A music content personalized pushing method, system, device and medium

Abstract

This invention relates to the field of personalized recommendation technology, specifically to a method, system, device, and medium for personalized music content push, comprising the following steps: collecting user playlists to generate a behavioral feature matrix, then filtering songs according to playback duration and interaction behavior, generating real genre tags and filtering irrelevant songs to obtain a preference matrix, assigning weighted cumulative ranking to song tags, and generating a music recommendation list based on preference tags from the music library. In this invention, multi-dimensional interaction and content attribute data are collected based on user playlists to construct a refined behavioral feature system. Invalid interference and non-preference-related content are eliminated through dual data filtering, reducing the proportion of noise samples from the source. Combined with the playback ratio and interaction type, differentiated weights are assigned, and related content tags are used to complete the weighted cumulative ranking, accurately anchoring the user's real music aesthetic preferences, significantly improving the matching degree between recommended content and the user's real listening needs, and optimizing the accuracy of music push and user experience.

Description

Technical Field

[0001] This invention relates to the field of personalized recommendation technology, and in particular to a method, system, device and medium for personalized music content delivery. Background Technology

[0002] Personalized recommendation technology encompasses core aspects such as user behavior data collection, music content feature extraction, user preference matching, music content filtering and delivery. This field collects user interaction data with music content, extracts the inherent features of the music content, establishes a matching relationship between users and music content, and delivers music content to users according to the matching results. Traditional personalized music content delivery methods refer to the technical aspects of matching user music preferences and distributing music content. These methods typically collect user behavior data such as playing, collecting, and downloading songs, extracting music genre, rhythm, and artist information features, calculating the similarity between users and music content, and filtering and delivering music based on the similarity ranking results.

[0003] Current personalized music recommendation technologies collect behavioral data using fixed dimensions, covering only three basic interactive behaviors: playback, favorites, and downloads. This fails to distinguish between user interaction scenarios and genuine motivations, making it difficult to identify non-subjective playback behaviors such as background activity or accidental play without switching songs. It also cannot differentiate between active listening in the foreground and passive playback in the background. Furthermore, it easily includes invalid interaction data in preference calculations, generating a large number of noisy samples that interfere with preference recognition accuracy. Current sample processing is undifferentiated, assigning equal weight to compliant behavioral data and failing to differentiate preference confidence levels for different behaviors. It cannot identify high-frequency playback behaviors driven by non-aesthetic motivations, such as idol worship or fan support, and treats chart-topping high-play and high-completion-time data directly as strong positive feedback, leading to fundamental distortion of user profiles and polluting the platform's overall song similarity calculation system. Summary of the Invention

[0004] To address the technical problems existing in the prior art, embodiments of the present invention provide a method for personalized music content recommendation based on user listening behavior, comprising the following steps: S1: Based on the user's playlist, collect the playback duration of each song, the foreground playback duration, genre tags, and interactive behaviors to generate a user listening behavior feature matrix; S2: Based on the user listening behavior feature matrix, remove songs whose playback time is less than half of the total duration of the corresponding song, select songs containing at least two interactive behaviors, extract genre tags from the corresponding songs, generate a set of user real genre tags, extract genre tags from the remaining songs, if all genre tags of the remaining songs do not appear in the set of user real genre tags, then remove the song, and summarize the songs that were not removed to obtain the preferred music matrix. S3: Based on the music preference matrix, assign weight values ​​to each song according to the percentage of playback time, the percentage of front-end playback, and the type of interaction behavior. Associate the tags carried by each song with the weight values ​​of the songs to obtain tag-associated weight data. S4: Accumulate the association weight values ​​corresponding to the same tag, sort the tags from high to low according to the final accumulated results, and obtain the user preference tag ranking results; S5: Based on the user preference tag sorting results, match song features from the music library and sort them to generate a music recommendation sorting list.

[0005] As a further aspect of the present invention, the specific steps of S1 are as follows: S101: Collect the playback duration of songs in the user's playlist, the playback duration in the foreground, genre tags and interactive behaviors, and process the above four types of data into numerical and identifiers respectively. Then, integrate the data corresponding to a single song horizontally to generate a single song behavior feature vector. Interactive behaviors include positive user feedback behaviors such as searching, liking, saving, and commenting; S102: Based on the single song behavior feature vector, arrange the songs in the user's playlist vertically, perform dimension alignment and matrix arrangement on all vectors, establish a data structure corresponding to rows and columns, and obtain the user listening behavior feature matrix.

