different session from the information of the immediately previous song. For example, in the example of information shown in FIG. 6, the information in the first and second lines constitute one session, the information in the third and fourth lines constitute one session, and the information in the fifth line constitute one session.

[0031] The learning data acquisition unit 11 associates each of the above-mentioned pieces of information about songs sung in each session with each piece of information about the songs using the song ID as a key. Specifically, it associates (combines) the song meta information including the singer ID and genre ID shown in Fig. 2(c), the lyric importance information shown in Fig. 4, and the spectrogram information shown in Fig. 5. Fig. 7(a) shows information before the pieces of information about songs for a certain session are associated, and Fig. 7(b) shows information after the pieces of information about songs for a certain session are associated.

[0032] Next, the learning data acquisition unit 11 generates learning time-series information and learning use song information based on the above information. The learning data acquisition unit 11 sets the songs indicated by each piece of information of the session as the first song, the second song, and so on in order of singing time. The learning data acquisition unit 11 sets the songs in the above order in the session as songs corresponding to the output (Y) of the model for song recommendation during machine learning. Information related to the songs is the learning use song information. Of the information related to the songs, the song ID and singer ID are set as the learning use song information. The learning data acquisition unit 11 converts the information indicating the song ID and singer ID in the learning use song information into a vector with a dimension corresponding to the number of ID types, where an element corresponding to the ID is set to 1 and other elements are set to 0. Note that the learning time-series information does not undergo the above conversion and is treated as numerical information.

[0033] The learning data acquisition unit 11 regards the song immediately preceding the song related to the learning song information in the session as the song corresponding to the input (X) to the model for song recommendation during machine learning. The information related to the song is the learning time series information. All information related to the song, except for the terminal ID, singing time, song, and song ID shown in FIG. 7(b), is regarded as the learning time series information. An example of the correspondence between the learning time series information and the learning song information is shown in FIG. 8. Note that there is no input song corresponding to the first output song. The input information corresponding to the first output song (information of "0" for X (previous song) in FIG. 8) is regarded as information that is a number (e.g., 0) indicating that the song does not exist in all information except for the terminal ID, singing time, song, and song ID shown in FIG. 7(b).

[0034] The learning data acquisition unit 11 acquires information indicating a location corresponding to each session. The location corresponding to each session is the location where the terminal 30 used when singing related to the session is located. The learning data acquisition unit 11 acquires information indicating the mesh in which the location exists as information indicating the location. For example, when the terminal 30 is installed in a store such as a karaoke shop, the learning data acquisition unit 11 stores in advance a correspondence relationship between the terminal ID of the terminal 30 and the mesh in which the terminal 30 is installed, and acquires information indicating the mesh based on the correspondence relationship from the terminal ID indicated by the information related to the session. Also, when the terminal 30 is carried by a user, the learning data acquisition unit 11 stores in advance a correspondence relationship between a location and a mesh including the location, acquires location information (e.g., latitude and longitude information) indicating the location of the terminal 30 together with the information related to the session, and acquires information indicating the mesh from the location information based on the correspondence relationship. In this case, the location information is transmitted from the terminal 30 when karaoke is used from the terminal 30, and is stored in the data server 40 together with the information related to the session.

[0035] The learning data acquisition unit 11 acquires information indicating the population of the mesh from the acquired information indicating the mesh (reads it from information stored by itself). The learning data acquisition unit 11 may also acquire information indicating the population according to the time related to the session. For example, the learning data acquisition unit 11 may acquire information on the population for a time period including the singing time of the first song in the session. The acquired information indicating the population is learning population information used as an input to a model for recommending songs during machine learning.

[0036] The learning data acquisition unit 11 acquires learning population information, learning time-series information, and learning used song information for a sufficient number of sessions to generate a model for song recommendation. That is, this information for a sufficient number of sessions for the model for song recommendation is prepared in advance.

[0037] The learning data acquisition unit 11 is a functional unit that acquires learning data used in machine learning to create a type estimation model. The learning data acquisition unit 11 acquires, as learning data, learning time series information, which is time series information regarding the multiple songs used, and learning type information indicating the type to which the user who used the multiple songs belongs. The learning data acquisition unit 11 acquires learning time series information including information on the singer of the used song. The learning data acquisition unit 11 acquires learning time series information including information on the genre of the used song. The learning data acquisition unit 11 acquires learning time series information including information on words included in the lyrics of the used song. The learning data acquisition unit 11 acquires learning time series information including information on the audio data of the used song. The learning data acquisition unit 11 acquires information that is time series information regarding the multiple songs used and includes information indicating the time when the songs were used, and generates learning time series information and learning used song information based on the time. The learning data acquiring unit 11 acquires learning time-series information for each of a plurality of users, performs clustering on the acquired learning time-series information to generate a plurality of types of estimation targets, and acquires learning type information based on the generated types. Note that the types generated here are pseudo types generated by a clustering process. Specifically, the learning data acquiring unit 11 acquires learning data as follows. Note that in acquiring the learning data below, each session is treated as corresponding to each user.

[0038] The learning data acquisition unit 11 generates learning time-series information from information (an example of which is shown in FIG. 7) in which information on songs sung in each session is associated with information on the songs used in generating learning data for generating a model for song recommendation. The learning data acquisition unit 11 sets the songs indicated by each piece of information on the session as the first song, the second song, and so on, in order of singing time. The learning data acquisition unit 11 sets information on the songs in the above order in the session as songs corresponding to the learning time-series information. All information on the song, except for the terminal ID, singing time, song, and song ID shown in FIG. 7(b), is set as the learning time-series information. FIG. 9(a) shows an example of learning time-series information for two sessions (session A, session B). This information is the learning time-series information for generating a model for song recommendation, shown as X (the previous song) in FIG. 8, with the information "0" removed.

