KR102100963B1 - Music recommendation service apparatus and system for serendipity, and recommendation method thereof - Google Patents

Abstract

The present invention is a music recommendation service device and a system using the same and a method and method for recommending the music for discovering music to make it easy to discover music that suits your tastes by recommending music that is different from the existing ones, but is suitable for preferences based on the user's preferences As to, the user preference is analyzed by analyzing user music history information obtained through various paths, and after selecting a mate having a high degree of similarity with the user preference, music is recommended using the common preference song and user experience song of the mate, In other words, it is possible to exclude a previously recommended song and to discover a new song of the user's preference through a new recommended music each time, or to rediscover an experienced song. In particular, as the recommendation is repeated due to the exclusion of the previous recommendation song, a certain percentage of mates among the selected mates such that music with a low preference is not recommended to the user is a new mate. It has the effect of making it possible.

Description

Music recommendation service device for discovering music, system using same, and method for recommending it {MUSIC RECOMMENDATION SERVICE APPARATUS AND SYSTEM FOR SERENDIPITY, AND RECOMMENDATION METHOD THEREOF}

The present invention relates to a music recommendation system, and in particular, a music recommendation service device for discovering music so that music suitable for preference can be easily discovered by recommending music that is different from the previous one and suitable for preference based on the user's preference And a system using the same and a method for recommending the same.

2. Description of the Related Art With the development of wireless communication technology and terminal technology, a life in which a user encounters various types of media through various paths using a terminal has become common.

In particular, by the mobility of digitized media, users can use their own media anywhere through a cloud service in addition to the media held in the terminal itself, and use new media in streaming format through their terminals through various media provision services. Share with friends.

With the advent of various media providing services, these digitized media are expanding exponentially in their types and numbers, and they can be used for free or at a low cost, so users can easily access them.

It is music that accounts for a significant portion of the vastly provided media, and the user's satisfaction is high compared to the communication cost or load. Accordingly, various online music services have emerged and are also evolving in the form of music social network services.

On the other hand, among these online music services, users can recommend similar sound sources using meta information such as the user's own sound source or genre, artist, album, etc. of the sound source that has been played before, so that the user can find a new sound that meets his or her preference in a vast sound source. Music is easily discovered and enjoyed. As access to vast media becomes common, users are expected to easily discover new music that suits their tastes, and service providers induce interest in or purchase of these recommended music, resulting in commercial effects. I want to get

On the other hand, in order to use the emotional similarity of a person when recommending music, a method of combining and recommending music of users similar to the user's preferred genre or taste is also used.

However, in the existing recommendation method, since the musical preference of a person is not easily changed, a significant portion of the music recommendation results suited to the user's preference is the same as the previous recommendation result, and the freshness is reduced when the recommendation service is repeatedly used. That is, the target preference matching is rather limited, and the adjacent region sound sources of the recommended object are not recommended, and only the recommended songs are re-recommended each time, and thus, it cannot respond to a user's request to discover new music.

Korean Patent Publication No. 10-2010-0022542 [Name: P2P based music recommendation system and method]

An object according to an embodiment of the present invention for improving the above-mentioned problems is to set user preferences in consideration of music log information including meta information of a sound source obtained through various paths and explicit preference information of a user, and similarity to such user preferences After selecting the high mates, the recommended songs are selected by reflecting the information on the preferred music and the user's preferred music, but the recommended songs are excluded until a certain condition is made so that new music is recommended each time. have.

Another object of the embodiment of the present invention is to expand the population so that whenever a user selects a mate similar to the user preference, a new mate is included at a predetermined rate each time, so that new music that is not desired to the user can be found at every recommendation.

Another object of the embodiment of the present invention is to select mates with similar preferences and then extract common preference songs of the corresponding mates and compare the user's preference songs to discover new music that they do not have, or to rediscover music that they and their mates have in common. It's about making it possible.

Another object of the embodiments of the present invention is to use weights when selecting a mate by assigning weights based on explicit preference information when analyzing user preferences, but to determine the similarity in consideration of the volume of history so as to increase the precision for selecting the mate. Has to do.

The music recommendation service device for discovering music according to an embodiment of the present invention for achieving the above object is the meta information about the sound source existing in the user device and the log information for the sound source through the user device where the music recommendation app is installed. A history analysis unit for analyzing user preferences by collecting information as history information; A mate selection unit for selecting a new mate group having a high degree of similarity with the user preference and containing a new mate at least a predetermined ratio or higher than a previous mate group; And a music recommendation unit that generates new recommended music information different from previously recommended music information based on common preference songs of the mate group selected by the mate selection unit and provides the recommended music information to a user device on which the music recommendation app is installed. .

The history analysis unit may analyze user preferences by receiving user subscription information through a user device on which the music recommendation app is installed, and collecting music history information of the user from a social network service server or an external music service server to which the user has subscribed. .

The music recommendation unit selects recommended music information from a difference set or intersection between a common preference song of the selected mate group and the user experience song of the collected history information.

More specifically, the mate selection unit calculates the similarity by comparing the music in the user's history information and the music in another person's history information, but applies weights according to preferences in the history information of the music, and the total number of songs viewed. Similarity can be adjusted accordingly.