[0006] As a further aspect of the present invention, the specific steps of S2 are as follows: S201: Based on the user listening behavior feature matrix, compare the song playback duration with half of the total song duration, remove songs that do not meet the duration condition, and obtain a set of songs with valid duration. S202: Based on the set of songs with valid duration, select songs with two or more interactive behaviors, extract the corresponding genre tags and aggregate them to generate a set of real user genre tags. S203: Based on the set of songs with valid duration and the set of user's real genre tags, determine the affiliation relationship between the genre tags of the remaining songs and the tag set, remove songs with tags that have no affiliation, and summarize the retained songs to obtain the preferred music matrix.

[0007] As a further aspect of the present invention, the specific steps of S3 are as follows: S301: Based on the music preference matrix, extract the playback duration of a single song and the total duration of the corresponding song, calculate the proportion of the playback duration in the total duration of the song, and obtain the playback duration proportion value. S302: Based on the music preference matrix, extract the front-end playback duration and corresponding playback duration of a single song, calculate the proportion of the front-end playback duration in the total playback duration, and perform weighted calculations on the values ​​by combining the number of interactive behaviors and the proportion of playback duration to generate a single song weight coefficient. S303: Based on the preference music matrix and the weight coefficient of a single song, the genre tags carried by the song are bound and mapped to the weight coefficient of the single song to establish the correspondence between the tags and the values, and the tag-related weight data is obtained.

[0008] As a further aspect of the present invention, the specific steps of S4 are as follows: S401: Obtain the tag association weight data, traverse the same tags in the tag association weight data, perform an accumulation operation on the weight values ​​corresponding to the same tags, and obtain the accumulated tag weight value. S402: Based on the numerical value of the accumulated tag weights, arrange all tags in descending order of value to determine the priority of each tag, forming an ordered tag sequence, and obtain the user preference tag ranking result.

[0009] As a further aspect of the present invention, the specific steps of S5 are as follows: S501: Based on the user preference tag sorting results, retrieve the genre feature data of all songs in the music library, compare the features of each song with the tags in the sorting one by one, filter out all songs with matching tags, and obtain a set of matching candidate songs.

[0010] As a further aspect of the present invention, the user preference tag sorting results and the matching candidate song set are called, and the candidate songs are arranged in order according to the tag sorting priority to determine the order in which the songs are played and generate a music recommendation sorting list.

[0011] A personalized music content recommendation system based on user listening behavior includes: a behavior feature acquisition module, a preferred track selection module, a tag weight assignment module, a preferred tag sorting module, and a music recommendation generation module; The behavioral feature acquisition module collects the playback duration of each song, the foreground playback duration, genre tags, and interactive behaviors based on the user's playlist, and generates a user listening behavior feature matrix. The preferred track filtering module is based on the user listening behavior feature matrix. It removes songs whose playback time is less than half of the total duration of the corresponding song, selects songs containing at least two interactive behaviors, extracts genre tags from the corresponding songs, generates a set of user real genre tags, and then extracts genre tags from the remaining songs. If none of the genre tags of the remaining songs appear in the user real genre tag set, the song is removed. The preferred music matrix is ​​obtained by summarizing the songs that are not removed. The tag weight assignment module assigns weight values ​​to each song based on the music preference matrix, according to the percentage of playback time, the percentage of front-end playback, and the type of interactive behavior. It associates the tags carried by each song with the weight values ​​of that song to obtain tag-associated weight data. The preference tag sorting module accumulates the association weight values ​​corresponding to the same tag, and sorts them from high to low according to the final accumulated results of each tag to obtain the user preference tag sorting result; The music recommendation generation module matches and sorts song features from the music library based on user preference tag sorting results, and generates a music recommendation sorting list.

[0012] A personalized music content delivery device based on user listening behavior includes: processor; Memory used to store the processor's executable instructions; The processor is used to execute any of the above-described methods for personalized music content delivery based on user listening behavior.

