A recommendation method and recommendation system based on user implicit feedback behavior

Abstract

The present invention discloses a recommendation method and a recommendation system based on implicit user feedback behavior, wherein the method comprises obtaining target user behavior log data and reference user behavior log data; obtaining reference user-reference song list-reference behavior-reference score vector table and target user-sample song list-sample behavior-sample score vector table according to a preset scoring rule, and calculating the similarity value between each reference user and the target user one by one, synthesizing the first K reference users with the highest similarity value with the target user into a similar user set; predicting the target user-reference song list-quasi-score vector table according to the similar user set and the reference user-reference song list-reference behavior-reference score vector table, and generating a recommendation list according to the target user-reference song list-quasi-score vector table. The present invention can form a song recommendation list according to the implicit feedback data in the behavior log data of the target user and the reference user.

Description

Technical Field

[0001] The present invention relates to a recommendation method and a recommendation system based on implicit feedback behavior of users, belonging to the technical field of big data recommendation. Background Art

[0002] The recommendation method is to analyze users' historical data, explore users' potential needs, and recommend information, thereby improving people's efficiency in obtaining information and effectively alleviating the increasingly serious problem of information overload.

[0003] Existing recommendation methods and recommendation systems recommend content to users based on the user's displayed ratings of the content and analyzing the user's interests and hobbies. In actual applications, it is difficult to obtain the user's ratings of songs, which makes it difficult to apply recommendation methods and recommendation systems in music.

[0004] Therefore, the present application proposes a recommendation method and a recommendation system based on user implicit feedback behavior to solve the above technical problems. Summary of the invention

[0005] The purpose of the present invention is to overcome the deficiencies in the prior art and to provide a recommendation method and a recommendation system based on user implicit feedback behavior, which can form a song recommendation list according to the implicit feedback data in the behavior log data of the target user and the reference user.

[0006] To achieve the above object, the present invention is implemented by adopting the following technical solutions:

[0007] In one aspect, the present invention provides a recommendation method based on user implicit feedback behavior, comprising the following steps:

[0008] Obtain target user behavior log data, and statistically analyze the target user behavior log data to obtain a target user-sample song list-sample behavior vector table;

[0009] Obtain reference user behavior log data, and statistically analyze the reference user behavior log data to obtain a reference user-reference song list-reference behavior vector table;

[0010] According to the preset scoring rules, the reference user-reference song-reference behavior vector table is converted into a reference user-reference song list-reference behavior-reference score vector table;

[0011] According to the preset scoring rules, the target user-sample song list-sample behavior vector table is converted into a target user-sample song list-sample behavior-sample score vector table;

[0012] According to the reference user-reference song list-reference behavior-reference score vector table and the target user-sample song list-sample behavior-sample score vector table, the similarity value between each reference user and the target user is calculated one by one, and the first K reference users with the highest similarity value to the target user are synthesized into a similar user set;

[0013] According to the similar user set and the reference user-reference song list-reference behavior-reference rating vector table, the target user-reference song list-pseudo-rating vector table is predicted, and a recommendation list is generated according to the target user-reference song list-pseudo-rating vector table.

[0014] Furthermore, the intersection of the reference playlist and the sample playlist of each reference user is a non-empty set.

[0015] Furthermore, the integration rules include:

[0016] In response to any reference behavior of any reference user on any reference song in the reference playlist, the reference score of the corresponding reference song of the corresponding reference user is updated according to the following formula:

[0017]

[0018] s play =λ u *num play *p play

[0019] s collect =num collect *p collect

[0020] s add =num add *p add

[0021] s comment =num comment *p comment

[0022] S u =s play *w play +s collect *w collect +s add *w add +s comment *w commet

[0023] Among them, λ u is the reference playback effectiveness factor, s play For reference, play pseudo-rating, num play is the reference playback number, p playis the reference playback score, s collect For reference collection pseudo rating, num collect is the reference number of favorites, p collect For reference, the collection points value, s add To add pseudo-ratings to the reference playlist, num add To add the reference playlist times, p add To add the reference song list points, s comment is the pseudo score of the reference review, num comment is the number of reference comments, p comment S is the reference comment score. u is the reference score, w play is the reference playback weight factor, w collect is the reference collection weight factor, w add To add the reference playlist weight factor, w commet is the reference comment weight factor.