Meanwhile, the history analysis unit collects a user history for the recommended music from a user device on which the music recommendation app is installed, and the music recommendation unit will not recommend the music afterwards if the user's explicit preference for the recommended music is low. Can be.

More specifically, the music recommendation unit includes a self-music recommendation unit that generates recommendation music information from among music that has been experienced by obtaining information about a song that has not been previously recommended while having high preference for music experienced by a user through the history analysis unit; A mate music recommendation unit that generates, as recommendation music information, a song with high preference not previously recommended from the intersection of the common preference song and the user experience song of the mate group selected by the mate selection unit, and the difference set; And a mixed recommendation unit providing one or a mixture of the recommended music information of the self-music recommendation unit and the recommended music information of the mate music recommendation unit as final recommended music information.

The music recommendation service device for discovering music according to another embodiment of the present invention records the meta information and log information about the sound source experienced by the user from the music recommendation app installed on the user device and the external service account to which the user has subscribed. Collecting and analyzing user preferences from the history information, select a mate with a high degree of similarity to the user preferences, but select a new mate group containing a new mate over a preset ratio compared to the previous one, and select the mate group's common preferences. It is characterized by generating recommended music information excluding previously recommended music information for a certain period based on the songs.

According to another embodiment of the present invention, a music recommendation method for discovering music includes meta information on music existing in a user device or streamed from a user device on which the music recommendation service device is installed, and log information on the music. A history collection step of collecting history information; A history analysis step in which the music recommendation service device configures user preference information through history information collected in the history collection step; A step of selecting a mate to select a mate group such that the music recommendation service device selects a mate having high similarity to the user preference information and includes a new mate at a preset rate each time; And a recommendation step in which the music recommendation service device generates and provides recommendation music information based on a common favorite song of a mate group selected in the mate selection step.

In the recommendation step, the previously recommended music information may be excluded for a predetermined period of time, or when a low preference for the corresponding song is specified in the user history of the previously recommended song, the relevant song may be excluded from the recommended music information.

In addition, in the history collection step, the music recommendation service device may collect music history information of the user from a social network service server or an external music service server.

More specifically, the history analysis step further includes a step of scoring preferences for each song by checking explicit preference information from log information of the collected history information, and the recommendation step is performed on the preference score for each song in the history analysis step. Accordingly, the self-recommended music for the preferred song among the songs experienced by the user may be provided as the recommended music information.

Meanwhile, the log information for the sound source collected as the history information in the history collection step may include at least one explicit preference information among play time, play date, play number, skip number, preference selection, sharing, purchase, and deletion. Can be.

A music recommendation system for discovering music according to another embodiment of the present invention includes a user device equipped with a music recommendation app that collects and transmits meta information on a sound source existing in a user device and log information on the sound source as history information; And analyzing user preferences through history information received from the user device on which the music recommendation app is installed, and selecting mates having high similarity with the user preferences, wherein the selected mates group includes new mates at a predetermined rate. And a music recommendation service device that generates recommended music information based on a common favorite song of the mate group and provides it to the music recommendation app of the user device.

In addition, it is preferable that the music recommendation service device manages the same music collectively by matching the music meta information collected as the history information with a standardized identifier or a unique identifier used in other external music services.

In addition, the music recommendation service device selectively scores the preferences according to a predetermined criterion when selecting mates having high user preferences and similarities, performs normalization by classifying levels according to a range of the scores, and then performs the normalization results. Mate similarity can be determined as a criterion.

According to the present invention, user preferences are identified by analyzing user music history information obtained through various paths, and after selecting a mate having a high degree of similarity with the user preferences, recommending music using a common preference song and a user experience song of the mate However, there is an effect of excluding the previously recommended songs to discover new songs of the user's preference through new recommended music each time, or to rediscover the experienced songs.

In particular, as the recommendation is repeated due to the exclusion of the previous recommendation song, a certain percentage of mates among the selected mates are made to be new mates so that music with a low preference rank is not recommended to the user. It has the effect of making it possible.

In addition, to determine the similarity of preference for selecting mates, by applying a scoring method for individual songs, a replay weight scoring method, or a normalization method, it is possible to effectively select mates with similar preferences while reducing the load on mating selection. In addition to individual songs, the preference for artists or genres is further considered, and the volume of the entire appreciation song is also considered, so the accuracy of judging the similarity of preference with various other users can be increased, which increases the user satisfaction with the recommended songs.

In addition, it is also possible to recommend non-repeatable music for songs that you have experienced by using a part of the configuration for selecting mate and creating a mate recommendation song, so users can recommend recommendations for familiar songs and new songs by mate recommendations. Since it can be easily listened to by selecting or mixing, there is an effect of providing a variety of recommended music suitable for the user's taste.

1 is a block diagram showing an outline of a music recommendation system according to an embodiment of the present invention.
2 is a conceptual diagram illustrating a music history collection method of a music recommendation system according to an embodiment of the present invention.
Figure 3 is a block diagram showing the configuration of a music recommendation server according to an embodiment of the present invention.
4 is a conceptual diagram showing a mate selection process of a music recommendation app according to an embodiment of the present invention.
5 is a conceptual diagram illustrating a mate music recommendation method according to an embodiment of the present invention.
6 is a flowchart illustrating an example of a music recommendation method according to an embodiment of the present invention.
7 is a flowchart illustrating an example of a method for providing self-recommended music according to an embodiment of the present invention.
8 is a flow chart showing an example of a mate selection method according to an embodiment of the present invention.
9 is a flowchart illustrating an example of a method for providing mate recommendation music according to an embodiment of the present invention.