[0013] A computer-readable storage medium storing a computer program for executing any one of the above-described methods for personalized music content delivery based on user listening behavior.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, a refined behavioral feature system is constructed by collecting multi-dimensional interaction and content attribute data from user playlists. Through dual data screening, invalid interference and non-preferred related content are eliminated, and the proportion of noise samples is compressed from the source. Combined with the play ratio and interaction type, differentiated weights are assigned, and related content tags are used to complete the weight accumulation and sorting. This accurately anchors the user's real music aesthetic preferences, greatly improves the matching degree between recommended content and the user's real listening needs, and optimizes the accuracy of music push and user experience. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of the steps of the present invention. Detailed Implementation

[0017] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0018] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0019] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0020] Please see Figure 1 This invention provides a method for personalized music content recommendation based on user listening behavior, comprising the following steps: S1: Based on the user's playlist, collect the playback duration of each song, the foreground playback duration, genre tags, and interactive behaviors to generate a user listening behavior feature matrix; The specific steps of S1 are as follows: S101: Collect the playback duration of songs in the user's playlist, the playback duration in the foreground, genre tags and interactive behaviors, and process the above four types of data into numerical and identifiers respectively. Then, integrate the data corresponding to a single song horizontally to generate a single song behavior feature vector. Interactive behavior data includes positive feedback actions such as searching, liking, saving, and commenting; Using user playlists as the data collection object, this method extracts the actual playback duration data for each song in the playlist, recording the actual duration from the start to the end of playback. It also extracts the foreground playback duration data, recording the effective playback duration of the song while it is in the application's foreground. Furthermore, it extracts genre tag data, unifying the tags for different genre types. Finally, it collects user actions such as searching, liking, favorited, and commenting on corresponding songs, marking each type of action data according to whether it occurred. The method converts both actual playback duration and foreground playback duration—two continuous data types—into standardized values ​​with unified dimensions. Genre tag data is then converted into uniquely identifiable classification codes. Finally, it categorizes search, like, favorited, and comment actions. Comment behavior data is converted into corresponding discrete identifier values. The playback duration value, front-end playback duration value, genre tag code, and interaction behavior identifier value of a single song are concatenated in the same dimension. The concatenated data is combined into a fixed-dimensional data sequence. For a popular style song in a user's playlist, the actual playback duration value is collected, the front-end playback duration value is collected when the actual playback duration is the same, the encoding value corresponding to the genre tag being popular style is collected, and the identifier value corresponding to the "like" behavior of the song is collected. The various values ​​processed above are combined and arranged in sequence to form a single-dimensional sequence structure that can represent the user's listening behavior of the song, thus completing the construction of the single song behavior feature vector. S102: Based on the single song behavior feature vector, arrange the songs in the user's playlist vertically, perform dimension alignment and matrix arrangement on all vectors, establish a data structure corresponding to rows and columns, and obtain the user listening behavior feature matrix. The order in which songs in a user's playlist are played is used as the basis for arrangement. The processed single-song behavioral feature vectors for each song are arranged sequentially according to their playback order. The number of dimensions in all single-song behavioral feature vectors is checked to ensure that the dimension length of each feature vector is consistent. Feature vectors with different number of dimensions are padded to achieve a uniform standard by adding fixed values. All single-song behavioral feature vectors with aligned dimensions are then arranged vertically, so that each row corresponds to the behavioral feature vector of one song, and each column corresponds to the same type of behavioral feature attribute, forming a structured data format with a fixed number of rows and columns. Taking a user's playlist containing multiple songs as an example, the single-song behavioral feature vectors corresponding to each song are arranged vertically sequentially. The number of dimensions in all vectors is uniformly verified and padded. All processed vectors are then arranged in a regular row and column arrangement to form a matrix structure data that can completely represent the user's listening behavior.

