CN109408665A - Information recommendation method and device and storage medium - Google Patents

Abstract

The embodiment of the present invention discloses an information recommendation method, a device, and a storage medium. The method includes: acquiring historical operation data of songs selected by a target user for a full history within a preset statistical period; based on the preset operation data and operation scores The corresponding relationship and the historical operation data, obtain the historical evaluation data of each historically selected song in the full amount of historically selected songs; the historical evaluation data is used to characterize the target user's liking for the song; The historical evaluation data is subjected to time decay processing to obtain the historical statistical evaluation data of each historically selected song; according to the historical statistical evaluation data, the full amount of historically selected songs is sorted to obtain the sorting result of the full amount of historically selected songs ; Recommend target songs according to the sorting result and the preset full playlist.

Description

Method and device for recommending information, and storage medium

technical field

The present invention relates to information processing technology, and in particular, to an information recommendation method and device, and a storage medium.

Background technique

With the development of terminal and Internet technology, users generally listen to music through music applications. Music applications recommend music that suits their personal hobbies through some music recommendation algorithms. The currently used music recommendation algorithm combines multiple interests. Users form a group, use the information that all users in the group are interested in and the information they are not interested in, generate recommended information, and recommend the recommended information to all users. In this way, the personal recommendation information obtained is greatly influenced by other users in the group. If there are abnormal users in other users or the information that other users are interested in is inaccurate, it will lead to a problem that the recommended information deviates greatly from personal hobbies, that is, the accuracy of the recommended information is low.

SUMMARY OF THE INVENTION

The main purpose of the present invention is to provide an information recommendation method, device, and storage medium, which can improve the accuracy of recommended information.

The technical scheme of the present invention is realized as follows:

An embodiment of the present invention provides an information recommendation method, and the method includes:

Obtain the historical operation data of the songs selected by the target users in the preset statistical period for the full amount of history;

Based on the preset corresponding relationship between the operation data and the operation score, and the historical operation data, the historical evaluation data of each historically selected song in the full amount of historically selected songs is obtained; the historical evaluation data is used to represent the target The user's liking to the song;

Performing time decay processing on the historical evaluation data to obtain historical statistical evaluation data of each of the historically selected songs;

According to the historical statistical evaluation data, sort the full amount of historically selected songs to obtain a sorting result of the full amount of historically selected songs;

According to the sorting result and the preset full playlist, the target song is recommended.

In the above solution, the preset statistical period includes at least one preset time period; the historical evaluation data includes period evaluation data corresponding to each time period in the at least one preset time period;

The time decay processing is performed on the historical evaluation data to obtain the historical statistical evaluation data of each historically selected song, including:

According to the corresponding relationship between the preset attenuation coefficient and time, and the time information represented by each time period, the attenuation coefficient corresponding to each time period is obtained;

According to the attenuation coefficient, the period evaluation data corresponding to the at least one preset time period are summed to obtain the historical statistical evaluation data.

In the above solution, the recommended target songs are recommended according to the sorting results and the preset full playlist, including:

According to the sorting result, a reference song is determined from the full amount of historically selected songs;

According to the reference song, determine a reference playlist from the preset full playlist;

According to the reference playlist, the target song is recommended.

In the above scheme, the reference song list includes a sub-reference song list corresponding to each reference song in the reference song;

The recommending the target song according to the reference playlist includes:

According to the repetition times of each song in the sub-reference playlist, the target song is determined from the reference playlist, and the target song is recommended.

In the above scheme, the target song is determined from the reference song list according to the number of repetitions of each song in the sub-reference song list, including:

Sort the historical statistical evaluation data of each of the reference songs, and obtain the data sorting sequence number of each of the reference songs;

Sort the repetition times of each song in the sub-reference playlist, obtain the ordering sequence number of the times of each song in the sub-reference playlist;

According to the sequence number of the times, determine the song to be recommended corresponding to each reference song from the sub-reference song list, and obtain at least one song to be recommended;

According to the data sorting sequence number and the frequency sorting sequence number of the songs to be recommended, determine the preference score of the songs to be recommended;

According to the preference score, the target song is determined from the at least one to-be-recommended song.

In the above solution, after determining the reference playlist from the preset full playlist according to the reference song, the method further includes:

According to the preset song attributes, the songs of each playlist in the reference playlist are quantified to obtain the reference attribute feature of each playlist in the reference playlist;

According to the preset playlist scoring model and the reference attribute feature, scoring data including the scoring result of each playlist in the reference playlist is obtained; the preset playlist scoring model represents the corresponding relationship between the attribute feature and the scoring data ;

When the first scoring data in the scoring data is smaller than the preset scoring threshold, delete the playlist corresponding to the first scoring data in the reference playlist from the reference playlist to obtain an updated reference song list.

In the above scheme, before the scoring data including the scoring result of each playlist in the reference playlist is obtained according to the preset playlist scoring model and the reference attribute feature, the method further includes:

During the preset training period, obtain the sample operation data of the target user for the full sample songs;

According to the sample operation data, determine a positive sample playlist and a negative sample playlist from the preset full playlist;

Obtain the positive scoring data corresponding to the positive sample playlist, and the positive sample playlist and the positive scoring data constitute a positive sample;

obtaining negative score data corresponding to the negative sample playlist, where the negative sample playlist and the negative score data constitute a negative sample;

Using the positive samples and the negative samples, the preset initial playlist scoring model is trained until the accuracy of the trained initial playlist scoring model is not less than the preset accuracy threshold;

The trained initial song list scoring model is used as the preset song list scoring model.