[0024] Furthermore, the integration rules include:

[0025] w play +w collect +w add +w commet =1

[0026] w commet =max{w play , w collect , w add , w commet}

[0027] p play <p collect <p add <p comment .

[0028] Furthermore, the integration rules include:

[0029] In response to any sample behavior of the target user on any sample song in the sample playlist, the sample score of the corresponding sample song is updated according to the following formula:

[0030]

[0031] s` play =λ` u *num` play *p` play

[0032] s` collect =num` collect *p` collect

[0033] s` add =num` add *p` add

[0034] s comment =num comment *p comment

[0035] S a =s` play *w` play +s` collect *w' collect +s' add *w' add +s' comment *w' commet

[0036] Among them, λ' u is the sample playback effective factor, s` play Play pseudo score for the sample, num` play is the number of sample playbacks, p` play is the sample playback score, s' collect Collect pseudo scores for samples, num` collect is the number of sample collections, p' ddlect is the sample collection score, s' add To add pseudo ratings to the sample playlist, num` add is the number of times the sample playlist is added, p' add To add the sample playlist points, s' comment Pseudo rating for sample reviews, num` comment is the number of sample comments, p' comment is the sample review score, S a Score the sample, w` play is the sample playback weight factor, w` collect is the sample collection weight factor, w` add To add the sample playlist weight factor, w' commet is the sample review weight factor.

[0037] Furthermore, the integration rules include:

[0038] w` play +w` collect +w` add +w` commet =1

[0039] w` commet =max{w` play ,w` collect ,w` add,w` commet}

[0040] p` play <p' collect <p' add <p' comment .

[0041] Furthermore, the similarity value between each reference user and the target user is calculated one by one by the following formula:

[0042]

[0043] Among them, sim(a, b) is the similarity value between the reference user and the target user, a is the target user, s a，i The sample score of song i given by the target user, I a，b is the intersection of the reference playlist and the sample playlist, s b，i is the reference user's reference rating for song i, U 1 is the reference user set;

[0044] The first K reference users with the highest similarity values ​​to the target user are synthesized into a similar user set.

[0045] Further, the predicting of the target user-reference song list-pseudo-score vector table according to the similar user set and the reference user-reference song list-reference behavior-reference score vector table includes:

[0046] The target user's proposed rating for each reference song in each reference playlist is predicted one by one using the following formula:

[0047]

[0048] Among them, sim(a,u′) is the similarity value between the similar user and the target user, l is the reference song in the reference playlist, S u′，l is the reference score of similar users for the reference song l, C a，l is the target user's proposed rating for the reference song l, and U is the set of similar users.

[0049] Further, generating a recommendation list according to the target user-reference song list-presumed rating vector table includes:

[0050] Remove the intersection of the reference playlist and the sample playlist in the reference playlist to obtain a recommended playlist;

[0051] According to the proposed ratings given by the target users to the recommended songs in the recommended playlist, the top N recommended songs with the highest proposed ratings are selected to generate a recommendation list.

[0052] On the other hand, the present invention provides a recommendation system based on user implicit feedback behavior, comprising:

[0053] The target data module is used to obtain the target user behavior log data, and statistically analyze the target user behavior log data to obtain the target user-sample song list-sample behavior vector table;

[0054] A reference data module is used to obtain reference user behavior log data, and statistically analyze the reference user behavior log data to obtain a reference user-reference song list-reference behavior vector table;

[0055] A reference points module, used to convert the reference user-reference song-reference behavior vector table into a reference user-reference song list-reference behavior-reference score vector table according to a preset points rule;

[0056] The sample points module is used to convert the target user-sample song list-sample behavior vector table into the target user-sample song list-sample behavior-sample score vector table according to the preset points rule;

[0057] A similar user module is used to calculate the similarity value between each reference user and the target user one by one according to the reference user-reference song list-reference behavior-reference score vector table and the target user-sample song list-sample behavior-sample score vector table, and synthesize the first K reference users with the highest similarity values ​​to the target user into a similar user set;

[0058] The recommendation module is used to predict the target user-reference song list-pseudo-rating vector table based on the similar user set and the reference user-reference song list-reference behavior-reference rating vector table, and generate a recommendation list based on the target user-reference song list-pseudo-rating vector table.