[0021] S2: Based on the user listening behavior feature matrix, remove songs whose playback time is less than half of the total duration of the corresponding song, select songs containing at least two interactive behaviors, extract genre tags from the corresponding songs, generate a set of user real genre tags, extract genre tags from the remaining songs, if all genre tags of the remaining songs do not appear in the set of user real genre tags, then remove the song, and summarize the songs that were not removed to obtain the preferred music matrix. The specific steps of S2 are as follows: S201: Based on the user listening behavior feature matrix, compare the song playback duration with half of the total song duration, remove songs that do not meet the duration condition, and obtain a set of songs with valid duration. The playback duration data and total duration data of each song are read from the matrix. The total duration data is substituted into the calculation to obtain half of the value. This value is used as the duration judgment benchmark. The playback duration of each song is compared with this benchmark value item by item. Songs whose playback duration is less than half of the total duration are marked and removed from the current processing set. Unmarked songs that meet the duration conditions are uniformly collected. Taking a specific song as an example, the total duration of the song is a set value, and its half duration is the corresponding calculated value. When the playback duration of the song is greater than or equal to the calculated value, the song is retained. When the playback duration is less than the calculated value, the song is removed. The duration filtering and removal operations of all songs are completed by traversing the feature matrix row by row, forming a set of effective duration songs that only contain songs that meet the duration conditions. S202: Based on the set of songs with valid duration, select songs with two or more interactive behaviors, extract the corresponding genre tags and aggregate them to generate a set of real user genre tags. The interaction behavior fields corresponding to each song in the set are statistically analyzed item by item. The actual number of search, like, favorite, and comment behaviors contained in a single song is counted. Songs with at least two behaviors are selected. The corresponding genre tag data of each selected song is extracted. All extracted genre tag data are deduplicated to remove duplicate tags. The deduplicated genre tags are classified and organized according to a unified format. Taking multiple songs in the effective duration song set as an example, songs with two or more interaction behaviors are identified, and their corresponding genre tags are extracted. The same genre tag that appears repeatedly is kept only once. All non-duplicate genre tags are integrated into a fixed tag group structure to form a set of user real genre tags that can represent the user's real preferences. S203: Based on the set of songs with valid duration and the set of user's real genre tags, determine the affiliation relationship between the genre tags of the remaining songs and the tag set, remove songs with tags that have no affiliation, and summarize the retained songs to obtain the preferred music matrix; In the valid song set, songs that have been included in the real genre tag are distinguished from the remaining songs that have not been included. The genre tags of the remaining songs are extracted one by one, and the genre tags of the remaining songs are matched item by item with each tag in the user's real genre tag set to determine whether the genre tags of the remaining songs exist in the real genre tag set. The remaining songs whose genre tags do not appear in the real genre tag set are marked and removed. The remaining songs with at least one genre tag belonging to the real genre tag set are retained. All the songs that have not been removed are summarized and integrated, and the feature vectors corresponding to these songs are rearranged according to the original matrix structure, keeping the matrix row and column structure and attribute correspondence unchanged. Through the traversal, matching, judgment and removal operations, the final preferred music matrix after multiple layers of screening and verification is obtained.

[0022] S3: Based on the music preference matrix, assign weight values ​​to each song according to the percentage of playback time, the percentage of front-end playback, and the type of interaction behavior. Associate the tags carried by each song with the weight values ​​of the songs to obtain tag-associated weight data. The specific steps for S3 are as follows: S301: Based on the music preference matrix, extract the playback duration of a single song and the total duration of the corresponding song, calculate the proportion of the playback duration in the total duration of the song, and obtain the playback duration proportion value. The system retrieves the playback duration data of a single song and the preset total duration data of the song from the preferred music matrix. It then reads and locates these two types of data, performing a calculation by dividing the playback duration by the total song duration. During the calculation, it maintains consistent data precision and standardizes the results, limiting them to a fixed numerical range to avoid overflow or uneven distribution. For each song, the actual calculation is performed, with the playback duration and total duration being specific values ​​within a set range. The corresponding proportion is obtained by dividing the two, and this proportion is determined as the playback duration percentage of the current song. The same data extraction, division, and standardization operations are performed on all songs in the preferred music matrix to ensure that each song generates a unique playback duration percentage value. This ensures that the calculation method and calculation process for the playback duration percentage values ​​of all songs are consistent, generating a stable and reusable playback duration percentage value for each song. S302: Based on the music preference matrix, extract the front-end playback duration and corresponding playback duration of a single song, calculate the proportion of the front-end playback duration in the total playback duration, and perform weighted calculations on the values ​​by combining the number of interactive behaviors and the proportion of playback duration to generate a single song weight coefficient. The front-end playback duration and corresponding playback duration of each song are extracted from the music preference matrix. The front-end playback duration is divided by the actual playback duration to obtain the front-end playback percentage. This percentage is then normalized to ensure it falls within a fixed range, preventing magnitude differences from affecting subsequent calculations. The previously calculated playback duration percentage for each song is extracted. The number of interactive behaviors corresponding to the song is counted, including searches, likes, favorites, and comments. Each type of behavior is counted as a valid count, and the result is a non-negative integer. A weighted summation formula is used to comprehensively calculate the sum of these three types of values. The weighting formula is as follows: ,in Let be the single-song weight coefficient for the i-th song. For the percentage of playback on the front end, This represents the percentage of playback time. For the number of interactive behaviors, , , The weight coefficients of the corresponding parameters are respectively and satisfy the following conditions: + + =1, substitute the corresponding value into the formula to complete the weighting operation, and perform interval constraint processing on the result so that the final generated single song weight coefficient is within a reasonable range. Perform the same extraction, calculation, weighting, and constraint operations on all songs in the preference music matrix one by one to obtain the single song weight coefficient corresponding to each song. S303: Based on the preference music matrix and the weight coefficient of a single song, the genre tags carried by the song are bound and mapped to the weight coefficient of the single song to establish the correspondence between the tags and the values, and the tag-related weight data is obtained. Using the song information in the preference music matrix and the corresponding calculated single-song weight coefficients as the processing objects, the system iterates through the genre tag information attached to each song in the matrix, binding all genre tags corresponding to a single song with the weight coefficient of that song one by one, establishing a direct mapping relationship between tags and weight values. For cases where the same tag corresponds to multiple songs, each independent mapping relationship is retained without pre-merging or averaging, ensuring that the correspondence between tags and weights is complete and traceable. The mapping results of all song genre tags and weight coefficients are uniformly organized and stored in a structured manner according to a fixed format, recording the tag content and associated weight value line by line, forming complete and directly callable tag-associated weight data.