The present invention as described above will be described in detail through the accompanying drawings and embodiments. In the detailed description to be described later, a representative embodiment of the present invention for achieving the above-described technical problems will be described. Here, in order to effectively describe the present invention, music meta information, music log information, and music history information are used separately, and the music meta information includes a song name, an artist name, an album name, and a genre included in a sound source. It means information about the sound source itself, and music log information, in addition to information that can directly or indirectly check the user's preferences, such as the date and time of playing the sound source, the duration, playback date, number of playbacks, and number of skips , Information including preference selection information, sharing, purchasing, deletion, etc., that explicitly identifies information about a user's sound source, and the music history refers to such music meta information and music log information. Of course, if the illustrated principle of division is applied, specific information not mentioned above may be further added.

In addition, a server will be described as an example of an embodiment of a service device providing a service, and a terminal will be described as an example of a user device.

On the other hand, an application installed and operated on a user terminal is expressed as an app, but it is noted that this is meant to download and install to a user terminal or to cover an application pre-installed on the terminal.

1 is a configuration diagram showing an outline of a music recommendation system according to an embodiment of the present invention, as shown, the user terminal 100 connects to a music providing system 200 that provides streaming music through the network 10 and desires I can listen to music.

In general, the music service server 210 configured in the music providing system provides a free streaming transmission service for all or part of the selected sound source.

The user can listen to a local sound source stored in the user terminal 100, access the music providing system 200, and listen to music. In addition, the SNS server 20 or another type of external music service server 30 ) To listen to music. Accordingly, reliability may be low when the user's musical taste or preference is simply analyzed based on the sound source of the user terminal 100.

In the embodiment of the present invention, a music recommendation app 110 is installed in the user terminal 100, and meta information (song name, artist name, album, genre, etc.) about the sound source in which the music recommendation app exists in the user terminal The information about the sound source itself) and log information for the sound source (information for knowing the user's preference for the sound source as related to the user's operation related to the sound source) are collected as history information to the music recommendation server 220 to provide. Although the illustrated music recommendation server 220 is shown to be configured separately from the music service server 210 that provides music-related information, this is for clarifying the description of the invention, and the music recommendation server 220 is substantially It may be included as a part of the music service server 210.

The music recommendation server 220 basically grasps a user's preference using history information provided by a music recommendation app installed on the user terminal 100 and then selects and recommends a song having a high preference from among songs experienced by the user. The user may select other users with similar preferences to the user, and then select and recommend the songs preferred by the users.

In an embodiment of the present invention, the music recommendation server 220 analyzes preference information for songs experienced by a user based on the user's history information, obtains a user preference weight for the song, and then sets preferences similar to the user. Eggplant selects other users as a mate in the sense of a music friend, and provides recommended music to the user based on songs commonly selected by the selected mate.

However, in the embodiment of the present invention, in that the musical preference does not change rapidly, a new mate is prescribed every time a mate is selected (a preset period, for example, every day) rather than the existing method in which a similar recommended song is provided every time. Select a mate group to be included as a ratio so that the population for recommendation will remain new, and once recommended songs will not be recommended again until certain conditions (e.g., over a period of time) are satisfied, so that each new recommended music Make information available.

That is, the possibility of discovering new music suited to the user's preference is greatly expanded, and the user's satisfaction with recommendation can be greatly increased by allowing users to re-discover the songs preferred by mates of similar taste among the experiences experienced by the user.

In order to recommend such music, it is necessary to collect and analyze various preferences of songs according to user preferences. To this end, the music recommendation app 110 may inquire subscription information regarding a social network service or other external music service to which the user has subscribed during installation, and the music recommendation server 220 is a user obtained through the music recommendation app Through the subscription information, the music history information of the user may be further collected from the social network service server 20 or the external music service server 30 and used as information for determining user preference.

FIG. 2 is a conceptual diagram for explaining how the music recommendation server 220 collects history information to determine a user's preference according to an embodiment of the present invention. First, the music recommendation server 220 is a user terminal 100 Through the music recommendation app installed in), it continuously collects meta information about the sound source and retained information, and log information about the sound source (including retained or deleted sound sources).

In addition, the social network service server 20 having a function related to sound source playback and an external linkage function collects meta information and usage log information (posting log, etc.) about the sound source used by the user, and the external music service server 30 ) Can be linked with the outside (for example, when providing an open API), collects meta information and usage log information of a sound source that has been listened to from the corresponding music service server 30. In this case, the music recommendation server 220 may collect accumulated sound source meta information and log information from the social network service server 20 and the external music service server 30 before the music recommendation app is installed. Through this, the user's preference information can be known in relatively detail.

Of course, when the user installs the music recommendation app, the user may not use a social network service, may not have subscribed to an external music service, or may have poor log information collected or accumulated even after using or subscribing to such a service. In order to grasp the original music propensity, the user may be asked to collect predetermined taste information, and the user may be asked to listen to a plurality of music of a certain style and select preferences for them to collect tastes. However, even if there is no such process, it is not necessary to collect such initial manual preference information because it is possible to make recommendations suitable for user preferences by gradually collecting the history including the user's explicit preference information for recommended music information according to a predetermined criterion.