[0059] Compared with the prior art, the present invention has the following beneficial effects:

[0060] The present invention obtains similar users with high similarity values ​​to the target user by analyzing implicit feedback data in the behavior log data of the target user and the reference user, so as to infer the target user's preference for songs, thereby forming a recommendation list and realizing the music recommendation function for the user. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 Shown is a flow chart of an embodiment of a recommendation method based on user implicit feedback behavior of the present invention. DETAILED DESCRIPTION

[0062] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

[0063] Example 1

[0064] The present invention provides a recommendation method based on user implicit feedback behavior, comprising the following steps:

[0065] S1 obtains the target user behavior log data, and statistically analyzes the target user behavior log data to obtain a target user-sample song list-sample behavior vector table;

[0066] S2 obtains reference user behavior log data, and statistically analyzes the reference user behavior log data to obtain a reference user-reference song list-reference behavior vector table;

[0067] S3 converts the reference user-reference song-reference behavior vector table into a reference user-reference song list-reference behavior-reference score vector table according to a preset scoring rule;

[0068] S4 converts the target user-sample song list-sample behavior vector table into a target user-sample song list-sample behavior-sample score vector table according to a preset scoring rule;

[0069] S5 calculates the similarity value between each reference user and the target user one by one according to the reference user-reference song list-reference behavior-reference score vector table and the target user-sample song list-sample behavior-sample score vector table, and synthesizes the first K reference users with the highest similarity values ​​to the target user into a similar user set;

[0070] S6 predicts the target user-reference song list-pseudo-rating vector table based on the similar user set and the reference user-reference song list-reference behavior-reference rating vector table, and generates a recommendation list based on the target user-reference song list-pseudo-rating vector table.

[0071] In the application, the target user behavior log data includes the target user's basic information, the songs that have been historically operated on, and the corresponding behaviors; the reference user behavior log data includes the basic information of multiple reference users, the songs that have been operated on by each reference user, and the corresponding behaviors. Among them, behaviors include playing songs, collecting songs, adding songs to playlists, and commenting on songs.

[0072] When this embodiment is applied, the intersection of the reference song list of each reference user and the sample song list of the target user is a non-empty set, that is, the target customer has performed behavioral operations on the same song with each reference user.

[0073] At present, most users are unwilling to actively rate songs, which makes it difficult to guess user preferences. The present invention obtains similar users with high similarity values ​​to the target user by analyzing the implicit feedback data in the behavior log data of the target user and the reference user, so as to infer the target user's preference for songs, thereby forming a recommendation list and realizing the music recommendation function for the user.

[0074] Example 2

[0075] Based on Example 1, this example introduces in detail the scoring rules, the similarity value calculation method and the recommendation list formation method.

[0076] (I) Points Rules

[0077] In order to explain the impact of each behavioral operation of the reference user or the target user in the process of selecting songs, this embodiment sets a scoring rule to quantify each behavior of the reference user or the target user, that is, different scoring values ​​are set for the four behaviors of playing, collecting, adding to playlists, and commenting. When the reference user or the target user performs any of the above behavioral operations on any song, the reference score or sample score of the corresponding song is updated.

[0078] In the application, the reference user or target user will comment on a song only if he or she really likes it. Therefore, different impact factors are set for different behaviors based on the probability of the four behaviors to balance the impact of different behaviors on the score. Since playing is the most likely to occur and is prone to frequent playing among the above four behaviors, not only is the impact factor of playing set much lower than the impact factor of commenting, but a play validity factor is also set to reasonably limit the impact of the number of plays on the score to avoid distorted scores due to too many plays for any song.

[0079] S11 considers any behavior of any reference user on any song in the reference playlist as positive feedback within a preset time, which is expressed as the preference of the reference user, that is, in response to any reference behavior of any reference user on any reference song in the reference playlist, the reference score of the corresponding reference user corresponding to the reference song is updated according to the following formula:

[0080]

[0081] s play =λ u *num play *p play

[0082] s collect =num collect *p collect

[0083] s add =num add *p add

[0084] s comment =num comment *p comment

[0085] S u =s play*w play +s collect *w collect +s add *w add +s comment *w commet

[0086] w play +w collect +w add +w commet =1

[0087] w commet =max{w play , w collect , w add , w commet}

[0088] p play <p collect <p add <p comment .