[0023] S4: Accumulate the association weight values ​​corresponding to the same tag, sort the tags from high to low according to the final accumulated results, and obtain the user preference tag ranking results; S401: Obtain the tag association weight data, traverse the same tags in the tag association weight data, perform an accumulation operation on the weight values ​​corresponding to the same tags, and obtain the accumulated tag weight value. All previously generated tag-related weight data were retrieved, and a dedicated data traversal framework was built. The content of the genre tag and the corresponding associated weight value in each data entry were read line by line. The entire data entry was traversed without omission, ensuring that no tag-weight correspondence was missed. A tag classification ledger was built simultaneously, and each genre tag was accurately matched and identified. Tags with completely identical content were grouped into the same category. For all associated weight values ​​under the same category, a cumulative calculation was performed, and each weight value under the same tag was summed up in turn. The calculation process retained complete decimal precision to avoid deviations in the numerical accumulation. For tags with different frequencies in the dataset, a unified classification, traversal, and accumulation process was performed, without special processing for individual tags. After the full tag processing was completed, the accumulation results of each category of tags were checked one by one to ensure that the values ​​were accurate. Finally, a unique tag weight accumulation value was generated for each category of tags. S402: Based on the numerical value of the accumulated tag weights, arrange all tags in descending order of value to determine the priority of each tag, forming an ordered tag sequence, and obtain the user preference tag ranking result; Extract all tags and their corresponding calculated cumulative tag weights. Bind each tag category to its cumulative weight one-to-one, forming an independent tag-weight data unit. Using the cumulative weight of each tag category as the core sorting criterion, initiate a descending sorting process. Compare the cumulative values ​​of all tags, placing tags with higher values ​​at the beginning of the sequence and those with lower values ​​at the end. If two or more tags have identical cumulative values, determine their order based on their first appearance in the associated weight data, without adjusting the original order. After sorting, organize and integrate all tags according to the predetermined order to form a continuous and ordered tag sequence. Preserve the relationship between the original tag content and its corresponding cumulative value throughout the process. Perform a comprehensive verification of the sorted sequence to confirm that the sorting logic is compliant and that no tags are missing or misplaced, ultimately forming a complete and standardized user preference tag sorting result.

[0024] S5: Based on the user preference tag sorting results, match song features from the music library and sort them to generate a music recommendation sorting list; S501: Based on the user preference tag sorting results, retrieve the genre feature data of all songs in the music library, compare the features of each song with the tags in the sorting one by one, filter out all songs with matching tags, and obtain a set of matching candidate songs. The system retrieves the generated user preference tag sorting results, synchronously connects to the backend music library database, extracts the complete genre feature data of each existing song in the music library, decomposes the genre features of each song into independent tag items, and then conducts precise comparisons with each tag in the user preference tag sorting results. The comparison process covers all related tags of the song, without omitting any feature information. All songs with at least one tag overlapping with the user preference tag are filtered out. The filtered songs are uniformly collected, and music library songs without any tag matching are removed. After collection, the candidate songs are initially deduplicated to remove duplicate entries of the same song. The entire process is carried out according to fixed comparison standards, without arbitrarily relaxing or tightening the matching conditions, and finally obtains a set of matching candidate songs with no duplicates and complete tag matching. The system retrieves the user preference tag sorting results and the matching candidate song set, arranges the candidate songs in order of tag sorting priority, determines the order of song playback, and generates a music recommendation sorting list. The system synchronously retrieves the user preference tag sorting results and the matching candidate song set, using tag sorting priority as the core arrangement basis. Each song in the candidate song set is labeled with the corresponding matching tag position, prioritizing the arrangement of songs matching higher-ranking preference tags. If the same song matches multiple tags, the highest-ranking tag is used to determine the basic sorting position. For multiple songs matching the same rank tag, they are arranged sequentially according to the order in which the songs were entered in the music library. The position labeling and order arrangement of all candidate songs are completed one by one, forming a coherent song sequence. Throughout the arrangement process, the song information and tag matching relationship are checked to avoid sorting misalignment or song omissions. The final song sequence is formatted and uniformly labeled with song numbers and corresponding matching tag information, forming a music recommendation sorting list with a clear structure and fixed order.