An embodiment of the present invention provides an information recommendation device, the device includes: an acquisition unit, a calculation unit, an attenuation processing unit, and a recommendation unit; wherein,

The obtaining unit is used to obtain the historical operation data of the songs selected by the target users in the preset statistical period for the full amount of history;

The computing unit is configured to obtain historical evaluation data of each historically selected song in the full amount of historically selected songs based on the preset corresponding relationship between the operation data and the operation score and the historical operation data; the historical evaluation The data is used to characterize the liking of the target user to the song;

The attenuation processing unit is used to perform time attenuation processing on the historical evaluation data to obtain the historical statistical evaluation data of each historically selected song;

The recommending unit is configured to sort the full amount of historically selected songs according to the historical statistical evaluation data to obtain a sorting result of the full amount of historically selected songs; target song.

In the above solution, the preset statistical period includes at least one preset time period; the historical evaluation data includes period evaluation data corresponding to each time period in the at least one preset time period;

The attenuation processing unit is specifically configured to obtain the attenuation coefficient corresponding to each time period according to the corresponding relationship between the preset attenuation coefficient and time and the time information represented by each time period; and according to the attenuation coefficient , summing the time period evaluation data corresponding to the at least one preset time period to obtain the historical statistical evaluation data.

In the above solution, the recommending unit is specifically configured to determine a reference song from the full historical selection songs according to the sorting result; and determine a reference song from the preset full song list according to the reference song a playlist; and recommending the target song based on the reference playlist.

In the above scheme, the reference song list includes a sub-reference song list corresponding to each reference song in the reference song;

The recommending unit is specifically configured to determine the target song from the reference playlist according to the repetition times of each song in the sub-reference playlist, and recommend the target song.

In the above scheme, the recommending unit is specifically used to sort the historical statistical evaluation data of each reference song to obtain the data sorting sequence number of each reference song; and to each song in the sub-reference song list. The number of repetitions of the songs is sorted to obtain the sequence number of the number of times of each song in the sub-reference playlist; recommending songs, and obtaining at least one song to be recommended; and determining the preference score of the song to be recommended according to the data sorting sequence number and the frequency sorting sequence number of the song to be recommended; and according to the preference score, from the The target song is determined from at least one song to be recommended.

In the above solution, the recommending unit is further configured to, after the reference song list is determined from the preset full-volume song list according to the reference song, according to the preset song attribute, perform an update on the reference song list. The songs of each playlist are quantified to obtain the reference attribute feature of each playlist in the reference playlist; The scoring data of the scoring result of the playlist; the preset playlist scoring model represents the corresponding relationship between the attribute feature and the scoring data; and when the first scoring data in the scoring data is smaller than the preset scoring threshold, the reference The playlist corresponding to the first scoring data in the playlist is deleted from the reference playlist to obtain an updated reference playlist.

In the above solution, the device further includes: a model training unit for obtaining a score including the scoring result of each playlist in the reference playlist according to the preset playlist scoring model and the reference attribute feature. Before the data, within the preset training period, obtain the sample operation data of the target user for the full sample songs; and according to the sample operation data, determine the positive sample playlist and the negative sample playlist from the preset full playlist and obtain the positive scoring data corresponding to the positive sample song list, the positive sample song list and the positive scoring data constitute a positive sample; and obtain the negative scoring data corresponding to the negative sample song list, the negative sample song list The list and the negative scoring data constitute a negative sample; and using the positive sample and the negative sample, the preset initial song list scoring model is trained until the accuracy of the trained initial song list scoring model is not less than the preset an accurate threshold; and using the trained initial playlist scoring model as the preset playlist scoring model.

An embodiment of the present invention provides an apparatus for recommending information. The apparatus includes: a processor, a memory, and a communication bus. The memory communicates with the processor through the communication bus, and the memory stores the information available to the processor. One or more programs to be executed, when the one or more programs are executed, the processor executes the steps of any one of the above information recommendation methods.

An embodiment of the present invention provides a computer-readable storage medium, where a program is stored in the computer-readable storage medium, and when the program is executed by at least one processor, the at least one processor is caused to execute any one of the foregoing The steps of the information recommendation method.

Embodiments of the present invention provide an information recommendation method, device, and storage medium. The method includes: acquiring historical operation data of songs selected by a target user for a full amount of history within a preset statistical period; based on the preset operation data and operation scores The corresponding relationship and the historical operation data, obtain the historical evaluation data of each historically selected song in the full amount of historically selected songs; the historical evaluation data is used to characterize the target user's liking for the song; The historical evaluation data is subjected to time decay processing to obtain the historical statistical evaluation data of each historically selected song; according to the historical statistical evaluation data, the full amount of historically selected songs is sorted to obtain the sorting result of the full amount of historically selected songs ; Recommend target songs according to the sorting result and the preset full playlist. Using the above technical implementation scheme, the historical evaluation data within the preset statistical period is subjected to time decay processing to obtain the historical statistical evaluation data after time decay processing, and then the sorting results of the full historical selection songs are obtained, and the target songs are recommended according to the sorting results. The historical evaluation data indicates how much the target user likes the song within the preset time period. Then, the historical evaluation data in the preset statistical period is subjected to time decay processing to obtain the historical statistical evaluation data representing the user's current preferences, and then use the full amount of history. The sorting result of the historical statistical evaluation data of songs, the obtained target songs are more in line with the current preferences of the target users, and the accuracy of the recommendation information is improved.