Meanwhile, in the embodiment of the present invention, music history information may be collected through various channels. As a specific example, the user terminal can collect log information and is an Android-based terminal or an iOS-based terminal having an externally linked music app. Last, FM (Last.fm), which can be linked with Facebook as a social network service that has functions related to music playback and supports external interoperability, and supports external interoperability as an external music service. Can be linked with. In addition, in the case of a music recommendation app, it is possible to generate log information for a retained sound source and a streaming sound source while supporting a playback function.

This is classified by collection source and summarized as an example, as shown in Table 1 below.

Collection source

possesion
Playback date
/ Play time
Play count
Skip count
preference
share
age
purchase iTunes Music Player

●
Last
Playing time
●
●
●



Music recommendation app

●
●
●
●
●
●
Upon inquiry
● Android Music
Player


●
●





Facebook


●
●



●
Last.fm


●
●

That is, for music recommendation, the music recommendation server 220 can effectively grasp user preferences by collecting user music histories through multiple paths, and collect user's explicit preference information for individual songs.

The user's explicit preference information can be checked from the log information generated by the music recommendation app or the log information included in the collected history information. The explicit preference information includes play time, play date, play count, and skip count. , Preference selection information, sharing, purchase, and deletion information may be used.

That is, through listening to a predetermined sound source or a predetermined music (same music, although the location or playback path of the sound source is different) for a certain period of time, it can be seen that the user is interested in the music, and it is possible to check when the music was played through the playback date and time, You can see the interest level more clearly through the number of replays, and if you select 'Like' explicitly, you can see that the preference is quite high. In addition, if the music is shared, it can be seen that the preference for the music is high, and if the music source is purchased, the preference is extremely high. In addition, if you skip after listening to a part of the music, you can see that the preference for the sound source is low, and if you select 'Dislike' or delete the sound source, you can see that the preference is very low. .

On the other hand, since the music recommendation server 220 must integrate and manage information on various pieces of music collected through a plurality of channels, the music history collected in this way is a history with at least basic meta information such as a song name or an artist name. It is desirable to collect only.

In addition, it is necessary to standardize music meta information obtained through various routes as described above for the determination of integrated preference. Therefore, if necessary, the music recommendation server utilizes an external music DB (for example, iTunes, AMG, The Echo Nest, etc.) having vast information about an individual sound source, and a sound source identifier (Song ID) for a sound source for which history is collected. ) Can be matched to standardize or correspond to an identifier of a sound source used in another commercial or famous music service server. Of course, it is preferable that the music recommendation server manages the standardized identifier information or matching information of the sound source as a separate DB, and standardizes the identifier by querying the external music DB only for sound sources that do not exist in the DB, and other external music. Matching information with a unique identifier used in the service can be configured.

Through this standardization, even if different sound sources are the same music, it is determined that they are the same, and the preference information can be accumulated and managed collectively.

Thereafter, the music recommendation server 220 recommends appropriate music to a corresponding user based on the collected and analyzed user preference information. Self-recommended music that provides a preferred music among users' experiences, and a similar tendency to the user Mate recommendation music based on a common preference song of mates having or a mixture of these may be provided to the music recommendation app of the user terminal 100 as recommendation music information.

Figure 3 shows the configuration of the music recommendation server 220 according to an embodiment of the present invention, as shown, to the music meta information database 201 and members having metadata and sound source identification information and matching information about the sound source Member database 202, which stores subscription information, preference information, mate group information, and recommended music information, and the recommended music database 203 that manages information about recommended music, can be linked to an external server for collecting history B. It can be interlocked with an external interlocking server, such as an external server to acquire information about the sound source.

The music recommendation server 220 includes a member management unit 221 for managing member information acquired through the music recommendation app and a history collection unit for collecting historical information about music experienced by a user, as shown in the illustrated embodiment. 222), a history analysis unit 223 that analyzes the collected history and configures user preference information of at least one of individual songs, artists, and genres, and user preference information obtained through the history analysis unit 223. The similarity determination unit 224 which selects a mate by calculating the similarity to the common history by using the user's preference information as a weight, and the mate management unit 225 that manages information on a group of mate groups for each user, and the similarity Through the mate similarity through the determination unit 224 and the mate history of the mate management unit 225, at least a predetermined ratio or more of the new mate is included compared to the previous mate group. And a mate selection unit 226, and a music recommendation unit 227 is to generate the music like mate selection unit 226 is based on the common preference for the selected song mate selecting a new formate group.

The music recommendation unit 227 is a self-music recommendation unit that generates recommendation music information from music that has been experienced by obtaining information about a song that has not been previously recommended while having high preference for music experienced by the user through the history analysis unit 223 , The mate music recommendation unit and the self-music for generating a previously unrecommended high-favorite song as recommended music information from the intersection and difference between the common preference song and the user experience song of the mate group selected by the mate selection unit 226 And a mixed recommendation unit providing one or a mixture of recommended music information of the recommendation unit and recommended music information of the mate music recommendation unit as final recommended music information.