[0025] A personalized music content recommendation system based on user listening behavior includes: a behavior feature acquisition module, a preferred track selection module, a tag weight assignment module, a preferred tag sorting module, and a music recommendation generation module; The behavioral feature collection module collects the playback duration of each song, the foreground playback duration, genre tags, and interactive behaviors based on the user's playlist, and generates a user listening behavior feature matrix. The preferred track selection module is based on the user listening behavior feature matrix. It removes songs whose playback time is less than half of the total duration of the corresponding song, selects songs containing at least two interactive behaviors, extracts genre tags from the corresponding songs, generates a set of user real genre tags, and then extracts genre tags from the remaining songs. If all genre tags of the remaining songs do not appear in the user real genre tag set, the song is removed. The preferred music matrix is ​​obtained by summarizing the songs that are not removed. The tag weight assignment module is based on the music preference matrix. It assigns weight values ​​to each song according to the percentage of playback time, the percentage of front-end playback, and the type of interaction behavior. It associates the tags carried by each song with the weight values ​​of that song to obtain tag-associated weight data. The preference tag sorting module accumulates the associated weight values ​​corresponding to the same tag, and sorts them from high to low according to the final accumulated results of each tag to obtain the user preference tag sorting results; The music recommendation generation module matches and sorts songs from the music library based on user preference tags, generating a music recommendation ranking list.

[0026] A personalized music content delivery device based on user listening behavior includes: processor; Memory used to store processor-executable instructions; The processor is used to execute any of the above-mentioned methods for personalized music content delivery based on user listening behavior.

[0027] A computer-readable storage medium storing a computer program for performing any of the above-mentioned methods for personalized music content delivery based on user listening behavior.

[0028] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for personalized music content recommendation based on user listening behavior, characterized in that, Includes the following steps: S1: Based on the user's playlist, collect the playback duration of each song, the foreground playback duration, genre tags, and interactive behaviors to generate a user listening behavior feature matrix; S2: Based on the user listening behavior feature matrix, remove songs whose playback time is less than half of the total duration of the corresponding song, select songs containing at least two interactive behaviors, extract genre tags from the corresponding songs, generate a set of user real genre tags, extract genre tags from the remaining songs, if all genre tags of the remaining songs do not appear in the set of user real genre tags, then remove the song, and summarize the songs that were not removed to obtain the preferred music matrix. S3: Based on the music preference matrix, assign weight values ​​to each song according to the percentage of playback time, the percentage of front-end playback, and the type of interaction behavior. Associate the tags carried by each song with the weight values ​​of the songs to obtain tag-associated weight data. S4: Accumulate the association weight values ​​corresponding to the same tag, sort the tags from high to low according to the final accumulated results, and obtain the user preference tag ranking results; S5: Based on the user preference tag sorting results, match song features from the music library and sort them to generate a music recommendation sorting list.

2. The method for personalized music content recommendation based on user listening behavior according to claim 1, characterized in that, The specific steps of S1 are as follows: S101: Collect the playback duration of songs in the user's playlist, the playback duration in the foreground, genre tags and interactive behaviors, and process the above four types of data into numerical and identifiers respectively. Then, integrate the data corresponding to a single song horizontally to generate a single song behavior feature vector. Interactive behaviors include positive user feedback behaviors such as searching, liking, saving, and commenting; S102: Based on the single song behavior feature vector, arrange the songs in the user's playlist vertically, perform dimension alignment and matrix arrangement on all vectors, establish a data structure corresponding to rows and columns, and obtain the user listening behavior feature matrix.