[0039] Further, the learning data acquiring unit 11 acquires the learning type information as follows. The learning data acquiring unit 11 performs clustering of the learning time-series information to generate a plurality of types of estimation targets, and acquires the learning type information based on the generated types. The learning data acquiring unit 11 makes the length of data of all sessions constant. For example, the length of information related to a preset number of songs (for example, three songs) is set. The information with the constant data length is shown in FIG. 9(b).

[0040] For example, when the learning time series information includes information on four or more songs as in session A shown in FIG. 9(a), only information on the first to third songs is used for clustering. When the learning time series information includes information on two or less songs as in session B shown in FIG. 9(a), new information is added for the number of songs that is lacking. The newly added information (information of "0" in session B in FIG. 9(b)) is information that is a numerical value (for example, 0) indicating that the song does not exist in all information except for the terminal ID, singing time, song, and song ID shown in FIG. 7(b). That is, in this case, zeros are added to the rear of the learning time series information. Furthermore, information that indicates an ID by a numerical value such as a singer ID and a genre ID among the information used for clustering is converted into a vector with a dimension corresponding to the type of ID, in which the element corresponding to the ID is set to 1 and the other elements are set to 0. Note that the learning time series information used for machine learning does not undergo the above conversion and is treated as numerical information.

[0041] The learning data acquisition unit 11 clusters the information of the sessions for clustering as shown in FIG. 10. FIG. 10(a) is a diagram showing the information of each session before clustering. One point corresponds to one session. FIG. 10(b) shows the information of the sessions for clustering. It is shown that sessions plotted at the same point belong to the same cluster. As described above, the information of the sessions for clustering is a vector composed of numerical values ​​of the same dimension (for example, information of three songs), so that the clustering can be performed in the same manner as the conventional method. For example, the clustering can be performed using the k-means method. The clusters generated by the clustering become the types (cluster types) to which the user belongs. The learning data acquisition unit 11 assigns identification numbers (cluster numbers) to each type, and associates the identification numbers of the types to which each session belongs to with each session as shown in FIG. 11.

[0042] The learning data acquisition unit 11 regards information on the identification numbers of types in a session, the number of which corresponds to the number of songs included in the session, as information to be output (Y) of a type estimation model in machine learning. This information is the learning type information. The learning data acquisition unit 11 converts the type identification numbers in the learning type information into a vector with a dimension corresponding to the number of types of identification numbers, in which an element corresponding to the identification number is set to 1 and other elements are set to 0. The learning data acquisition unit 11 regards the learning time series information in a session as information to be input (X) to a type estimation model in machine learning. An example of the correspondence between the learning time series information and the learning type information is shown in FIG. 12(a) and FIG. 12(b). As shown in FIG. 12(a) and FIG. 12(b), information related to one song corresponds to information on one identification number.

[0043] The learning data acquisition unit 11 acquires learning time-series information and learning type information for a sufficient number of sessions for generating a type estimation model, similar to information for generating a music recommendation model. Note that the learning data acquisition unit 11 may acquire the learning time-series information and the learning type information by a method other than the above. The learning data acquisition unit 11 outputs the acquired learning data to the model generation unit 12.

[0044] The model generation unit 12 is a functional unit that generates a model for music recommendation by performing machine learning using the learning data acquired by the learning data acquisition unit 11. The model generation unit 12 also generates a model for category estimation by performing machine learning using the learning data acquired by the learning data acquisition unit 11. Fig. 13 shows a schematic diagram of how the model for music recommendation and the model for category estimation generated by the model generation unit 12 are used.

[0045] To generate a model for music recommendation, the model generation unit 12 performs machine learning using information based on the learning population information acquired by the learning data acquisition unit 11 as an input to the model for music recommendation, and information based on the learning-use music information acquired by the learning data acquisition unit 11 as an output of the model for music recommendation. To generate a model for music recommendation, the model generation unit 12 performs machine learning using information based on the learning time-series information acquired by the learning data acquisition unit 11 as an input to the model for music recommendation in chronological order for each song, and information based on the learning-use music information acquired by the learning data acquisition unit 11 as an output of the model for music recommendation. To generate a model for music recommendation, the model generation unit 12 performs machine learning using information based on the learning type information acquired by the learning data acquisition unit 11 as an input to the model for music recommendation.

[0046] As shown in FIG. 13, the model for music recommendation generated by the model generation unit 12 is a model that inputs information based on recommendation population information (population information in FIG. 13) indicating the population for each human attribute in a place where the music is used, recommendation time-series information (music time-series information in FIG. 13) which is time-series information on a plurality of used music pieces, and recommendation type information (type information in FIG. 13) indicating the type to which the user belongs, and outputs information about the music to be recommended (recommendation information in FIG. 13). That is, the model for music recommendation is a model that predicts information about the music to be recommended, which is Y, from the population information, type information, and time-series information about the music, which is X, shown in FIG. 8. The model for music recommendation includes, for example, a neural network. The neural network may be multi-layered. That is, the model generation unit 12 may generate the model for music recommendation by performing deep learning.

