Music recommendation method and device based on multiple platforms, equipment and storage medium

Abstract

The invention relates to the technical field of music recommendation, and discloses a music recommendation method and device based on multiple platforms, equipment and a storage medium, which are used for improving the accuracy of music recommendation. The multi-platform-based music recommendation method comprises the steps of obtaining multi-platform music data, wherein the multi-platform music data comprises music listening histories of a user collected from at least two different music platforms; performing metadata completion processing on the music data of the multiple platforms to obtain target music data; performing cross-platform user portrait construction according to the target music data; and performing personalized recommendation algorithm processing according to the constructed cross-platform user portrait to obtain a target recommended music list.

Description

Technical Field

[0001] The present invention relates to the technical field of music recommendation, and in particular to a multi-platform-based music recommendation method, device, equipment and storage medium. Background Art

[0002] The recommendation algorithms used by most platforms are based only on the user's behavioral data on a single platform, ignoring the user's preferences and collections on other platforms, resulting in one-sided and difficult-to-personalize recommendation effects. Users' music subscription content, personal music library and recommendation system are often independent modules, lacking an effective integration mechanism. This separation limits the user experience and makes it difficult for users to obtain unified services between different platforms, resulting in inaccurate music recommendation results and a large deviation between the recommendation results and the user's immediate needs. Summary of the invention

[0003] The present invention provides a multi-platform music recommendation method, device, equipment and storage medium to solve the problem in the prior art that the accuracy of music recommendation is low due to limitation to single platform data and lack of cross-platform integration.

[0004] A first aspect of the present invention provides a multi-platform music recommendation method, comprising: obtaining multi-platform music data, the multi-platform music data including a user's music listening history collected from at least two different music platforms; performing metadata completion processing on the multi-platform music data to obtain target music data; constructing a cross-platform user portrait based on the target music data; performing personalized recommendation algorithm processing based on the constructed cross-platform user portrait to obtain a target recommended music list.

[0005] In a feasible implementation, the acquisition of music data for multiple platforms includes: determining the target music platforms for which music data needs to be collected; extracting music data for each target music platform according to the type of each target music platform and the corresponding extraction strategy, to obtain music data for multiple platforms.

[0006] In a feasible implementation, the music data of each target music platform is extracted according to the type of each target music platform and the corresponding extraction strategy to obtain music data of multiple platforms, including: when the type of the target music platform is a first music platform that supports API, an API request is sent to the first music platform, and the response data returned by the first music platform is parsed to obtain the music data of the first music platform; when the type of the target music platform is a second music platform that does not support API, a crawler program is used to capture corresponding data to obtain the music data of the second music platform; when the target music platform is a third music platform where a local database is located, corresponding data is extracted from the local database to obtain the music data of the third music platform.

[0007] In a feasible implementation, metadata completion processing is performed on the music data of the multiple platforms to obtain target music data, including: extracting audio features of each music track from the music data of the multiple platforms; performing similarity matching in a preset database based on the audio features of each music track, and removing duplicate music tracks to obtain multiple first music tracks, where the first music tracks are music tracks included in the music data of the multiple platforms; and using data sources, text analysis technology and the user's music listening history to complete the metadata of each first music track to obtain target music data.

[0008] In a feasible implementation manner, the similarity matching is performed in a preset database based on the audio features of each music track, and duplicate music tracks are removed to obtain multiple first music tracks, including: in a preset music fingerprint database, using a local sensitive hashing algorithm, similarity matching is performed based on the audio features of each music track, and duplicate music tracks are removed to obtain multiple first music tracks.

[0009] In a feasible implementation manner, the use of data sources, text analysis technology and the user's music listening history to complete the metadata of each first music track to obtain target music data includes: obtaining first metadata of each first music track from multiple online data sources; using natural language processing technology to perform deep text analysis on the lyrics of each first music track to extract second metadata of each first music track; based on each first music track, analyzing according to the user's music listening history to obtain third metadata of each first music track; performing metadata completion processing on each first music track according to the first metadata, the second metadata and the third metadata to obtain target music data.

[0010] In a feasible implementation, the cross-platform user portrait is constructed based on the target music data, including: extracting music information and user behavior information from the target music data to obtain target information; analyzing the corresponding behavior feature vector, preference model and situational awareness model based on the target information; and constructing a cross-platform user portrait based on the behavior feature vector, the preference model and the situational awareness model, wherein the cross-platform user portrait comprehensively reflects the user's behavior characteristics, music preferences and contextualized music needs on multiple music platforms.