In the illustrated configuration, the history collection unit 222 continuously collects the history of music experienced by the user through various paths as described in FIG. 2, and the collected history is basically a history analysis unit 223 ). Basically, the user preference of the corresponding song can be calculated through a scoring method through explicit preference for a specific song.

For example, as shown in Table 2 below, 2 points are added for each playback, and 10 points are added if 'Like' is explicitly selected. You can add 5 points multiplied and 20 points when you purchase the song. In addition, if the song is skipped within a preset play time, it will be scored as -10 points when skipped, -20 points when explicitly selecting 'Dislike', -20 points when deleting the song, etc. Can be. Of course, the specific score can be variously determined.

Preference score item Add song
(Add) good
(Like) I do not like it
(Dislike) evaluation
(Rate) play
(Play) Skip
(Skip) purchase
(Buy) delete
(Delete) One 10 -10 Rating * 5 2 -10 20 -20

The music recommendation unit of the music recommendation unit 227 may recommend the music to the user based on the preference score for each song analyzed by the history analysis unit 223. The old music may not be recommended again until a predetermined condition is satisfied. For example, once recommended songs are not recommended for one week, the exclusion period is variable according to the user's music activity, or if there are no more songs to recommend above the preset preferences, various conditions can be recommended. Can be applied.

Meanwhile, the recommended song may be further filtered using additional preference information such as user age information, style (genre), mood, tempo, similarity, and artist.

The self-music recommendation can be expected to have the effect of allowing the user to easily discover and listen to his / her favorite songs by allowing the user to newly recommend his / her favorite songs from among his / her experiences.

In addition to the self-music recommendation, the music recommendation server 220 according to an embodiment of the present invention increases the user's recommendation music satisfaction through mate music recommendation, and the mate music recommendation is selected by selecting a mate group having similar preferences to the user. It is based on the group's common favorite music. Here, in the embodiment of the present invention, when selecting a mate group that is a population for recommendation by selecting a mate having a similar preference with a user, a new mate different from the existing mate in a certain percentage of mates composing the mate group each time (periodically) Make sure that the population for recommendation is always new. In addition, the recommended music through the mate group also excludes the previous recommended music until a certain condition is satisfied so that new music can be recommended.

That is, since the musical preference does not change rapidly, when a user selects a mate similar in preference to the user, the object may be mostly similar. In the embodiment of the present invention, the preferences as well as specific mates having high similarity to the preferences are not limited. You can also diversify the highly recommended music items by including the surrounding mates that are slightly lower than the standard in the population for recommendation, and new music is provided every time, so users can find the music they like. You will be able to provide an opportunity to have unexpected fun through this recommended music information. For example, new music can be recommended based on a different population each time by selecting a mate group such that 30% of the mates are new every day.

If this is described with reference to FIG. 4, as shown in FIG. 4A, through the analysis result of the history analysis unit 223, the music history 301 experienced by the user and the history 302 of another user are similar to the similar determination unit 224 ) Compares to calculate similarity for each.

Using the calculated similarity, the mate selection unit 226 may select (303) K mates having a high similarity among other user histories 302, as shown in FIG. 4B, the existing mate selection history of the mate management unit 225 ( 310), instead of a previously selected mate (such as a mate of a lower similarity or a mate with a lot of repeated history), a new mate (a mate having a higher similarity) is selected, and this new mate has a preset ratio (for example, 30). %) To select a mate group 320 containing a certain percentage (or number) of new mates compared to the previous one.

 The similarity determination of the similarity determination unit 224 is simply based on the number of music common among the music of the self-history 301 and the music of other user histories, or the genre, artist, mood, tempo, and similar pattern songs in the self-history. Although it may be based on the distribution of Korean music and the distribution of music in other user histories, in the embodiment of the present invention, the preference score (for example, a table) obtained for each user's favorite songs in the history analysis unit 223 in order to effectively determine similarity 2) can be used as a weight to determine the similarity between user histories.

That is, it may be determined that other user histories having a song having a high weight among the user histories in common have high similarity. Of course, it is possible to use the preference scores of individual songs, but for quick mate selection, the similarity can be determined by simply scoring the number of replays separately as a weight.

In the case of simply scoring the number of replays, the method shown in Table 3 below can be used.

Replay weight score item 3 times or less 4 ~ 9 times 10 ~ 50 times 50 or more times One 2 3 4

On the other hand, the similarity of other users who have experienced the same number of users' favorite music can be additionally adjusted. For example, considering the total number of songs enjoyed by others, the total number of songs enjoyed is less than the number of users having many songs. The user's similarity can be judged higher. That is, in the case of n (A∩B) = n (A 에 C) for the history of users A and B, C, if n (A∪B)> n (A∪C), the history of C is higher than B. It can be judged that the similarity is high.

In addition to these common songs, similarity can be calculated based on a common artist, and similarity can be calculated based on a common genre. Since the likelihood of matching the taste is slightly lower than that of a song-based, song-based, artist-based, and common genre-based Similarity can be determined in order, and after selecting a song-based similarity mate, it can be filtered using artist or genre (or mood, beat, etc.) similarity. In addition, when deriving similarity, if the number of common songs is the same, it may be determined that the similarity is higher as it overlaps with the user's preferred genre considering the genre of songs excluding the common song.