[0089] Among them, λ u is the reference playback effectiveness factor, s play For reference, play pseudo-rating, num play is the reference playback number, p play is the reference playback score, s collect For reference collection pseudo rating, num collect is the reference number of favorites, p collect For reference, the collection points value, s add To add pseudo-ratings to the reference playlist, num add To add the reference playlist times, p add To add the reference song list points, s comment is the pseudo score of the reference review, num comment is the number of reference comments, p comment S is the reference comment score. u is the reference score, w play is the reference playback weight factor, w collect is the reference collection weight factor, w add To add the reference playlist weight factor, w commet is the reference comment weight factor.

[0090] S12: Within a preset time, any sample behavior of the target user on any sample song in the sample playlist is considered to be positive feedback, which is expressed as the preference of the target user. In response to any sample behavior of the target user on any sample song in the sample playlist, the sample score of the corresponding sample song is updated according to the following formula:

[0091]

[0092] s` play =λ` u *num` play *p` play

[0093] s` collect =num` collect *p` collect

[0094] s` add =num` add *p` add

[0095] s` comment =num` comment *p` comment

[0096] S a =s` play *w` play +s` collect *w` collect +s` add *w` add +s` comment *w` commet

[0097] w` play +w` collect +w` add +w` commet =1

[0098] w' commet =max{w` play , w' collect ,w` add ,w` commet}

[0099] p` play <p` collect <p` add <p` comment .

[0100] Among them, λ` u is the sample playback effective factor, s` play Play pseudo score for the sample, num` play is the number of sample playbacks, p` play is the sample playback score, s` collect Collect pseudo scores for samples, num` collect is the number of sample collections, p` collect is the sample collection score, s'add To add pseudo ratings to the sample playlist, num` add is the number of times the sample playlist is added, p' add To add the sample playlist points, s' comment Pseudo rating for sample reviews, num` comment is the number of sample comments, p' comment is the sample review score, S a Score the sample, w` play is the sample playback weight factor, w' collect is the sample collection weight factor, w' add To add the sample playlist weight factor, w' commet is the sample review weight factor.

[0101] (II) Similarity value

[0102] S21 calculates the similarity value between each reference user and the target user one by one through the following formula:

[0103]

[0104] Among them, sim(a, b) is the similarity value between the reference user and the target user, a is the target user, s a.i The sample score of song i given by the target user, I a，b is the intersection of the reference playlist and the sample playlist, s b，i is the reference user's reference rating for song i, U 1 is the reference user set;

[0105] S22 synthesizes the first K reference users with the highest similarity values ​​to the target user into a similar user set.

[0106] (III) Formation of the Recommendation List

[0107] According to the similar user set and the reference user-reference song list-reference behavior-reference score vector table, the prediction of the target user-reference song list-quasi-score vector table in S31 includes:

[0108] The target user's proposed rating for each reference song in each reference playlist is predicted one by one using the following formula:

[0109]

[0110] Among them, sim(a,u′) is the similarity value between the similar user and the target user, l is the reference song in the reference playlist, S u′，l is the reference score of similar users for the reference song l, C a，l is the target user's proposed rating for the reference song l, and U is the set of similar users.

[0111] S32 generates a recommendation list according to the target user-reference song list-presumed rating vector table, including:

[0112] Remove the intersection of the reference playlist and the sample playlist in the reference playlist to obtain a recommended playlist;

[0113] According to the proposed ratings given by the target users to the recommended songs in the recommended playlist, the top N recommended songs with the highest proposed ratings are selected to generate a recommendation list.

[0114] Example 3

[0115] This embodiment introduces in detail a recommendation system based on user implicit feedback behavior, including:

[0116] The target data module is used to obtain the target user behavior log data, and statistically analyze the target user behavior log data to obtain the target user-sample song list-sample behavior vector table;

[0117] A reference data module is used to obtain reference user behavior log data, and statistically analyze the reference user behavior log data to obtain a reference user-reference song list-reference behavior vector table;

[0118] A reference points module, used to convert the reference user-reference song-reference behavior vector table into a reference user-reference song list-reference behavior-reference score vector table according to a preset points rule;

[0119] The sample points module is used to convert the target user-sample song list-sample behavior vector table into the target user-sample song list-sample behavior-sample score vector table according to the preset points rule;

[0120] A similar user module is used to calculate the similarity value between each reference user and the target user one by one according to the reference user-reference song list-reference behavior-reference score vector table and the target user-sample song list-sample behavior-sample score vector table, and synthesize the first K reference users with the highest similarity values ​​to the target user into a similar user set;

[0121] The recommendation module is used to predict the target user-reference song list-pseudo-rating vector table based on the similar user set and the reference user-reference song list-reference behavior-reference rating vector table, and generate a recommendation list based on the target user-reference song list-pseudo-rating vector table.