[0047] For example, in a music recommendation model, the input layer is provided with neurons for inputting population information, the number of which is equal to the number of dimensions of the population information. Specifically, as described above, neurons are provided in the number of occupations x the number of preferences.

[0048] Also, for example, in the model for music recommendation, a neuron is provided in the input layer as a neuron for inputting information about a song, the number of neurons being equal to the number of dimensions of information about a single song. Specifically, as described above, neurons are provided as the number of dimensions of the song meta information, lyric importance information, and spectrogram information. The model for music recommendation inputs information about a plurality of songs in time series in the order of the songs in time series. Also, information indicating that no song has been sung before the information about the sung song is input is input to the model for music recommendation. This information is information equivalent to the song meta information, lyric importance information, and spectrogram information in which all values ​​are set to 0, and is the same information as the information of "0" in X (the previous song) in FIG. 8. By inputting this information, the model for music recommendation can output information about a song to be recommended even when no song has been sung by the user.

[0049] Also, for example, in the music recommendation model, the input layer is provided with neurons for inputting type information, the number of which corresponds to the dimension of the type information (for example, the number of types). As will be described later, the type information is also time-series information like the information about music, so the music recommendation model inputs time-series type information.

[0050] For example, the model for song recommendation outputs, as information on the songs to be recommended, a numerical value indicating the degree to which each song is recommended and a numerical value indicating the degree to which a singer is recommended. In the model for song recommendation, the output layer is provided with neurons for outputting a numerical value indicating the degree to which a song is recommended, the number of neurons being the number of songs to be recommended. In machine learning, when a vector is used as the song information to be used for learning, in which the element of the sung song is set to 1 and the elements of the other songs are set to 0, the larger the output numerical value, the higher the degree of recommendation. In the model for song recommendation, the output layer is provided with neurons for outputting a numerical value indicating the degree to which a singer is recommended, the number of neurons being the number of singers to be recommended. In machine learning, when a vector is used as the song information to be used for learning, in which the element of the singer of the sung song is set to 1 and the elements of the other singers are set to 0, the larger the output numerical value, the higher the degree of recommendation.

[0051] The model for song recommendation outputs information about a song to be recommended each time information about a song is input. The information about a song to be recommended that is output is information about a song to be recommended after the song related to the input information. First, when the model for song recommendation inputs information indicating that no song has been sung previously, the model for song recommendation outputs information about a song to be recommended as the first song. Next, when information about a song sung as the first song is input, the model for song recommendation outputs information about a song to be recommended as the second song. In this way, the model for song recommendation outputs information about a song to be recommended next in response to the sequential input of information about sung songs.

[0052] The model for recommending music generated by the model generation unit 12 may be configured by something other than a neural network.

[0053] The model generation unit 12 inputs the learning data from the learning data acquisition unit 11. The model generation unit 12 uses the respective numerical values ​​indicated by the learning time-series information, learning population information, and learning type information, which are information on time-series songs, among the input learning data, as input values ​​to a model for song recommendation, and performs machine learning using the learning used song information as an output value of the model for song recommendation to generate a model for song recommendation. The learning time-series information and the learning type information are input values ​​to the model for song recommendation in the order of the time series of the session in units of songs. Note that the first information of the learning type information (information corresponding to information indicating that there is no song, which is the first information of the learning time-series information) is information indicating that it is not any type (for example, information in which elements corresponding to all types are set to 0).

[0054] In addition, the learning song information is set as the output value of the song recommendation model in the chronological order of the session for each song. At this time, as shown in the correspondence relationship in Fig. 8, when a certain learning time-series information and learning type information are set as input values, the learning song information corresponding to these information is set as the output value. In other words, information on the next song to be recommended is output based on the history of songs sung in the session.

[0055] In the model for music recommendation, the singer ID input as a single numerical value is converted into a feature quantity, which is a numerical value with a preset number of dimensions, as shown in FIG. 3(a). When starting machine learning of the model for music recommendation, the model generation unit 12 gives the initial value of the feature quantity as a random numerical value. In the model for music recommendation, the converted feature quantity is used to generate an output value. By converting into a feature quantity, the model for music recommendation can appropriately handle the ID as a numerical value, and an appropriate output value can be obtained. The genre ID is also converted into a feature quantity in the same way as the singer ID. Furthermore, the learning population information may also be converted into a feature quantity. Note that, in the model for music recommendation, each piece of information does not need to be converted into a feature quantity.

[0056] As described above, the learning time-series information and the learning type information are input to the song recommendation model in units of songs in sequence. In the song recommendation model, the information is treated as time-series information, so that a GRU (Gated Recurrent Unit) is used inside the song recommendation model.

[0057] The above machine learning itself can be performed in the same manner as conventional machine learning methods. The model generation unit 12 outputs the generated music recommendation model to the model utilization system 20.

[0058] In order to generate a type estimation model, the model generation unit 12 performs machine learning using information based on the learning time-series information acquired by the learning data acquisition unit 11 as an input to the type estimation model in chronological order for each song, and uses the information based on the learning type information acquired by the learning data acquisition unit 11 as an output of the type estimation model.

[0059] As shown in FIG. 13, the type estimation model generated by the model generation unit 12 is a model that inputs information based on type estimation time series information (time series information of songs in FIG. 13), which is time series information on a plurality of songs used by a user, and outputs information on the type to which the estimated user belongs (type information in FIG. 13). That is, the type estimation model is a model that predicts type information Y from time series information of songs X shown in FIG. 12(a) and FIG. 12(b). The type estimation model includes, for example, a neural network. The neural network may be multi-layered. That is, the model generation unit 12 may generate the type estimation model by performing deep learning.