Description of drawings

FIG. 1 is a schematic structural diagram of an information recommendation system according to an embodiment of the present invention;

FIG. 2 is a flowchart 1 of an information recommendation method provided by an embodiment of the present invention;

FIG. 3 is a second flowchart of an information recommendation method provided by an embodiment of the present invention;

FIG. 4 is a schematic structural diagram 1 of an information recommendation apparatus provided by an embodiment of the present invention;

FIG. 5 is a second schematic structural diagram of an information recommendation apparatus according to an embodiment of the present invention.

Detailed ways

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

In the following description, suffixes such as 'module', 'component' or 'unit' used to represent elements are used only to facilitate the description of the present invention and have no specific meaning per se. Thus, "module", "component" or "unit" may be used interchangeably.

As shown in FIG. 1, which is a schematic structural diagram of an information recommendation system for implementing various embodiments of the present invention, the system 1 may include: a server 10 and a terminal 11; wherein, the server 10 may store data files of application programs, application programs User data, etc., can also broadcast and synchronize user data; when the terminal 11 runs the application program, it interacts with the server 10, and can download the data file of the application program from the server 10 through the network, and can also send to the server. 10 Upload the operation data generated by the user operating the application, the terminal 11 can be implemented in various forms, for example, can include mobile terminals such as mobile phones, tablet computers, notebook computers, PDAs, etc., and fixed terminals such as desktop computers; The program should include music applications.

Those skilled in the art can understand that the structure of the information recommendation system shown in FIG. 1 does not constitute a limitation on the recommended information recommendation system, and the information recommendation system may include more or less components than those shown in the figure, or combine some components, Or a different component arrangement.

It should be noted that the embodiment of the present invention may be implemented based on the information recommendation system shown in FIG. 1 , and the information recommendation device may be the server 10 .

Example 1

An embodiment of the present invention provides an information recommendation method, as shown in FIG. 2 , the method includes:

S201: Acquire historical operation data of songs selected by a target user for a full amount of history within a preset statistical period.

The information recommendation device obtains, from the terminal according to a preset time period, the time period operation data of the target user's selection of songs for the full history, and stores the preset time period and the time period operation data correspondingly until the historical operation data within the preset statistical period is obtained, The preset statistical period includes at least one preset time period, and the historical operation data includes period operation data in each time period in the at least one preset time period; for example, the preset time period may be one day or one hour.

Exemplarily, taking a music application as an example, when a user uses a music application on the terminal, the terminal records the user's login account information (for example, user identification number (ID, IDentity)), and the terminal also records the operation information. The data is recorded, and the operation data is the data generated after the user selects the songs in the music application. The terminal stores the user ID and the operation data correspondingly, and sends the operation data of each time period in combination with the user ID. to the server, and the server saves the time period operation data of each user according to the user ID; wherein, the selection operations for songs include: like operation, collection operation, comment operation, sharing operation, listening operation, song cutting operation, audition operation, For the purchase operation, the setting of the CRBT operation, the download operation, etc., the operation data includes the song identification (eg, song ID) of the song, and the operation type of the selection operation.

S202: Based on the preset corresponding relationship between the operation data and the operation score, as well as the historical operation data, obtain the historical evaluation data of each historically selected song in the full amount of historically selected songs; the historical evaluation data is used to represent the target user's liking of the song. .

The information recommendation device determines the historical operation data of each user in the target users from the historical operation data, and operates on the time period corresponding to each time period in the historical operation data of each user according to the preset corresponding relationship between the operation data and the operation score. The operation score is calculated on the data to obtain the time period evaluation data corresponding to each time period, and then the historical evaluation data of each user composed of the time period evaluation data corresponding to all time periods is obtained.

Exemplarily, the preset time period is one day, and calculating the time period evaluation data corresponding to each time period is to calculate the user's liking for different songs in a single day. The preference score reflects the user's liking degree expressed by the selection operation of each operation type, and the corresponding relationship between the preset operation data and the operation score as shown in the following table 1 is obtained:

Table 1

operational data Score Song cut operation -0.5 mark like action 2 mark dislike action -2 download operation 4 Comment action 1 share action 2 Favorite operation 2 Audition operation 0.5 listen to action 4 Set as CRBT operation 3 search operation 1

Exemplarily, the information recommendation device calculates the historical evaluation data of each historically selected song in the full amount of historically selected songs according to the corresponding relationship in Table 1 and the operation type of the selection operation in the historical operation data, and records it in the following form: " User ID-Song ID-Operation Score-Date", the date is the time information corresponding to each time period in the at least one preset time period.

Exemplarily, after acquiring the period operation data of each time period, the server uses the period operation data to calculate the period evaluation data of each user for each historically selected song, for example, the period operation data of the historically selected song A includes: If user B searches 5 times, shares 4 times, favorites 1 time, comments 2 times, downloads 2 times, and cuts songs 3 times, then according to the corresponding relationship in Table 1, formula (1) is obtained, and formula (1) is used to calculate The time period evaluation data of user B's historical selection of song A in one day is 23.5:

5×1+4×2+1×2+2×1+2×4+3×(-0.5)=23.5 (1)

When the time information corresponding to this day is October 11, 2018, the time period evaluation data of the historically selected song A can be recorded in the following form: user B—song A—23.5—20181011.

S203: Perform time decay processing on the historical evaluation data to obtain historical statistical evaluation data of each historically selected song.