Previously, we looked at how to use weights for songs and weights for the whole listening song to calculate similarity, and how to use additional similarity judgments for artists or genres in addition to songs. Preference for each song for faster mate selection It is also possible to use a result normalized using a score or a reproduction weight score. Table 4 below divides the level for the song unit preference or reproduction weight, the artist unit level, and the genre unit level into relatively large groups, so that when using such level information, the mate selection load can be reduced. The normalization number may be determined in consideration of the average number of users watching, the number of artists, and the number of genres.

Song unit artist  unit Genre unit level Song point Song Point Total Song Point Total 2 or less 2 or less 2 or less One 3-5 2 ~ 10 11 ~ 100 2 5 ~ 10 11-20 101 ~ 300 3 11-20 21-50 301 ~ 500 4 21 or more 51 or more 501 or more 5

As described above, the mates selected through the similarity determination described above select the final mate group by making a mate of a predetermined ratio a new mate, and the number of mates in the mate group is fixed to K (or set by the user). The similarity of the selection criteria may be adjusted to be the corresponding K, or the results of similarity determination at a sequential lower level may be reflected in the order of song criteria, artist criteria, genre criteria, and the like.

5 is a conceptual diagram for explaining a method for generating recommended music information using a mate group when a mate group is selected as a result of similarity determination described through FIG. 4, based on a common preference song of mates belonging to the mate group, as shown in FIG. It has been shown that the recommended music can be extracted.

The common preference song of mates is basically based on the intersection 330 of the songs (music) included in the mates' history, the higher the preference score sum or average among the songs in the intersection (or the sum of playback weight scores or The higher the average, the higher the ranking of recommendations.

The songs included in the mate intersection 330 include songs experienced by the user and songs not experienced by the user, of the difference 331 and intersection 332 of the self history 301 and the mate intersection 330 Recommended music can be determined based on the content.

The difference 331 of the self-history 301 and the mate intersection 330 is music that the user has not experienced among common preference songs of the mate group, and among them, new music is found in the order of songs with high preference (mate preference) Recommended music information for can be generated. Here, among the recommended music, music of a genre such as a user's preferred genre may be first recommended. Of course, the previously recommended music may be excluded from re-recommendation until a predetermined criterion is satisfied.

The self-history 301 and the intersection 332 of the mate intersection 330 are music that the user has experienced among common preference songs of the mate group, and among them, rediscovery of existing music in the order of songs with high preference (mate preference) Recommended music information for is generated. Of course, the previously recommended music may be excluded from re-recommendation until a predetermined criterion is satisfied.

Therefore, by appropriately setting the ratio of recommended music information for rediscovering existing music and recommended music information for rediscovering existing music according to taste, it is possible to discover newly suited music each time, and rediscover familiar songs. have.

On the other hand, the user's log information is collected even when the recommended music is played through the recommended recommended music information. According to the user's explicit preference for the recommended music, the re-recommendation criteria for the recommended music is adjusted or the corresponding music is You can also make it no longer recommended. For example, when skipping while playing a recommended music, the music can no longer be recommended, and if 'Like' is selected while playing the recommended music, the period for excluding re-recommendation of the music can be reduced.

6 to 9 is a flow chart showing an exemplary embodiment of music recommendation according to an embodiment of the present invention described above, Figure 6 is a flow chart of the overall recommendation music providing method, Figure 7 is the experience included in the user's history 8 is a flowchart of a method for recommending magnetic music for a song, FIG. 8 is a flowchart for a method for selecting a mate, and FIG. 9 is a flowchart for a method for recommending a mate music.

First, referring to FIG. 6, installing a music recommendation app on a user terminal (S10), and past accumulated history information and user history information after installation are collected through the music recommendation app and transmitted to a music recommendation server Step (S11), to standardize (e.g., SongID) or match the meta-information of the music history to quantitatively analyze the music history, and standardize the music identification (S12), and analyze the user's history information By calculating the preference score for each song and calculating the sum of preference scores for an artist or genre, if necessary, a self-history analysis step (S13) of constructing user preferences, and based on the results of the self-history analysis, the user's and other user's similarities are selected for each song. After selecting a mate with high similarity by reflecting the score (Table 2) or the replay weight score (Table 3), the new mate has a certain ratio. A mate selection step (S14) for selecting a mate group included above, a self-music recommendation according to the self-music preferences analyzed in the self-history analysis step, and a mate based on a common preference song of the mate group selected in the mate selection step It consists of a music recommendation step (S15) in which one or more of the music recommendations are made.

The past / continuous history information collecting step (S11) may collect user music history through a social network service having an external linkage function or an external music service in addition to a user terminal in which a music recommendation app is installed.

FIG. 7 shows a method for self-music recommendation in the music recommendation step (S15) of FIG. 6, the collecting history confirmation step (S21) and the preference history scoring for each song shown in the magnetic history analysis step (S13). S22 may be performed, and other steps may be performed in the music recommendation step S15.

Referring to the illustrated flowchart, first, collecting history confirmation step (S21) of checking user history information of various paths collected and arranging histories for each song, and using the history information for each song, explicit preference information for the song is exemplified. For example, it includes a preference history scoring step for each song (S22) for calculating the score for each song by checking using Table 2.

The preference for each song is scored as an explicit score in the preference-scoring step for each song (S22). Using this score for each song, an accumulated artist score or a cumulative score for each genre may be further obtained. When selecting a recommended song, it can be used as reference information.