[0011] In a feasible implementation, the corresponding behavior feature vector, preference model and situational awareness model are analyzed based on the target information, including: data cleaning and formatting of the target information to remove duplicate, invalid or abnormal data to form a unified behavior feature vector; using collaborative filtering and content-based hybrid recommendation models, as well as topic models to conduct in-depth analysis of the target information to construct a preference model that reflects the user's music preference; based on the target information, using a conditional random field model to model the user's music needs in different situations to obtain a situational awareness model that can predict the user's situational music needs.

[0012] In a feasible implementation, the personalized recommendation algorithm is processed according to the constructed cross-platform user portrait to obtain a target recommended music list, including: extracting a latent vector reflecting the user's cross-platform music taste according to the cross-platform user portrait; constructing a music feature vector for each second music track according to the music content of the local database, the second music track being the music track contained in the local database; calculating the similarity between each second music track and the user's preference according to the latent vector and each music feature vector set to obtain a music candidate set; generating a target recommended music list according to the music candidate set in combination with at least one music prediction model.

[0013] In a feasible implementation, the extracting a latent vector reflecting the user's cross-platform music taste based on the cross-platform user portrait includes: extracting features reflecting the user's music taste from the cross-platform user portrait, and quantizing them to obtain a target feature vector; performing matrix decomposition on the target feature vector, and extracting the user's latent vector from the result of the matrix decomposition.

[0014] In a feasible implementation, the similarity between each second music track and the user preference is calculated based on the latent vector and each music feature vector to obtain a music candidate set, including: calculating the similarity between each second music track and the user preference using a cosine similarity algorithm based on the latent vector and each music feature vector; sorting all second music tracks according to the calculated similarities, and generating a music candidate set based on the sorting results.

[0015] In a feasible implementation, when there are two or more music prediction models, the target recommended music list is generated based on the music candidate set in combination with at least one music prediction model, including: assigning weights to at least one music prediction model; processing the music candidate set according to the at least one music prediction model to obtain prediction results of each music prediction model; and analyzing the weights and prediction results of each music prediction model to obtain the target recommended music list.

[0016] The second aspect of the present invention provides a multi-platform music recommendation device, including: an acquisition module, used to acquire music data from multiple platforms, the music data from multiple platforms including the user's music listening history, preference information and interaction records collected from at least two different music platforms; a first processing module, used to perform metadata completion processing on the music data from multiple platforms to obtain target music data; a construction module, used to construct a cross-platform user portrait based on the target music data; and a second processing module, used to perform personalized recommendation algorithm processing based on the constructed cross-platform user portrait to obtain a target recommended music list.

[0017] In a feasible implementation, the acquisition module includes: a determination unit, used to determine the target music platform for which music data needs to be collected; a first extraction unit, used to extract music data from each target music platform according to the type of each target music platform and the corresponding extraction strategy, to obtain music data for multiple platforms.

[0018] In a feasible implementation, the first extraction unit is specifically used for: when the type of the target music platform is a first music platform that supports API, sending an API request to the first music platform, parsing the response data returned by the first music platform, and obtaining the music data of the first music platform; when the type of the target music platform is a second music platform that does not support API, using a crawler program to capture corresponding data to obtain the music data of the second music platform; when the target music platform is a third music platform where the local database is located, extracting corresponding data from the local database to obtain the music data of the third music platform.

[0019] In a feasible implementation, the first processing module includes: a second extraction unit, used to extract audio features of each music track from the music data of the multiple platforms; a matching unit, used to perform similarity matching in a preset database based on the audio features of each music track, and remove duplicate music tracks to obtain multiple first music tracks, where the first music tracks are music tracks included in the music data of the multiple platforms; a completion unit, used to complete the metadata of each first music track using the data source, text analysis technology and the user's music listening history to obtain target music data.

[0020] In a feasible implementation, the matching unit is specifically used to: use a local sensitive hashing algorithm in a preset music fingerprint database to perform similarity matching according to audio features of each music track, and remove duplicate music tracks to obtain multiple first music tracks.

[0021] In a feasible implementation, the completion unit is specifically used to: obtain first metadata for each first music track from multiple online data sources; use natural language processing technology to perform deep text analysis on the lyrics of each first music track to extract second metadata for each first music track; based on each first music track, analyze according to the user's music listening history to obtain third metadata for each first music track; perform metadata completion processing on each first music track according to the first metadata, the second metadata and the third metadata to obtain target music data.