The historical evaluation data of each user obtained by the information recommendation device is composed of the time period evaluation data corresponding to each time period, and the time decay processing is performed on the time period evaluation data corresponding to all time periods to obtain the historical statistical evaluation corresponding to each user. data.

It should be noted that, since the user's preferences for songs may change dynamically, it is often impossible to accurately determine the user's current preferences only based on the period evaluation data within a preset time period. For example, a preset statistical period of 90 days is used to determine the user's 90-day liking scores for each song selected by the user within 90 days. The total likes score for the selected song in history.

In some embodiments, the information recommendation apparatus obtains the attenuation coefficient corresponding to each time period according to the corresponding relationship between the preset attenuation coefficient and time, and the time information represented by each time period; according to the attenuation coefficient, for at least one preset time period The evaluation data corresponding to the segment is summed to obtain historical statistical evaluation data.

Exemplarily, since the user's preference for songs changes dynamically, the user's recent period evaluation data tends to reflect the user's current preference for songs more accurately than the earlier period evaluation data. As the time interval of the current time increases, the accuracy of reflecting the user's current preferences gradually decreases. Therefore, at the end of the preset statistical period, the information recommending device uses the time information corresponding to each time period and the initial time corresponding to the preset statistical period. According to the attenuation coefficient, the weighted summation of the period evaluation data corresponding to all time periods is carried out to obtain the historical statistical evaluation data.

Exemplarily, the corresponding relationship between the preset attenuation coefficient and time is expressed as formula (2):

N(t)=N ₀ e ^-t/T (2)

Among them, N ₀ represents the initial attenuation coefficient at the start of the preset statistical period, t represents the time interval between the time information corresponding to each time period and the start of the preset statistical period, and N(t) represents the corresponding Attenuation coefficient, e represents the exponent, and T represents the end time of the preset statistical period.

S204: Rank the full historically selected songs according to the historical statistical evaluation data, and obtain a sorting result of the full historically selected songs.

According to the historical statistical evaluation data of each user and the size of the historical statistical evaluation data, the information recommendation device performs ascending or descending sorting on all historically selected songs to obtain a sorting result.

S205: Recommend target songs according to the sorting result and the preset full playlist.

For each user, the information recommendation device selects n reference songs from the full history based on the sorting result, where n is an integer greater than 0, and the historical statistical evaluation data of the n reference songs is greater than the historical statistical evaluation of other songs in the full historical selection. data, the n reference songs are the favorite n reference songs of each user, and then the target songs are determined according to the n reference songs and the preset full playlist of the music application.

It should be noted that, for an information recommendation method as shown in FIG. 3 , the specific implementation of step S205 is to execute S2051-S2053, including:

S2051: According to the sorting result, determine the reference song from the full amount of historically selected songs;

The information recommendation device selects n reference songs from the full amount of historically selected songs.

S2052: According to the reference song, determine the reference playlist from the preset full playlist;

For each reference song in the reference song, the information recommendation device determines all the playlists including each reference song from the preset full playlist, and uses all the playlists as the sub-reference playlists of each reference song, and then obtains all the playlists from all the reference songs. A reference playlist composed of sub-reference playlists.

S2053: Recommend the target song according to the reference song list.

The information recommendation device determines the target song from the songs in the reference playlist, and recommends the target song to the user by sending it to the terminal.

In some embodiments, the reference song list includes a sub-reference song list corresponding to each reference song in the reference song; the information recommendation device determines the target song from the reference song list according to the repetition times of each song in the sub-reference song list, Recommend target songs.

Exemplarily, taking a reference song A _L among the n reference songs that the user likes the most as an example, where L is an integer greater than 0 and not greater than n, the information recommendation device may determine that the reference song A is included in the preset full playlist. The sub-reference playlist of _L , the sub-reference playlist is expressed as {playlist L1, playlist L2, ..., playlist LN}, N is an integer greater than 2, and count the number of repetitions of each song in the sub-reference playlist , sort the repetition times of all songs in the sub-reference playlist, and obtain the sequence number of each song in the sub-reference playlist; determine the song whose sequence number belongs to the top q as the reference song _AL corresponding to the The recommended songs _TL and q are positive integers greater than 0, and then at least one to-be-recommended song composed of to-be-recommended songs corresponding to n reference songs is obtained, wherein the to-be-recommended song _TL may include at least one song.

In some embodiments, the information recommendation device sorts the historical statistical evaluation data of each reference song to obtain the data sorting serial number of each reference song; sorts the repetition times of each song in the sub-reference playlist to obtain the sub-reference The number of times of each song in the playlist is sorted; according to the number of times, the song to be recommended corresponding to each reference song is determined from the sub-reference playlist, and at least one song to be recommended is obtained; The number of times is sorted and the preference score of the song to be recommended is determined; according to the preference score, the target song is determined from at least one of the songs to be recommended.

Exemplarily, the information recommendation device sorts the historical statistical evaluation data of n reference songs in descending order, and obtains the data sorting sequence number g of each reference song; Get the number h of the songs to be recommended corresponding to each reference song; calculate the preference score Score of the songs to be recommended according to formula (3):

According to the preference score of the song to be recommended, a target song with a higher preference score is determined from at least one of the songs to be recommended, and is recommended to the user.

It should be noted that, after step S2052, the information recommendation device quantifies the songs of each playlist in the reference playlist according to the preset song attributes, and obtains the reference attribute features of each playlist in the reference playlist; Playlist scoring model and reference attribute features to obtain scoring data including the scoring results of each playlist in the reference playlist; the preset playlist scoring model represents the corresponding relationship between attribute features and scoring data; when there is a first score in the scoring data When the data is less than the preset score threshold, the playlist corresponding to the first scoring data in the reference playlist is deleted from the reference playlist to obtain an updated reference playlist.