[0060] For example, in the type estimation model, neurons for inputting information about a single song are provided in the input layer. These neurons are the same as the neurons for inputting information about a song in the song recommendation model, except for the following: Unlike the song recommendation model, the type estimation model is not trained or estimated (predicted) when no song has been sung by the user.

[0061] For example, the type estimation model outputs a numerical value indicating the degree to which the user belongs to each type as information on the type to be estimated. In the type estimation model, the output layer is provided with neurons equal to the number of types as neurons for outputting numerical values ​​indicating the degree to which the user belongs. In machine learning, when a vector is used as learning type information in which the element of the identification number of the type to which the user belongs is set to 1 and the elements of other songs are set to 0, the larger the output numerical value, the greater the degree to which the user belongs to that type.

[0062] The category estimation model outputs information about the category each time information about a song is input. The output information about the category is information about the category estimated by the time-series song related to the input information. First, when information about the song sung as the first song is input, the category estimation model outputs information about the category estimated from the information. Next, when information about the song sung as the second song is input, the category estimation model outputs information about the category estimated from the previous information (information about the first and second songs). In this way, the category estimation model outputs information about the estimated category sequentially according to the sequential input of information about the sung songs. Note that the degree to which the user belongs to each category may also change according to the sequential input of information about the sung songs.

[0063] The information on the type output by the type estimation model is input to a music recommendation model as shown in FIG. 13 when recommending music to a user.

[0064] The type estimation model generated by the model generating unit 12 may be configured by something other than the above-mentioned neural network.

[0065] The model generation unit 12 performs machine learning using the type information as an output value of the type estimation model, and generates a type estimation model by using the learning time-series information, which is information on time series music pieces among the input learning data, as input values ​​to the type estimation model. The learning time-series information is input to the type estimation model in the chronological order of the sessions in units of music pieces.

[0066] Moreover, the type information is set as an output value of the type estimation model in a time-series order in units corresponding to the units of the learning time-series information as shown in Fig. 12. In this case, when a certain piece of learning time-series information is set as an input value, the type information corresponding to that information is set as an output value, as shown in the correspondence relationship in Fig. 12. In other words, type information estimated from the history of songs sung in the session is output.

[0067] As described above, the learning time-series information is input to the music type estimation model in sequence for each song. In the music type estimation model, the information is treated as time-series information, so that a GRU is used inside the music type estimation model.

[0068] The above machine learning itself can be performed in the same manner as conventional machine learning methods. The model generation unit 12 outputs the generated type estimation model to the model utilization system 20. The processing by the learning data acquisition unit 11 and the model generation unit 12 may be performed as daily batch processing, for example. The above are the functions of the model generation system 10 according to this embodiment.

[0069] Next, a description will be given of the functions of the model utilization system 20 according to this embodiment. As shown in FIG.

[0070] The model usage data acquisition unit 21 is a functional unit that acquires model usage data used for a music recommendation model and a type estimation model. The model usage data acquisition unit 21 acquires recommendation population information indicating the population for each human attribute in the place where the music is used as the model usage data. The model usage data acquisition unit 21 acquires recommendation time series information, which is time series information regarding the multiple songs used, as the model usage data. The model usage data acquisition unit 21 acquires type estimation time series information, which is time series information regarding the multiple songs used, as the model usage data.

[0071] The model usage data acquisition unit 21 acquires data when a user sings karaoke at a store such as a karaoke parlor, that is, when the user uses a song. When the user sings karaoke, a request for recommended content is made from the terminal 30 used by the user to the recommendation server 1. The request for recommended content is made first before the user registers a song to be sung.

[0072] The model usage data acquisition unit 21 acquires information indicating a place where karaoke is performed. For example, the model usage data acquisition unit 21 acquires information indicating a mesh in which the place exists as information indicating the place. The request for the recommended content includes the terminal ID of the terminal 30 or location information (e.g., latitude and longitude information) indicating the location of the terminal 30, and the model usage data acquisition unit 21 acquires information indicating a mesh from the terminal ID or location information of the terminal 30 in the same manner as the method by the learning data acquisition unit 11. The model usage data acquisition unit 21 acquires information indicating the population of the mesh from the acquired information indicating the mesh as population information for recommendation (reads it from information stored by itself) in the same manner as the method by the learning data acquisition unit 11. Note that if the stored population information is information for each time period, the model usage data acquisition unit 21 acquires population information associated with the current time.

[0073] When the user starts karaoke and sings a song, the terminal 30 includes information indicating the song sung by the user in a request for recommended content and transmits the information to the recommendation server 1. The transmitted information is, for example, the information shown in each row of FIG. 6. The transmitted information may include information related to a plurality of songs (for example, the most recent N logs). The model utilization data acquisition unit 21 receives the information transmitted from the terminal 30. As shown in FIG. 7, the model utilization data acquisition unit 21 associates each piece of information related to the song with the information of the song received from the terminal 30, using the song ID as a key. The associated information is song meta information, lyric importance information, and spectrogram information. The association is performed in the same manner as the method by the learning data acquisition unit 11.