In order to self-refer to recommend a preferred song among his songs to the user based on the history information about the song that the user has experienced, the candidate song selection step of selecting the recommended candidate songs in the order of highest score for each song (S23) And, identifying the songs (or songs whose exclusion condition has not been resolved) within a preset period among the selected candidate songs and excluding the corresponding songs (S24, 25), and self-recommended songs as recommendable songs A self-recommended song providing step (S26) of providing music recommendation information by generating the may be performed.

Whether or not to include the self-recommended song in the recommended music information may be determined according to a user's selection, and may be mixed with a mate recommended song as necessary to configure recommended music information.

Even if the self-recommended song as described above is recommended once, the self-recommended song recommended within a predetermined period is different every time because it is not recommended again until a certain condition is satisfied, so that the user can easily listen to his or her favorite music. can do.

8 shows a process of selecting a music friend, ie, a mate, who has similar preferences to himself, by utilizing the results of the self-history analysis performed for each user in the music recommendation server.

As shown in the figure, a music propensity information configuration step (S31) of preparing historical analysis information as a reference for mate similarity determination, and using the historical analysis information to obtain a playback weight score for each song or extracting a preference score for each song, as necessary In this case, based on this, the song regeneration-based weighting step of performing normalization (S32) and the preference score or replay weight score or normalization level for each song obtained for each song are used, and the total number of appreciation songs, artist or genre similarity, and common songs are excluded. A similarity calculation step (S33) between a user and a mate that calculates the similarity to another user in consideration of the genre matching degree of a song, and a higher mate selection step (S34), which selects a mate having a high level of similarity to a predetermined level, Check if new mates are included at a preset ratio among the selected mates. Sublethal to a step (S35, S36, S37) to be replaced by new mates select formate group.

In the step of replacing the new mate (S36), a new mate may be selected by replacing the previously selected mate with a mating not previously selected while having a similar similarity while having a low similarity, and the corresponding mate is selected in order of the number of times the previous mate is selected frequently. Instead, you could choose a mate that wasn't selected, with similarity to the next level.

That is, if the new mate is set to 30% for each update of the mate group (for example, every day), the population that will provide the recommended music to the user will always be able to maintain a certain part of the newness.

9 is an example showing a process of generating a mate recommendation music from a corresponding mate group when a mate group is selected, as shown, extracting a common song between mates from the mate group (S41), and the mate common song and the user By extracting the difference set and intersection with the user song experienced by the user, the mates with similar inclinations have experienced common experiences, but the user selects new songs that have not been experienced as the difference set, and the preferences commonly experienced by the mates and users with similar inclination The step of selecting a high song as an intersection (S42), and among the songs included in one or more sets of the selected difference or intersection, a mate preference (preferred weight score for each song or a reproduction weight score) sum or a song having a high average is a candidate for recommendation Step (43) for selecting songs, and other criteria within a set period (or a preset period) Excluding recommendations) (S44, S45), and mate recommendation song provision step (S46) of generating and providing a mate recommendation song with a song different from the existing recommendation song among the recommended candidate songs. .

Since the user's musical preferences are not changed much, selecting a highly recommended mate each time to select a highly recommended recommendation song will make most of the recommended songs similar, and if you simply exclude the previously recommended song from this recommendation, the recommendation will be repeated The more it becomes, the farther away from the user's preference, or the lack of a common mate song may occur. However, in the embodiment of the present invention, the mate in the mate group, which is a population for selecting the recommended songs, is changed at least at a certain rate every time, thereby diversifying the common songs for the mates close to the user preference, and accordingly, the previous recommended songs are fixed for a certain period or Even if it is excluded according to certain conditions, it is possible to increase the likelihood that a recommended song that is differently recommended each time matches the user's preference.

As a result, despite providing new recommendation results every time, similar music is recommended to the user's preferences, so that new music can be discovered or rediscovered experiences, thereby increasing user satisfaction.

In particular, explicit preference information is also provided for recommended music, and since the information is continuously collected from the music recommendation server, the grasp of preference through the recommended music is rapidly accumulated, and if the recommended music is skipped, the music is not recommended afterwards Since the recommendation result can be directly manipulated, the satisfaction with the recommended music can be further increased according to use.

In the above, preferred embodiments according to the present invention have been illustrated and described. However, the present invention is not limited to the above-described embodiments, and various modifications can be made to anyone having ordinary knowledge in the technical field to which the present invention pertains without departing from the gist of the present invention attached in the claims. .

The present invention analyzes user music history information, grasps user preferences, selects a mate with a high degree of similarity to the user preferences, and recommends music using the mate's common preference songs and user experience songs. Excluded by allowing users to discover new songs of their preference through new recommended music each time, or to rediscover the songs they have experienced. It can be widely used in the service provision field, music social network service provision field, telematics field where streaming music service is introduced, media advertising field, media sales and diffusion field.