Exemplarily, in order to avoid the influence of inferior playlists, the information recommendation device quantifies each song in each playlist according to the preset song attributes, obtains a multi-dimensional attribute vector of each song, and presets each song in the song attributes. Attributes correspond to one-dimensional data; use the attribute vectors of all songs in each playlist to generate a multi-dimensional attribute matrix, which is the reference attribute feature of each playlist; use the preset playlist scoring model and the multi-dimensional attributes of each playlist Matrix, calculate the scoring results of each playlist; finally, filter the reference playlists according to the scoring results of each playlist.

Exemplarily, the preset song attributes include song release year, singer age, singer gender information, song rhythm information, song pitch information, etc. The singer gender information includes singer gender information of male singers equal to 0, and singer gender information of female singers equal to 0. 1. The rhythm information of the song is the number of beats per minute of the song, and the tone information of the song includes the CDEFGAB key of the song corresponding to 1-7 respectively. The male singer born in 1979 released the song in 2007. The released song is in the key of A and 60 per minute. beat, the multi-dimensional attribute vector of the released song is expressed as {2007, 1979, 0, 60, 6}; wherein, the position of each song attribute in the preset song attribute in the multi-dimensional attribute vector can be changed, and it is only necessary to ensure that the multi-dimensional attribute vector The order of the attributes of the song is the same as the order of the attributes of the songs corresponding to the preset playlist scoring model.

In some embodiments, the information recommendation device obtains sample operation data of the target user for the full sample songs within a preset training period; according to the sample operation data, determines the positive sample song list and the negative sample song from the preset full song list single; obtain the positive score data corresponding to the positive sample playlist, and the positive sample playlist and the positive score data constitute a positive sample; obtain the negative score data corresponding to the negative sample playlist, and the negative sample playlist and the negative score data constitute a negative sample; use the positive sample and negative samples, train the preset initial playlist scoring model until the accuracy of the trained initial playlist scoring model is not less than the preset accuracy threshold; use the trained initial playlist scoring model as the preset playlist scoring Model.

Exemplarily, within a preset training period, the information recommendation device obtains sample operation data of the target user for the full amount of sample songs; obtains the sample statistical evaluation data of each sample song according to the sample operation data; Sort the sample songs to obtain the sample sorting results of the full sample songs; according to the sample sorting results, determine the positive sample playlist and the negative sample playlist from the preset full playlist, and the positive sample playlist contains the user's favorite songs. The playlist of songs, the negative sample playlist is the playlist that contains the most disliked songs of the user; according to the preset song attributes, quantify the songs of each playlist in the positive sample playlist, and get the positive sample playlist. The positive attribute features of each playlist, according to the preset song attributes, quantify the songs of each playlist in the negative sample playlist to obtain the negative attribute features of each playlist in the negative sample playlist; positive attribute features and positive scores The data constitutes a positive sample, and the negative attribute features and negative rating data constitute a negative sample.

Exemplarily, when the positive score data is 1 and the negative score data is -1, the information recommendation apparatus deletes the playlists whose score results are less than 0 from the reference playlist according to the score result of each playlist.

It should be noted that the full sample songs correspond to the aforementioned full historical selection songs, and the full sample songs and the full historical selection songs are only at different acquisition times. Similarly, the sample operation data corresponds to the aforementioned historical operation data, sample operation data and historical operation songs. Only the acquisition time is different; therefore, the process of acquiring sample statistical evaluation data and sample sorting results based on sample operation data is consistent with the process of acquiring historical statistical evaluation data and sorting results based on historical operation data.

Exemplarily, the Adboost algorithm can be used to obtain a preset song list scoring model, which specifically includes: setting an initial song list scoring model, and the initial song list scoring model includes a weak classifier corresponding to each adjacent two-dimensional data; The initial data weight of each dimension data, use each adjacent two dimension data corresponding to the positive sample and the negative sample to train a weak classifier, calculate the classifier weight of each weak classifier, use each weak classifier and the classifier weight Generate a strong classifier; calculate the accuracy of the strong classifier, when the accuracy is less than the preset accuracy threshold, adjust the initial data weight until the accuracy of the strong classifier is not less than the preset accuracy threshold; use the strong classifier as the preset song Single scoring model.

Exemplarily, the preset song attributes include 5 song attributes, which are the release year of the song, the singer's age, the singer's gender information, the song rhythm information, and the song pitch information, and the strong classifier is generated by K=5 weak classifiers, for example, When every two adjacent dimensional data is represented as {first song attribute x, second song attribute y}, the sample input is obtained according to positive samples and negative samples as {(x ₁ , y ₁ ), (x ₂ , y ₂ ),...,(x _m ,y _m )}, the sample output is {-1,+1}, m is an integer greater than 0; according to the initial data weight, use the sample input and sample output to train the preset initial playlist The scoring model obtains the weak classifier G ₁ (x) corresponding to the first song attribute x, and then obtains the weak classifier corresponding to each song attribute; wherein, the initial data weight D(1) is shown in formula (4):

D(1)=(w ₁₁ ,w ₁₂ ,...,w _1m ) (4)

in,

Further, training the weak classifier G _k (x) specifically includes: using the data weight D(k) corresponding to the weak classifier G _k (x) to train the initial song list scoring model to obtain the weak classifier G _k (x), where k=1,2,...K, and the data weight D(k) is shown in formula (5):

D(k)=(w _k1 ,w _k2 ,...,w _km ) (5)

in, Use formula (6) to calculate the classification error rate _ek of the weak classifier G _k (x):

Wherein, after inputting (x _i , y _i ) to the weak classifier G _k ( _xi ), a scoring result is output, and the scoring result is not equal to the preset threshold I corresponding to _yi .