[0074] The model utilization data acquisition unit 21 receives information transmitted from the terminal 30 one by one and performs the above-mentioned association. The model utilization data acquisition unit 21 may also determine whether information on each piece of music acquired in the same manner as the method by the learning data acquisition unit 11 constitutes the same session as information on another piece of music. The model utilization data acquisition unit 21 generates information on the time series of music for each session based on the determination. The model utilization data acquisition unit 21 sets the generated time series information on the music as type estimation time series information. The model utilization data acquisition unit 21 also adds the above-mentioned information indicating that no music has been sung before the generated time series information on the music to set it as recommendation time series information. The information indicating that no music has been sung is for determining information to be recommended using a music recommendation model when the user has not sung any music. Note that, when information indicating a music piece is not transmitted from the terminal 30, the recommendation time series information is only information indicating that no music has been sung, and there is no type estimation time series information. The model usage data acquisition unit 21 uses each of the above information as information to be used for making recommendations to the user related to the session.

[0075] The model utilization data acquisition unit 21 outputs the acquired recommendation population information, recommendation time series information, and type estimation time series information to the model utilization unit 22.

[0076] The model utilization unit 22 is a functional unit that uses a model for music recommendation and a model for type estimation to determine information about music to be recommended to a user. The model utilization unit 22 inputs information based on population information for recommendation acquired by the model utilization data acquisition unit 21 into the model for music recommendation to determine information to be recommended. The model utilization unit 22 inputs information based on time-series information for recommendation acquired by the model utilization data acquisition unit 21 into the model for music recommendation in chronological order for each song to determine information to be recommended. The model utilization unit 22 inputs information based on time-series information for type estimation acquired by the model utilization data acquisition unit 21 into the model for type estimation in chronological order for each song to estimate the type to which the user belongs.

[0077] The model utilization unit 22 inputs and stores the music recommendation model and the type estimation model generated by the model generation system 10, and uses them to determine information about music to be recommended to users. The model utilization unit 22 inputs recommendation population information and recommendation time-series information (type estimation time-series information) from the model utilization data acquisition unit 21.

[0078] The model utilization unit 22 obtains type information, which is an output value from the type estimation model, using the type estimation time series information as an input value to the type estimation model. When the type estimation time series information is used as an input value to the type estimation model, it is performed in the order of the time series of the session in units of songs. The type information that becomes an output value is a numerical value (vector) indicating the degree to which the user belongs to each type. The type information that becomes an output value is obtained every time the type estimation time series information is input to the type estimation model in units of songs. In other words, the type information that becomes an output value is obtained for each song related to the type estimation time series information. The numerical value corresponds to the estimation result of the type to which the user belongs. Note that if information indicating a song is not transmitted from the terminal 30, there is no type estimation time series information, and therefore type estimation is not performed.

[0079] As shown in Fig. 13, the model utilization unit 22 obtains recommendation information, which is an output value from the model for song recommendation, using population information for recommendation (population information in Fig. 13), time series information for recommendation (time series information of songs in Fig. 13), and type information obtained from the model for type estimation (type information in Fig. 13) as input values ​​to the model for song recommendation. When the time series information for recommendation and the type information are input values ​​to the model for type estimation, they are done in the order of the time series of the session for each song. Note that since there is no type information corresponding to the information indicating that no song has been sung in the time series information for recommendation, information indicating that it is not any type is used for the information indicating that no song has been sung.

[0080] The recommendation information that serves as an output value is a numerical value indicating the degree to which each song is recommended and a numerical value (vector) indicating the degree to which each singer is recommended. The recommendation information that serves as an output value is obtained each time recommendation time-series information is input to the song recommendation model on a song-by-song basis. In other words, the recommendation information that serves as an output value is obtained for each song related to the recommendation time-series information (including the case where there is no song).

[0081] The model utilization unit 22 determines information to be recommended to the user based on the output value. For example, it determines that the song and singer with the largest numerical value are to be recommended to the user. Alternatively, it determines that a preset number of songs and singers are to be recommended to the user in descending order of numerical value. The model utilization unit 22 makes this determination every time new recommendation information is obtained, that is, every time information indicating a new song is transmitted from the terminal 30 as a sung song. The model utilization unit 22 transmits the determined information to the terminal 30. The terminal 30 receives the information related to the recommendation transmitted from the recommendation server 1 and performs output such as display. The user can refer to the display to determine the next song to sing.

[0082] The processing by the model utilization data acquisition unit 21 and the model utilization unit 22 is performed, for example, as described above, as real-time processing in response to reception of information from the terminal 30. The above is the function of the model utilization system 20 according to this embodiment.

[0083] Next, the process executed by the recommendation server 1 according to this embodiment (the operation method performed by the recommendation server 1) will be described with reference to the flowcharts of Fig. 14 and Fig. 15. First, the process executed when generating a music recommendation model and a type estimation model, that is, the process executed by the model generation system 10 according to this embodiment, will be described with reference to the flowchart of Fig. 14.

[0084] In this process, first, the learning data acquisition unit 11 acquires learning data used in machine learning to create a music recommendation model and learning data used in machine learning to create a type estimation model (S01). Next, the learning data acquisition unit 11 performs clustering of the learning time series information of the acquired learning data to generate a plurality of types, and acquires learning type information based on the generated types (S02). Next, the model generation unit 12 performs machine learning based on the learning time series information and the learning type information of the learning data to generate a type estimation model (S03). During this machine learning, information based on the learning time series information is input to the type estimation model, and information based on the learning type information is output to the type estimation model. Information based on the learning time series information is input to the type estimation model in chronological order in units of music.