3. The method for personalized music content recommendation based on user listening behavior according to claim 2, characterized in that, The specific steps of S2 are as follows: S201: Based on the user listening behavior feature matrix, compare the song playback duration with half of the total song duration, remove songs that do not meet the duration condition, and obtain a set of songs with valid duration. S202: Based on the set of songs with valid duration, select songs with two or more interactive behaviors, extract the corresponding genre tags and aggregate them to generate a set of real user genre tags. S203: Based on the set of songs with valid duration and the set of user's real genre tags, determine the affiliation relationship between the genre tags of the remaining songs and the tag set, remove songs with tags that have no affiliation, and summarize the retained songs to obtain the preferred music matrix.

4. The method for personalized music content recommendation based on user listening behavior according to claim 3, characterized in that, The specific steps for S3 are as follows: S301: Based on the music preference matrix, extract the playback duration of a single song and the total duration of the corresponding song, calculate the proportion of the playback duration in the total duration of the song, and obtain the playback duration proportion value. S302: Based on the music preference matrix, extract the front-end playback duration and corresponding playback duration of a single song, calculate the proportion of the front-end playback duration in the total playback duration, and perform weighted calculations on the values ​​by combining the number of interactive behaviors and the proportion of playback duration to generate a single song weight coefficient. S303: Based on the preference music matrix and the weight coefficient of a single song, the genre tags carried by the song are bound and mapped to the weight coefficient of the single song to establish the correspondence between the tags and the values, and the tag-related weight data is obtained.

5. The method for personalized music content recommendation based on user listening behavior according to claim 4, characterized in that, The specific steps of S4 are as follows: S401: Obtain the tag association weight data, traverse the same tags in the tag association weight data, perform an accumulation operation on the weight values ​​corresponding to the same tags, and obtain the accumulated tag weight value. S402: Based on the numerical value of the accumulated tag weights, arrange all tags in descending order of value to determine the priority of each tag, forming an ordered tag sequence, and obtain the user preference tag ranking result.

6. The method for personalized music content recommendation based on user listening behavior according to claim 5, characterized in that, The specific steps of S5 are as follows: S501: Based on the user preference tag sorting results, retrieve the genre feature data of all songs in the music library, compare the features of each song with the tags in the sorting one by one, filter out all songs with matching tags, and obtain a set of matching candidate songs.

7. The method for personalized music content recommendation based on user listening behavior according to claim 6, characterized in that, The user preference tag sorting results and the matching candidate song set are called, and the candidate songs are arranged in order according to the tag sorting priority to determine the order in which the songs are played and generate a music recommendation sorting list.

8. A personalized music content recommendation system based on user listening behavior, characterized in that, The system is used to implement the personalized music content push method based on user listening behavior as described in any one of claims 1-7. The system includes: a behavior feature acquisition module, a preferred track filtering module, a tag weight assignment module, a preferred tag sorting module, and a music push generation module. The behavioral feature acquisition module collects the playback duration of each song, the foreground playback duration, genre tags, and interactive behaviors based on the user's playlist, and generates a user listening behavior feature matrix. The preferred track filtering module is based on the user listening behavior feature matrix. It removes songs whose playback time is less than half of the total duration of the corresponding song, selects songs containing at least two interactive behaviors, extracts genre tags from the corresponding songs, generates a set of user real genre tags, and then extracts genre tags from the remaining songs. If none of the genre tags of the remaining songs appear in the user real genre tag set, the song is removed. The preferred music matrix is ​​obtained by summarizing the songs that are not removed. The tag weight assignment module assigns weight values ​​to each song based on the music preference matrix, according to the percentage of playback time, the percentage of front-end playback, and the type of interactive behavior. It associates the tags carried by each song with the weight values ​​of that song to obtain tag-associated weight data. The preference tag sorting module accumulates the association weight values ​​corresponding to the same tag, and sorts them from high to low according to the final accumulated results of each tag to obtain the user preference tag sorting result; The music recommendation generation module matches and sorts song features from the music library based on user preference tag sorting results, and generates a music recommendation sorting list.

9. A personalized music content push device based on user listening behavior, characterized in that, include: processor; Memory used to store the processor's executable instructions; The processor is used to execute a personalized music content push method based on user listening behavior as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which is used to execute the personalized music content push method based on user listening behavior as described in any one of claims 1-7.

Images