Calculate the classifier weight _ak of the weak classifier G _k (x) using formula (7):

Use formula (8) to calculate the data weight D(k+1) corresponding to the weak classifier k+1:

where Z _k is the normalization factor, as shown in Equation (9):

Use formula (10) to generate a strong classifier:

It should be noted that the larger the classification error rate of the weak classifier, the smaller the classifier weight of the weak classifier.

It can be understood that the information recommendation device performs time decay processing on the historical evaluation data within the preset statistical period, obtains the historical statistical evaluation data after the time decay process, and then obtains the sorting result of all the historically selected songs, and recommends the target song according to the sorting result. , since the historical evaluation data represents the target user's liking of the song within the preset time period, then the historical evaluation data in the preset statistical period is subjected to time decay processing to obtain the historical statistical evaluation data representing the user's current preferences, and then use The sorting result of the historical statistical evaluation data of the full amount of historical songs, the obtained target songs are more in line with the current preferences of the target users, and the accuracy of the recommendation information is improved.

Embodiment 2

Based on the same inventive concept of the first embodiment, further description will be made.

An embodiment of the present invention provides an information recommendation device 4. As shown in FIG. 4, the device 4 includes: an acquisition unit 40, a calculation unit 41, an attenuation processing unit 42, and a recommendation unit 43; wherein,

The obtaining unit 40 is used to obtain the historical operation data of the target user's selection of songs for the full amount of history within the preset statistical period;

The computing unit 41 is used to obtain historical evaluation data of each historically selected song in the full amount of historically selected songs based on the preset corresponding relationship between the operation data and the operation score and the historical operation data; the historical The evaluation data is used to represent the liking of the target user to the song;

The attenuation processing unit 42 is configured to perform time attenuation processing on the historical evaluation data to obtain historical statistical evaluation data of each historically selected song;

The recommending unit 43 is configured to sort the full amount of historically selected songs according to the historical statistical evaluation data to obtain a sorting result of the full amount of historically selected songs; and according to the sorting result and the preset full amount of playlists, Recommend target songs.

In some embodiments, the preset statistical period includes at least one preset time period; the historical evaluation data includes period evaluation data corresponding to each time period in the at least one preset time period;

The attenuation processing unit 42 is specifically configured to obtain the attenuation coefficient corresponding to each time period according to the corresponding relationship between the preset attenuation coefficient and time, and the time information represented by each time period; and according to the attenuation coefficient, and summing the period evaluation data corresponding to the at least one preset period of time to obtain the historical statistical evaluation data.

In some embodiments, the recommending unit 43 is specifically configured to, according to the sorting result, determine a reference song from the full historical selection of songs; and according to the reference song, select a reference song from the preset full song list A reference playlist is determined; and the target song is recommended according to the reference playlist.

In some embodiments, the reference playlist includes a sub-reference playlist corresponding to each reference song in the reference songs;

The recommending unit 43 is specifically configured to determine the target song from the reference playlist according to the repetition times of each song in the sub-reference playlist, and recommend the target song.

In some embodiments, the recommending unit 43 is specifically configured to sort the historical statistical evaluation data of each reference song to obtain the data sorting sequence number of each reference song; and to sort the sub-reference song list Sort the number of repetitions of each song in the sub-reference playlist to obtain the number of times sorting sequence number of each song in the sub-reference playlist; Corresponding songs to be recommended, obtain at least one song to be recommended; and determine the preference score of the song to be recommended according to the data sorting sequence number and the frequency sorting sequence number of the song to be recommended; and according to the preference score, The target song is determined from the at least one song to be recommended.

In some embodiments, the recommending unit 43 is further configured to, after the reference playlist is determined from the preset full playlist according to the reference song, according to the preset song attribute The songs of each playlist in the playlist are quantified to obtain the reference attribute feature of each playlist in the reference playlist; and according to the preset playlist scoring model and the reference attribute feature, obtain the reference playlist containing the reference playlist The scoring data of the scoring result of each playlist in the playlist; the preset playlist scoring model represents the corresponding relationship between the attribute feature and the scoring data; and when there is a first scoring data in the scoring data that is less than the preset scoring threshold, the The playlist corresponding to the first scoring data in the reference playlist is deleted from the reference playlist to obtain an updated reference playlist.

In some embodiments, the model training unit is configured to, before obtaining the scoring data including the scoring result of each playlist in the reference playlist according to the preset playlist scoring model and the reference attribute feature, perform a During the preset training period, obtain the sample operation data of the target user for the full sample songs; and according to the sample operation data, determine a positive sample playlist and a negative sample playlist from the preset full playlist; and obtain all the sample playlists. the positive score data corresponding to the positive sample playlist, the positive sample playlist and the positive score data constitute a positive sample; and obtain the negative score data corresponding to the negative sample playlist, the negative sample playlist and the The negative scoring data constitutes a negative sample; and using the positive sample and the negative sample, a preset initial playlist scoring model is trained until the accuracy of the trained initial playlist scoring model is not less than a preset accuracy threshold; and The trained initial song list scoring model is used as the preset song list scoring model.