[0085] Next, the model generation unit 12 performs machine learning based on the learning time-series information, learning population information, learning type information, and learning used song information from the learning data to generate a song recommendation model (S04). During this machine learning, information based on the learning time-series information, learning population information, and learning type information is input to the song recommendation model, and information based on the learning used song information is output from the song recommendation model. Information based on the learning time-series information is input to the song recommendation model in chronological order for each song. The above is the process executed when generating a song recommendation model and a type estimation model.

[0086] Next, a process executed when recommending music to a user, that is, a process executed by the model utilization system 20 according to this embodiment, will be described with reference to the flowchart in Fig. 15. In this process, first, the model utilization data acquisition unit 21 acquires model utilization data used for the music recommendation model and model utilization data used for the type estimation model (S11). The model utilization data is acquired when a request for recommended content transmitted from the terminal 30 is received as a trigger.

[0087] Next, the model utilization unit 22 inputs information based on the type estimation time series information from among the model utilization data to the type estimation model, and estimates the type to which the user to be recommended belongs (S12). The information based on the type estimation time series information is input to the type estimation model in chronological order in units of songs. As an estimation result, type information, which is an output value from the type estimation model, is obtained.

[0088] Next, the model utilization unit 22 inputs information based on the population information for recommendation, information based on the time-series information for recommendation, and type information from the model utilization data into the model for song recommendation, and determines recommendation information regarding songs to be recommended to the user (S13). Information based on the information based on the time-series information for recommendation is input to the model for song recommendation in chronological order for each song. Next, the model utilization unit 22 recommends songs to the user based on the time-series information for recommendation (S14). The user can refer to the recommended information to decide which song to sing next.

[0089] Note that when a request for recommended content is received before the first song is sung, there is no time series information on the time series of the multiple songs used, so the time series information is not used in the above process. In this case, type estimation (S12) is not performed. The above is the process executed when recommending songs to a user.

[0090] According to this embodiment, it is possible to recommend songs using population information of the place where the song is used based on a song recommendation model generated by machine learning. Even if the user to whom the recommendation is to be made does not sing and has no singing history, it is possible to make an appropriate recommendation according to the population of each person's attribute in the place, even if the user has a singing history. Therefore, according to this embodiment, it is possible to make an appropriate recommendation of songs. For example, compared to a case where a popularity ranking generated in a karaoke store chain is presented to a user, it is possible to make an appropriate recommendation according to the attributes of users in the place where the song is used, such as the location of the karaoke store, from the first song.

[0091] In this embodiment, information other than population information is also used for the recommendation, but it is not necessary to use information other than population information. Furthermore, in addition to population information, information other than the information shown in this embodiment may be used for the recommendation.

[0092] In addition, population information according to time may be used as in the present embodiment. With this configuration, appropriate recommendations according to time can be made. However, the population information does not necessarily have to be according to time.

[0093] In addition, as in the present embodiment, the attributes of a person related to population information can be the person's occupation and preferences. With this configuration, recommendations can be made reliably and appropriately. Also, only one of the person's occupation and preferences may be used. Also, attributes other than those described above may be used.

[0094] In addition, in this embodiment, the same music recommendation model can recommend both songs and singers. However, it is not necessary to recommend both of them, and it may be possible to recommend only one of them. In addition, any information related to music other than the above may be recommended.

[0095] Furthermore, in this embodiment, the recommendation server 1 includes the model generation system 10 and the model usage system 20, but the model generation system 10 and the model usage system 20 may each be implemented independently.

[0096] In addition, each of the music recommendation model and the type estimation model generated by the model generation system 10 is expected to be used as a program module that is a part of artificial intelligence software. The music recommendation model and the type estimation model are used in a computer equipped with a CPU and a memory. Specifically, the CPU of the computer operates in accordance with instructions from the music recommendation model and the type estimation model stored in the memory to input information to the input layer of the neural network, perform calculations based on the learned weighting coefficients in the neural network, and output results from the output layer of the neural network.

[0097] The block diagrams used to explain the above embodiments show functional blocks. These functional blocks (components) are realized by any combination of at least one of hardware and software. The method of realizing each functional block is not particularly limited. That is, each functional block may be realized by using one device that is physically or logically combined, or may be realized by using two or more devices that are physically or logically separated and directly or indirectly connected (for example, by wire, wirelessly, etc.). The functional blocks may be realized by combining the one device or the multiple devices with software.

[0098] Functions include, but are not limited to, judgment, determination, judgment, calculation, computation, processing, derivation, investigation, search, confirmation, reception, transmission, output, access, resolution, selection, selection, establishment, comparison, assumption, expectation, consideration, broadcasting, notifying, communicating, forwarding, configuring, reconfiguring, allocating, mapping, and assignment. For example, a functional block (component) that performs the function of transmission is called a transmitting unit or a transmitter. In either case, as described above, there is no particular limitation on the method of realization.

[0099] For example, the recommendation server 1 in an embodiment of the present disclosure may function as a computer that performs the information processing of the present disclosure. Fig. 16 is a diagram showing an example of a hardware configuration of the recommendation server 1 according to an embodiment of the present disclosure. The recommendation server 1 described above may be physically configured as a computer device including a processor 1001, a memory 1002, a storage 1003, a communication device 1004, an input device 1005, an output device 1006, a bus 1007, and the like. The terminal 30 and the data server 40 may also have a similar device configuration.

[0100] In the following description, the term "apparatus" can be replaced with a circuit, a device, a unit, etc. The hardware configuration of the recommendation server 1 may be configured to include one or more of the devices shown in the figure, or may be configured to exclude some of the devices.