[0022] In a feasible implementation, the construction module includes: a third extraction unit, used to extract music information and user behavior information from the target music data to obtain target information; an analysis unit, used to analyze the corresponding behavior feature vector, preference model and situational perception model based on the target information; a first construction unit, used to construct a cross-platform user portrait based on the behavior feature vector, the preference model and the situational perception model, and the cross-platform user portrait comprehensively reflects the user's behavior characteristics, music preferences and contextualized music needs on multiple music platforms.

[0023] In a feasible implementation, the analysis unit is specifically used to: perform data cleaning and formatting on the target information, remove duplicate, invalid or abnormal data, and form a unified behavioral feature vector; use a hybrid recommendation model based on collaborative filtering and content, and a topic model to conduct in-depth analysis of the target information and construct a preference model that reflects the user's music preference; based on the target information, use a conditional random field model to model the user's music needs in different situations, and obtain a situational awareness model that can predict the user's situational music needs.

[0024] In a feasible implementation, the second processing module includes: a fourth extraction unit, used to extract a latent vector reflecting the user's cross-platform music taste based on the cross-platform user portrait; a second construction unit, used to construct a music feature vector of each second music track based on the music content of the local database, and the second music track is a music track contained in the local database; a calculation unit, used to calculate the similarity between each second music track and the user's preference based on the latent vector and each music feature vector set, and obtain a music candidate set; a generation unit, used to generate a target recommended music list based on the music candidate set in combination with at least one music prediction model.

[0025] In a feasible implementation, the fourth extraction unit is specifically used to: extract features reflecting the user's music taste from the cross-platform user portrait, and perform quantization processing to obtain a target feature vector; perform matrix decomposition on the target feature vector, and extract the user's latent vector from the result of the matrix decomposition.

[0026] In a feasible implementation, the calculation unit is specifically used to: calculate the similarity between each second music track and the user preference using a cosine similarity algorithm based on the latent vector and each music feature vector; sort all second music tracks according to the calculated similarity, and generate a music candidate set based on the sorting result.

[0027] In a feasible implementation, the generation unit is specifically used to: when there are two or more music prediction models, assign weights to at least one music prediction model; process the music candidate set according to the at least one music prediction model to obtain prediction results of each music prediction model; and analyze the weights and prediction results of each music prediction model to obtain a target recommended music list.

[0028] A third aspect of the present invention provides a multi-platform music recommendation device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the multi-platform music recommendation device executes the above-mentioned multi-platform music recommendation method.

[0029] A fourth aspect of the present invention provides a computer-readable storage medium, in which instructions are stored, which, when executed on a computer, enable the computer to execute the above-mentioned multi-platform-based music recommendation method.

[0030] In the technical solution provided by the present invention, music data from multiple platforms are obtained, and the music data from multiple platforms include the user's music listening history collected from at least two different music platforms; metadata completion processing is performed on the music data from multiple platforms to obtain target music data; a cross-platform user portrait is constructed based on the target music data; and a personalized recommendation algorithm is processed based on the constructed cross-platform user portrait to obtain a target recommended music list. In an embodiment of the present invention, a detailed cross-platform user portrait is constructed by collecting and integrating the user's music data across platforms and after metadata completion processing, thereby generating a target recommended music list, which fully reflects the user's music preferences and immediate needs, and significantly improves the personalization and accuracy of music recommendations. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 A schematic diagram of an embodiment of a music recommendation method based on multiple platforms in an embodiment of the present invention; Figure 2 It is a schematic diagram of another embodiment of the music recommendation method based on multiple platforms in an embodiment of the present invention; Figure 3 It is a schematic diagram of another embodiment of the music recommendation method based on multiple platforms in an embodiment of the present invention; Figure 4 It is a schematic diagram of another embodiment of the music recommendation method based on multiple platforms in an embodiment of the present invention; Figure 5 A schematic diagram of an embodiment of a music recommendation device based on multiple platforms in an embodiment of the present invention; Figure 6 It is a schematic diagram of another embodiment of a multi-platform music recommendation device according to an embodiment of the present invention; Figure 7 Schematic diagram of an embodiment of a multi-platform music recommendation device in an embodiment of the present invention. DETAILED DESCRIPTION

[0032] The embodiments of the present invention provide a multi-platform music recommendation method, apparatus, device and storage medium, which achieve more accurate and personalized music recommendations by integrating user music data across platforms and building user portraits.