[0122] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0123] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0124] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0125] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0126] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the enlightenment of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the purpose of the present invention and the claims, which all fall within the protection of the present invention.

Claims

1. A recommendation method based on user implicit feedback behavior, characterized in that: The following steps are involved: Obtain target user behavior log data, and statistically analyze the target user behavior log data to obtain a target user-sample song list-sample behavior vector table; Obtain reference user behavior log data, and statistically analyze the reference user behavior log data to obtain a reference user-reference song list-reference behavior vector table; According to the preset scoring rules, the reference user-reference song-reference behavior vector table is converted into a reference user-reference song list-reference behavior-reference score vector table; According to the preset scoring rules, the target user-sample song list-sample behavior vector table is converted into a target user-sample song list-sample behavior-sample score vector table; According to the reference user-reference song list-reference behavior-reference score vector table and the target user-sample song list-sample behavior-sample score vector table, the similarity value between each reference user and the target user is calculated one by one, and the first K reference users with the highest similarity value to the target user are synthesized into a similar user set; According to the similar user set and the reference user-reference song list-reference behavior-reference score vector table, the target user-reference song list-quasi-score vector table is predicted, and a recommendation list is generated according to the target user-reference song list-quasi-score vector table; The intersection of the reference playlist and the sample playlist of each reference user is a non-empty set; The integration rules include: In response to any reference behavior of any reference user on any reference song in the reference playlist, the reference score of the corresponding reference song of the corresponding reference user is updated according to the following formula: s play =λ u *number play *p play s collect =number collect *p collect yes add =number add *p add s comment =number comment *p comment S u =s play *w play +s collect *w colect +s add *w add +s comment *w commet Among them, λ u is the reference playback effectiveness factor, s play For reference, play pseudo-rating, num play is the reference playback number, p play is the reference playback score, s collect For reference collection pseudo rating, num collect is the reference number of favorites, p collect For reference, the collection points value, s add To add pseudo-ratings to the reference playlist, num add To add the reference playlist times, p add To add the reference song list points, s comment is the pseudo score of the reference review, num comment is the number of reference comments, p comment S is the reference comment score. u is the reference score, w play is the reference playback weight factor, w collect is the reference collection weight factor, w add To add the reference playlist weight factor, w commet is the reference comment weight factor; In response to any sample behavior of the target user on any sample song in the sample playlist, the sample score of the corresponding sample song is updated according to the following formula: yes play =λ` u *number` play *p` play s` collect =num` collec t*p` collect yes add =number add *p add Yes comment =number` comment *p` comment S a =s` play *w` play +s` colect *w` collect +s` add *w` add +s` comment *w` commet Among them, λ` u is the sample playback effective factor, s` play Play pseudo score for the sample, num` play is the number of sample playbacks, p` play is the sample playback score, s` collect Collect pseudo scores for samples, num` collect is the number of sample collections, p` collect is the sample collection score, s` add To add pseudo ratings to the sample playlist, num` add is the number of times the sample playlist is added, p` add To add sample playlist points, s` comment Pseudo rating for sample reviews, num` comment is the number of sample comments, p` comment is the sample comment score, S a Score the sample, w` play is the sample playback weight factor, w` collect is the sample collection weight factor, w` add To add the sample playlist weight factor, w` commet is the sample review weight factor.

2. The recommendation method based on user implicit feedback behavior according to claim 1, characterized in that: The integration rules include: In play +in collect +in add +in commet =1 w commet =max{w play ,w collect ,w add ,w commet } p play <p collect <p add <p comment 。 3. The recommendation method based on user implicit feedback behavior according to claim 1, characterized in that: The integration rules include: in` play +in` collect +in` add +in` commet =1 w` commet =max{w` play ,w` collect ,w` add ,w` commet } p` play <p` collect <p` add <p` comment 。 4. The recommendation method based on user implicit feedback behavior according to claim 1, characterized in that: The similarity value between each reference user and the target user is calculated one by one using the following formula: Among them, sim(a,b) is the similarity value between the reference user and the target user, a is the target user, s a,i The sample score of song i given by the target user, I a,b is the intersection of the reference playlist and the sample playlist, s b,i is the reference rating of song i by the reference user, and U1 is the reference user set; The first K reference users with the highest similarity values ​​to the target user are synthesized into a similar user set.