[0101] Each function in the recommendation server 1 is realized by loading a specific software (program) onto hardware such as a processor 1001 and a memory 1002, so that the processor 1001 performs calculations, controls communications via a communication device 1004, and controls at least one of reading and writing of data in the memory 1002 and the storage 1003.

[0102] The processor 1001, for example, operates an operating system to control the entire computer. The processor 1001 may be configured with a central processing unit (CPU) including an interface with peripheral devices, a control device, an arithmetic unit, a register, etc. For example, each function in the recommendation server 1 described above may be realized by the processor 1001.

[0103] Furthermore, the processor 1001 reads out a program (program code), a software module, data, etc. from at least one of the storage 1003 and the communication device 1004 into the memory 1002, and executes various processes according to the read out programs. As the program, a program that causes a computer to execute at least a part of the operations described in the above-mentioned embodiment is used. For example, each function in the recommendation server 1 may be realized by a control program stored in the memory 1002 and operated in the processor 1001. Although the above-mentioned various processes have been described as being executed by one processor 1001, they may be executed simultaneously or sequentially by two or more processors 1001. The processor 1001 may be implemented by one or more chips. The program may be transmitted from a network via a telecommunication line.

[0104] The memory 1002 is a computer-readable recording medium, and may be composed of at least one of, for example, a read only memory (ROM), an erasable programmable ROM (EPROM), an electrically erasable programmable ROM (EEPROM), a random access memory (RAM), etc. The memory 1002 may be called a register, a cache, a main memory (primary storage device), etc. The memory 1002 can store a program (program code), software modules, etc. that are executable to perform information processing according to an embodiment of the present disclosure.

[0105] Storage 1003 is a computer-readable recording medium, and may be composed of at least one of, for example, an optical disk such as a CD-ROM (Compact Disc ROM), a hard disk drive, a flexible disk, a magneto-optical disk (e.g., a compact disk, a digital versatile disk, a Blu-ray (registered trademark) disk), a smart card, a flash memory (e.g., a card, a stick, a key drive), a floppy (registered trademark) disk, a magnetic strip, and the like. Storage 1003 may be called an auxiliary storage device. The storage medium provided in recommendation server 1 may be, for example, a database, a server, or other appropriate medium including at least one of memory 1002 and storage 1003.

[0106] The communication device 1004 is hardware (transmission / reception device) for performing communication between computers via at least one of a wired network and a wireless network, and is also called, for example, a network device, a network controller, a network card, or a communication module.

[0107] The input device 1005 is an input device (e.g., a keyboard, a mouse, a microphone, a switch, a button, a sensor, etc.) that accepts input from the outside. The output device 1006 is an output device (e.g., a display, a speaker, an LED lamp, etc.) that performs output to the outside. The input device 1005 and the output device 1006 may be integrated into one device (e.g., a touch panel).

[0108] In addition, each device such as the processor 1001 and the memory 1002 is connected by a bus 1007 for communicating information. The bus 1007 may be configured using a single bus, or may be configured using different buses between each device.

[0109] The recommendation server 1 may be configured to include hardware such as a microprocessor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a programmable logic device (PLD), a field programmable gate array (FPGA), etc., and some or all of the functional blocks may be realized by the hardware. For example, the processor 1001 may be implemented using at least one of these pieces of hardware.

[0110] The order of the steps, sequences, flow charts, etc. of each aspect / embodiment described in this disclosure may be changed unless inconsistent. For example, the methods described in this disclosure present elements of various steps using an example order, and are not limited to the particular order presented.

[0111] The input and output information may be stored in a specific location (e.g., memory) or may be managed using a management table. The input and output information may be overwritten, updated, or added to. The output information may be deleted. The input information may be transmitted to another device.

[0112] The determination may be based on a value represented by a single bit (0 or 1), a Boolean value (true or false), or a numerical comparison (e.g., comparison with a predetermined value).

[0113] Each aspect / embodiment described in this disclosure may be used alone, in combination, or switched according to execution. In addition, notification of specific information (e.g., notification that "X is the case") is not limited to being done explicitly, but may be done implicitly (e.g., not notifying the specific information).

[0114] Although the present disclosure has been described in detail above, it is clear to those skilled in the art that the present disclosure is not limited to the embodiments described in the present disclosure. The present disclosure can be implemented in modified and altered forms without departing from the spirit and scope of the present disclosure as defined by the claims. Therefore, the description of the present disclosure is intended to be illustrative and does not have any limiting meaning on the present disclosure.

[0115] Software shall be construed broadly to mean instructions, instruction sets, code, code segments, program code, programs, subprograms, software modules, applications, software applications, software packages, routines, subroutines, objects, executable files, threads of execution, procedures, functions, etc., whether referred to as software, firmware, middleware, microcode, hardware description language, or otherwise.

[0116] Additionally, software, instructions, information, etc. may be transmitted or received over a transmission medium. For example, if the software is transmitted from a website, server, or other remote source using wired and / or wireless technologies (such as coaxial cable, fiber optic cable, twisted pair, Digital Subscriber Line (DSL)), and / or wireless technologies (such as infrared, microwave, etc.), then these wired and / or wireless technologies are included within the definition of transmission media.

[0117] As used in this disclosure, the terms "system" and "network" are used interchangeably.

[0118] Furthermore, the information, parameters, etc. described in this disclosure may be expressed using absolute values, may be expressed using relative values ​​from a predetermined value, or may be expressed using other corresponding information.