10: network 20: SNS server
30: external music service server 100: user terminal
110: music recommendation app 200: music service system
210: music service server 220: music recommendation server
221: member management unit 222: history collection unit
223: history analysis unit 224: similarity determination unit
225: Mate management unit 226: Mate selection unit
227: music recommendation department

Claims (13)

A history analysis unit for analyzing user preferences by collecting meta information about sound sources existing in the user device and log information for the sound sources as history information through a user device on which the music recommendation app is installed;
A mate selection unit for selecting a new mate group having a high degree of user preference and similarity and containing a new mate at least a predetermined ratio or higher than a previous mate group; And
Includes a music recommendation unit that generates new recommended music information different from previously recommended music information based on common preference songs of the mate group selected by the mate selection unit and provides the music recommendation app to the installed user device.
The mate selection unit calculates the similarity by comparing the music in the user's history information and the music in another person's history information, applies weights according to the preferences in the history information of the music, and applies the similarity according to the number of songs viewed. Adjust,
Select a mate with high similarity based on a preset criterion from the history information of the other person, and select the new mate above a preset ratio instead of a previously selected mate using the user's history information among the selected mate, Music recommendation service device for music discovery, characterized by selecting the new mate group containing the new mate. According to claim 1, The history analysis unit collects the user's music history information from a social network service server or an external music service server to which the user subscribes by receiving user subscription information through a user device on which the music recommendation app is installed. Music recommendation service device for discovering music, characterized by analyzing preferences. The music for music discovery of claim 1, wherein the music recommendation unit selects recommended music information from a difference or intersection between a common preference song of the selected mate group and a user experience song of the collected history information. Recommended service device. delete The method of claim 1, wherein the history analysis unit collects a user history for the recommended music from a user device on which the music recommendation app is installed, and the music recommendation unit plays the music when the user's explicit preference for the recommended music is low. Music recommendation service device for discovering music, which is not recommended afterwards. The method of claim 1, wherein the music recommendation unit
A self-music recommendation unit that generates recommendation music information from music that has been experienced by acquiring information on songs that have not been previously recommended while having high preference for music experienced by the user through the history analysis unit;
A mate music recommendation unit that generates, as recommendation music information, a previously unrecommended high preference song from the intersection and difference between the common preference song and the user experience song of the mate group selected by the mate selection unit; And
Music recommendation service device for music discovery, characterized in that it comprises a recommendation information of the music recommendation of the self-music recommendation unit and the recommendation music information of the mate music recommendation unit or a mixture of the recommendation information as the final recommendation music information. A music collection service step in which the music recommendation service device collects meta information on music existing in the user device or streamed from the user device on which the music recommendation app is installed and log information on the music as history information;
A history analysis step in which the music recommendation service device configures user preference information through history information collected in the history collection step;
A step of selecting a mate to select a mate group such that the music recommendation service device selects a mate having high similarity to the user preference information and includes a new mate at a preset rate each time; And
The music recommendation service device includes a recommendation step of generating and providing recommendation music information based on a common favorite song of a mate group selected in the mate selection step,
In the selecting of the mate, the similarity is calculated by comparing the music in the user's history information and the music in another person's history information, but applying weights according to preferences in the history information of the corresponding music, and similarity according to the number of songs viewed. Adjust the
A mate with high similarity is selected based on a preset criterion among the other person's history information, and a new mate having a predetermined ratio or higher is selected instead of a previously selected mate using the user's history information among the selected mate. Music recommendation method for discovering music, characterized by selecting a new group of mates that contain mates. The method of claim 7, wherein the recommendation step excludes the song from the recommended music information when the previously recommended music information is excluded for a preset limit period or when a low preference for the song is specified in the user history of the previously recommended song. Music recommendation method for discovering music, characterized in that. The music recommendation method of claim 7, wherein in the history collection step, the music recommendation service device collects music history information of the user from a social network service server or an external music service server. The method of claim 7, wherein the history analysis step further comprises the step of scoring the preferences for each song by checking explicit preferences from log information of the collected history information, and the recommendation step comprises the preference scores for each song in the history analysis step. A music recommendation method for discovering music, characterized in that self-recommended music for a preferred song among the songs experienced by a user is provided as the recommended music information. The log information of the sound source collected as the history information in the history collection step is at least one explicit preference among play time, play date, play number, skip number, preference selection, sharing, purchase, and deletion. Music recommendation method for discovering music, characterized by including information. A user device having a music recommendation app installed to collect and transmit meta information on a sound source existing in the user device and log information on the sound source as history information; And
The user preference is analyzed through the history information received from the user device on which the music recommendation app is installed, and the mates having high similarity with the user preference are selected, but the mate group that is selected each time includes a new mate at a predetermined rate, and the And a music recommendation service device that generates recommendation music information based on a common preferred song of the mate group and provides it to the music recommendation app of the user device.
The music recommendation service device calculates the similarity by comparing music in the user's history information and music in another person's history information, and applies weights according to preferences in the history information of the corresponding music, and similarity according to the number of songs viewed. Adjust the
A mate with high similarity is selected based on a preset criterion among the other person's history information, and a new mate having a predetermined ratio or higher is selected instead of a previously selected mate using the user's history information among the selected mate. A music recommendation system for discovering music, characterized by selecting a new mate group containing mate. The method of claim 12, wherein the music recommendation service device scores the preferences according to a preset criterion when selecting mates having a high degree of user preference and similarity, and performs normalization after classifying levels according to a range of the corresponding scores. Music recommendation system for discovering music, characterized by determining the mate similarity based on the results.

Images