It should be noted that, in practical applications, the above-mentioned acquisition unit 40 , calculation unit 41 , attenuation processing unit 42 and recommendation unit 43 may be implemented by the processor 44 located on the information recommendation device 4 , specifically a CPU (Central Processing Unit, central processing unit). device), MPU (Microprocessor Unit, microprocessor), DSP (Digital SignalProcessing, digital signal processor) or Field Programmable Gate Array (FPGA, Field Programmable GateArray) and other implementations.

The embodiment of the present invention also provides an information recommendation device 4. As shown in FIG. 5, the device 4 includes: a processor 44, a memory 45, and a communication bus 46. The memory 45 communicates with the processor 44 through the communication bus 46, and the memory 45 stores one or more programs executable by the processor 44. When the one or more programs are executed, any one of the information recommendation methods described in the foregoing embodiments is executed by the processor 44.

In practical applications, the memory 45 may be a volatile first memory (volatile memory), such as a random-access first memory (Random-Access Memory, RAM); or a non-volatile first memory (non-volatile memory), For example, a read-only first memory (Read-Only Memory, ROM), a flash first memory (flashmemory), a hard disk (Hard Disk Drive, HDD) or a solid-state drive (Solid-State Drive, SSD); A combination of memories and provides programs and data to processor 44 .

An embodiment of the present invention provides a computer-readable storage medium, where the computer-readable storage medium stores one or more programs, the one or more programs can be executed by one or more processors, and the programs are executed by the processor 44 At the same time, any one of the information recommendation methods described in the foregoing embodiments is implemented.

As will be appreciated by one skilled in the art, embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the invention may take the form of a hardware embodiment, a software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media having computer-usable program code embodied therein, including but not limited to disk storage, optical storage, and the like.

The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each process and/or block in the flowchart illustrations and/or block diagrams, and combinations of processes and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to the processor of a general purpose computer, special purpose computer, embedded processor or other programmable data processing device to produce a machine such that the instructions executed by the processor of the computer or other programmable data processing device produce Means for implementing the functions specified in a flow or flow of a flowchart and/or a block or blocks of a block diagram.

These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory result in an article of manufacture comprising instruction means, the instructions The apparatus implements the functions specified in the flow or flow of the flowcharts and/or the block or blocks of the block diagrams.

These computer program instructions can also be loaded on a computer or other programmable data processing device to cause a series of operational steps to be performed on the computer or other programmable device to produce a computer-implemented process such that The instructions provide steps for implementing the functions specified in the flow or blocks of the flowcharts and/or the block or blocks of the block diagrams.

The above descriptions are merely preferred embodiments of the present invention, and are not intended to limit the protection scope of the present invention.

Claims (16)

1. An information recommendation method, characterized in that the method comprises:

acquiring historical operation data of a target user for selecting songs aiming at the full amount of history in a preset statistical period;

obtaining historical evaluation data of each historical selected song in the total historical selected songs based on the corresponding relation between preset operation data and operation scores and the historical operation data; the historical evaluation data is used for representing the likeness degree of the target user to the song;

performing time attenuation processing on the historical evaluation data to obtain historical statistical evaluation data of each historical selected song;

sorting the full-scale historical selected songs according to the historical statistical evaluation data to obtain a sorting result of the full-scale historical selected songs;

and recommending the target song according to the sequencing result and a preset full-scale song list.

2. The method according to claim 1, wherein the preset statistical period comprises at least one preset time period; the historical evaluation data comprises time interval evaluation data corresponding to each time interval in the at least one preset time interval;

the time attenuation processing is performed on the historical evaluation data to obtain the historical statistical evaluation data of each historical selected song, and the method comprises the following steps:

obtaining an attenuation coefficient corresponding to each time period according to the corresponding relation between a preset attenuation coefficient and time and the time information represented by each time period;

and summing the time interval evaluation data corresponding to the at least one preset time interval according to the attenuation coefficient to obtain the historical statistical evaluation data.

3. The method of claim 1, wherein recommending the target song according to the ranking result and the preset full-volume menu comprises:

according to the sorting result, determining a reference song from the songs selected from the full-scale history;

determining a reference song list from the preset total song list according to the reference song;

and recommending the target song according to the reference song list.

4. The method of claim 3, wherein the reference menu comprises sub-reference menus corresponding to each of the reference songs;

the recommending the target song according to the reference song list comprises the following steps:

and determining the target song from the reference list according to the repeated times of each song in the sub-reference list, and recommending the target song.

5. The method of claim 4, wherein said determining the target song from the reference menu based on the number of repetitions of each song in the sub-reference menu comprises:

sorting the historical statistical evaluation data of each reference song to obtain a data sorting sequence number of each reference song;

sequencing the repeated times of each song in the sub-reference song list to obtain a sequencing sequence number of the times of each song in the sub-reference song list;

determining the songs to be recommended corresponding to each reference song from the sub-reference song list according to the time sequencing sequence number to obtain at least one song to be recommended;

determining the favorite score of the song to be recommended according to the data sorting sequence number and the frequency sorting sequence number of the song to be recommended;

and determining the target song from the at least one song to be recommended according to the preference score.

6. The method of claim 3, wherein after said determining a reference menu from said preset full volume menu based on said reference song, said method further comprises:

according to preset song attributes, quantifying the songs of each of the reference singals to obtain reference attribute characteristics of each of the reference singals;

obtaining scoring data containing scoring results of each song in the reference song list according to a preset song list scoring model and the reference attribute characteristics; the preset singing bill grading model represents the corresponding relation between attribute characteristics and grading data;

and when first scoring data in the scoring data is smaller than a preset scoring threshold value, deleting the singing sheet corresponding to the first scoring data in the reference singing sheet from the reference singing sheet to obtain an updated reference singing sheet.