[0119] At least one of the server and the client may be called a transmitting device, a receiving device, a communication device, etc. At least one of the server and the client may be a device mounted on a moving body, the moving body itself, etc. The moving body may be a vehicle (e.g., a car, an airplane, etc.), an unmanned moving body (e.g., a drone, an autonomous vehicle, etc.), or a robot (manned or unmanned). At least one of the server and the client may include a device that does not necessarily move during communication operations. For example, at least one of the server and the client may be an IoT (Internet of Things) device such as a sensor.

[0120] Furthermore, the server in the present disclosure may be read as a client terminal. For example, each aspect / embodiment of the present disclosure may be applied to a configuration in which communication between a server and a client terminal is replaced with communication between a plurality of user terminals (which may be called, for example, D2D (Device-to-Device) or V2X (Vehicle-to-Everything)). In this case, the client terminal may be configured to have the functions of the above-mentioned server.

[0121] Similarly, the client terminal in the present disclosure may be read as a server. In this case, the server may have the functions of the client terminal described above.

[0122] As used in this disclosure, the terms "determining" and "determining" may encompass a wide variety of actions. "Determining" and "determining" may include, for example, judging, calculating, computing, processing, deriving, investigating, looking up, search, inquiry (e.g., searching in a table, database, or other data structure), ascertaining, and the like. "Determining" and "determining" may also include receiving (e.g., receiving information), transmitting (e.g., sending information), input, output, accessing (e.g., accessing data in a memory), and the like. In addition, "judgment" and "decision" can include considering resolving, selecting, choosing, establishing, comparing, etc., to be a "judgment" or "decision." In other words, "judgment" and "decision" can include considering some action to be a "judgment" or "decision." In addition, "judgment" can be interpreted as "assuming," "expecting," "considering," etc.

[0123] The terms "connected," "coupled," or any variation thereof, refer to any direct or indirect connection or coupling between two or more elements, and may include the presence of one or more intermediate elements between two elements that are "connected" or "coupled" to each other. The coupling or connection between elements may be physical, logical, or a combination thereof. For example, "connected" may be read as "access." As used in this disclosure, two elements may be considered to be "connected" or "coupled" to each other using at least one of one or more wires, cables, and printed electrical connections, as well as electromagnetic energy having wavelengths in the radio frequency range, microwave range, and optical (both visible and invisible) range, as some non-limiting and non-exhaustive examples.

[0124] As used in this disclosure, the phrase "based on" does not mean "based only on," unless expressly stated otherwise. In other words, the phrase "based on" means both "based only on" and "based at least on."

[0125] Any reference to an element using a designation such as "first," "second," etc., used in this disclosure does not generally limit the quantity or order of those elements. These designations may be used in this disclosure as a convenient method of distinguishing between two or more elements. Thus, a reference to a first and a second element does not imply that only two elements may be employed or that the first element must precede the second element in some way.

[0126] When the terms "include," "including," and variations thereof are used in this disclosure, these terms are intended to be inclusive, similar to the term "comprising." Further, when used in this disclosure, the term "or" is not intended to be an exclusive or.

[0127] In this disclosure, where articles have been added due to translation, such as a, an, and the in English, the disclosure may include that the nouns following these articles are in the plural form.

[0128] In the present disclosure, the term "A and B are different" may mean "A and B are different from each other." The term may also mean "A and B are each different from C." Terms such as "separate" and "coupled" may also be interpreted in the same way as "different." [Explanation of symbols]

[0129] 1...recommendation server, 10...model generation system, 11...learning data acquisition unit, 12...model generation unit, 20...model utilization system, 21...model utilization data acquisition unit, 22...model utilization unit, 30...terminal, 40...data server, 1001...processor, 1002...memory, 1003...storage, 1004...communication device, 1005...input device, 1006...output device, 1007...bus.

Claims

1. A music recommendation model generation system that generates a music recommendation model used for music recommendation, comprising: a learning data acquisition unit that acquires learning population information indicating the population of each person for each attribute in a place where a song is used, and learning song information related to the song, which are learning data used in machine learning; a model generation unit that performs machine learning using information based on the learning population information acquired by the learning data acquisition unit as an input to a model for music recommendation and information based on the learning music information acquired by the learning data acquisition unit as an output of the model for music recommendation to generate a model for music recommendation; Equipped with A model generation system for recommending songs, wherein the learning population information is information for locations where karaoke stores are known to exist in advance.

2. The system for generating a model for recommending music according to claim 1 , wherein the learning data acquisition unit acquires learning population information indicating people's attributes and population at each time in a place where the music is used.

3. The system for generating a model for music recommendation according to claim 1 or 2, wherein the person's attributes include at least one of the person's occupation and preferences.

4. A music recommendation system that determines information to be recommended using a music recommendation model generated by the music recommendation model generation system according to any one of claims 1 to 3, a model usage data acquisition unit that acquires recommendation population information indicating the population of each person attribute in a place where the song is used; a model utilization unit that inputs information based on the population information for recommendation acquired by the model utilization data acquisition unit into the model for music recommendation to determine information to be recommended; Equipped with The music recommendation system, wherein the recommendation population information is information for a location where it is known in advance that a karaoke store exists.

Citation Information

Patent Citations

  • Karaoke system

    JP2012078387A

  • System and method for associating music with brain-state data

    US20150297109A1

  • Cognitive music selection system and method

    US20180314959A1