5. The recommendation method based on user implicit feedback behavior according to claim 1, characterized in that: The method of predicting the target user-reference song list-quasi-score vector table according to the similar user set and the reference user-reference song list-reference behavior-reference score vector table includes: The target user's proposed rating for each reference song in each reference playlist is predicted one by one using the following formula: Among them, sim(a,u′) is the similarity value between the similar user and the target user, l is the reference song in the reference playlist, S u′,l is the reference score of similar users for the reference song l, C a,l is the target user's proposed rating for the reference song l, and U is the set of similar users.

6. The recommendation method based on user implicit feedback behavior according to claim 1, characterized in that: The generating of the recommendation list according to the target user-reference song list-presumed rating vector table includes: Remove the intersection of the reference playlist and the sample playlist in the reference playlist to obtain a recommended playlist; According to the proposed ratings given by the target users to the recommended songs in the recommended playlist, the top N recommended songs with the highest proposed ratings are selected to generate a recommendation list.

7. A recommendation system based on user implicit feedback behavior, characterized in that: include: The target data module is used to obtain the target user behavior log data, and statistically analyze the target user behavior log data to obtain the target user-sample song list-sample behavior vector table; A reference data module is used to obtain reference user behavior log data, and statistically analyze the reference user behavior log data to obtain a reference user-reference song list-reference behavior vector table; A reference points module, used to convert the reference user-reference song-reference behavior vector table into a reference user-reference song list-reference behavior-reference score vector table according to a preset points rule; The sample points module is used to convert the target user-sample song list-sample behavior vector table into the target user-sample song list-sample behavior-sample score vector table according to the preset points rule; A similar user module is used to calculate the similarity value between each reference user and the target user one by one according to the reference user-reference song list-reference behavior-reference score vector table and the target user-sample song list-sample behavior-sample score vector table, and synthesize the first K reference users with the highest similarity values ​​to the target user into a similar user set; The recommendation module is used to predict the target user-reference song list-quasi-rating vector table based on the similar user set and the reference user-reference song list-reference behavior-reference rating vector table, and generate a recommendation list based on the target user-reference song list-quasi-rating vector table; The intersection of the reference playlist and the sample playlist of each reference user is a non-empty set; The integration rules include: In response to any reference behavior of any reference user on any reference song in the reference playlist, the reference score of the corresponding reference song of the corresponding reference user is updated according to the following formula: s play =λ u *number play *p play s collect =number collect *p collect yes add =number add *p add s comment =number comment *p comment S u =s play *w play +s collect *w collect +s add *w add +s comment *w commet Among them, λ u is the reference playback effectiveness factor, s play For reference, play pseudo-rating, num play is the reference playback number, p play is the reference playback score, s collect For reference collection pseudo rating, num collect is the reference number of favorites, p collect For reference, the collection points value, s add To add pseudo-ratings to the reference playlist, num add To add the reference playlist times, p add To add the reference song list points, s comment is the pseudo score of the reference review, num comment is the number of reference comments, p comment S is the reference comment score. u is the reference score, w play is the reference playback weight factor, w collect is the reference collection weight factor, w add To add the reference playlist weight factor, w commet is the reference comment weight factor; In response to any sample behavior of the target user on any sample song in the sample playlist, the sample score of the corresponding sample song is updated according to the following formula: yes paly =λ` u *number` play *p` play yes collect =number` collect *p` collect yes add =number` add *p* add yes comment =number` comment *p` comment S a =s` play *w` play +s` collect *w` collect +s` add *w` add +s` comment *w` commet Among them, λ` u is the sample playback effective factor, s` play Play pseudo score for the sample, num` play is the number of sample playbacks, p` play is the sample playback score, s` collect Collect pseudo scores for samples, num` collect is the number of sample collections, p` collect is the sample collection score, s` add To add pseudo ratings to the sample playlist, num` add is the number of times the sample playlist is added, p` add To add sample playlist points, s` comment Pseudo rating for sample reviews, num` comment is the number of sample comments, p` comment is the sample comment score, S is the sample score, w` play is the sample playback weight factor, w` collect is the sample collection weight factor, w` add To add the sample playlist weight factor, w` comment is the sample review weight factor.

Images