7. The method as claimed in claim 6, wherein before the obtaining of the scoring data including the scoring result of each of the reference singing tickets according to the preset singing ticket scoring model and the reference attribute feature, the method further comprises:

in a preset training period, obtaining sample operation data of a target user for a full amount of sample songs;

determining a positive sample song list and a negative sample song list from the preset full-volume song list according to the sample operation data;

acquiring positive scoring data corresponding to the positive sample song list, wherein the positive sample song list and the positive scoring data form a positive sample;

acquiring negative scoring data corresponding to the negative sample song list, wherein the negative sample song list and the negative scoring data form a negative sample;

training a preset initial singing bill scoring model by using the positive sample and the negative sample until the accuracy of the trained initial singing bill scoring model is not less than a preset accuracy threshold;

and taking the trained initial singing bill scoring model as the preset singing bill scoring model.

8. An information recommendation apparatus, characterized in that the apparatus comprises: the device comprises an acquisition unit, a calculation unit, an attenuation processing unit and a recommendation unit; wherein,

the acquisition unit is used for acquiring historical operation data of a target user for selecting songs aiming at the full amount of history in a preset statistical period;

the computing unit is used for obtaining historical evaluation data of each historical selected song in the total historical selected songs based on the corresponding relation between preset operation data and operation scores and the historical operation data; the historical evaluation data is used for representing the likeness degree of the target user to the song;

the attenuation processing unit is used for carrying out time attenuation processing on the historical evaluation data to obtain historical statistical evaluation data of each historical selected song;

the recommending unit is used for sequencing the full-amount historical selected songs according to the historical statistical evaluation data to obtain a sequencing result of the full-amount historical selected songs; and recommending the target song according to the sequencing result and the preset full-scale song list.

9. The apparatus according to claim 8, wherein the preset statistical period comprises at least one preset time period; the historical evaluation data comprises time interval evaluation data corresponding to each time interval in the at least one preset time interval;

the attenuation processing unit is specifically configured to obtain an attenuation coefficient corresponding to each time period according to a corresponding relationship between a preset attenuation coefficient and time and the time information represented by each time period; and according to the attenuation coefficient, summing the time interval evaluation data corresponding to the at least one preset time interval to obtain the historical statistical evaluation data.

10. The apparatus of claim 8,

the recommending unit is specifically used for determining a reference song from the full-amount history selected songs according to the sorting result; determining a reference song list from the preset full-volume song list according to the reference song; and recommending the target song according to the reference song list.

11. The apparatus of claim 10, wherein the reference menu comprises sub-reference menus corresponding to each of the reference songs;

and the recommending unit is specifically used for determining the target song from the reference list according to the repetition times of each song in the sub-reference list and recommending the target song.

12. The apparatus of claim 11,

the recommending unit is specifically configured to sort the historical statistical evaluation data of each reference song to obtain a data sorting sequence number of each reference song; sequencing the repeated times of each song in the sub-reference song list to obtain a sequencing sequence number of the times of each song in the sub-reference song list; determining the songs to be recommended corresponding to each reference song from the sub-reference song list according to the time sequencing sequence number to obtain at least one song to be recommended; determining the favorite score of the song to be recommended according to the data sorting serial number and the frequency sorting serial number of the song to be recommended; and determining the target song from the at least one song to be recommended according to the preference score.

13. The apparatus of claim 10,

the recommending unit is further configured to quantify the songs of each of the reference singals according to preset song attributes after the reference singals are determined from the preset total singals according to the reference songs, so as to obtain reference attribute features of each of the reference singals; obtaining scoring data containing scoring results of each song in the reference song list according to a preset song list scoring model and the reference attribute characteristics; the preset singing bill grading model represents the corresponding relation between attribute characteristics and grading data; and when first scoring data in the scoring data is smaller than a preset scoring threshold value, deleting the singing sheet corresponding to the first scoring data in the reference singing sheet from the reference singing sheet to obtain an updated reference singing sheet.

14. The apparatus of claim 13, further comprising:

the model training unit is used for acquiring sample operation data of a target user for a full amount of sample songs in a preset training period before obtaining scoring data containing a scoring result of each song in the reference song list according to a preset song list scoring model and the reference attribute characteristics; determining a positive sample song list and a negative sample song list from the preset full-volume song list according to the sample operation data; acquiring positive scoring data corresponding to the positive sample song list, wherein the positive sample song list and the positive scoring data form a positive sample; acquiring negative scoring data corresponding to the negative sample song list, wherein the negative sample song list and the negative scoring data form a negative sample; training a preset initial singing bill scoring model by using the positive sample and the negative sample until the accuracy of the trained initial singing bill scoring model is not less than a preset accuracy threshold; and taking the trained initial singing bill scoring model as the preset singing bill scoring model.

15. An information recommendation apparatus, characterized in that the apparatus comprises: a processor, a memory and a communication bus, the memory in communication with the processor through the communication bus, the memory storing one or more programs executable by the processor, the one or more programs, when executed, causing the processor to perform the method of any of claims 1-7.

16. A computer-readable storage medium, characterized in that the computer-readable storage medium stores a program which, when executed by at least one processor, causes the at least one processor to perform the method of any one of claims 1-7.