[0033] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0034] It is understandable that the execution subject of the present invention may be a music recommendation device based on multiple platforms, or may be a terminal or a server, which is not limited here. The embodiment of the present invention is described by taking the server as the execution subject as an example.

[0035] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 , an embodiment of the multi-platform music recommendation method in the embodiment of the present invention includes: 101. Acquire music data from multiple platforms, where the music data from multiple platforms includes a user's music listening history collected from at least two different music platforms; With the user's explicit consent and authorization, data communication connections are established with various music platforms through preset API interfaces, data exchange protocols or user authorization, and the user's music listening history data is synchronized accordingly; the synchronized data is cleaned, deduplicated and anonymized to ensure the accuracy, uniqueness and protection of user privacy; when storing and processing data, relevant data security and privacy protection regulations are strictly followed, as well as the scope of user authorization, to ensure the legality and security of the data; the user's music listening history data is updated from various music platforms regularly or based on preset trigger conditions, such as changes in user behavior or new music releases, to keep the data timely and accurate.

[0036] The multi-platform music data may include music data from digital music service platforms, such as music data from online music streaming media and music download websites; it may also include music data from physical media, such as music data downloaded and converted from CDs.

[0037] 102. Perform metadata completion processing on music data of multiple platforms to obtain target music data; Identify and mark missing or inconsistent metadata fields in the music data of each platform, and establish a unified metadata standard that covers key information such as song name, singer, album, genre, release year, lyrics, cover image, etc.; according to the metadata standard, convert and standardize the music data obtained from each platform to ensure the consistency of all data in metadata fields and formats; in the standardization process, compare the music data of multiple platforms and use algorithms to automatically fill in the missing metadata. For example, find matching songs on other platforms through song names and singer information and obtain complete metadata; for metadata that cannot be automatically filled in, provide manual review and manual completion interfaces to ensure the integrity and accuracy of the target music data; at the same time, in the process of completing metadata, mark and track the source of the data to facilitate subsequent data quality assessment and error troubleshooting.

[0038] Through the establishment of metadata standards, data format conversion and standardization, the combination of automatic filling and manual review, and the importance of data tracking, a more comprehensive and detailed metadata completion solution is provided for multi-platform music recommendation methods.

[0039] In one embodiment, lyrics service platforms such as LyricFind and Musixmatch can be used to input the song name and singer information in the music data to be completed, so as to obtain the corresponding lyrics content. These platforms have a huge lyrics database and cooperate with multiple music streaming platforms to provide accurate and real-time lyrics information. After obtaining the lyrics, the lyrics content is filled into the corresponding fields of the music data. In addition, core metadata such as songs, singers, albums, genres and release years can be obtained and checked through metadata platforms such as MusicBrainz and Discogs. These platforms provide detailed music databases and metadata standards, which can help ensure the accuracy and consistency of data. Consistency, using their search function, according to key information such as song name, singer name, etc., to find and obtain the corresponding metadata in the database. After obtaining the metadata, the data can be checked and cleaned to ensure the accuracy and completeness of the data. Furthermore, for audio files that cannot directly obtain metadata, Shazam and other audio recognition technologies can be used to identify songs. Shazam uses audio fingerprint technology to quickly and accurately identify the song being played and return detailed information about the song. Using Shazam's recognition results, find and complete the missing metadata on other music platforms. At the same time, audio processing technology can also be used to pre-process the audio files to improve Shazam's recognition accuracy.

[0040] During the entire metadata completion process, a combination of automation and manual review can be used. For metadata that can be automatically acquired and verified, automated processing can be used to improve processing efficiency. For metadata that cannot be automatically acquired or requires manual verification, manual review and manual completion interfaces are provided to ensure the accuracy and completeness of the data. At the same time, the data source can also be marked and tracked to facilitate subsequent data quality assessment and error troubleshooting.

[0041] By using platforms such as LyricFind, Musixmatch, MusicBrainz, Discogs and Shazam, we have achieved comprehensive completion of music data metadata. Not only has it automatically and efficiently filled in key fields such as lyrics and album information, it has also solved the problem of not being able to directly obtain metadata through audio recognition technology. At the same time, combined with manual review, it ensures the high accuracy of metadata and the clear traceability of data sources, thereby greatly improving the quality and integrity of music data.

[0042] 103. Build cross-platform user portraits based on target music data; According to the target music data, the user's preferred tracks, favorite lists, playlists, interactive behaviors, social information and geographic location information are integrated and analyzed, and key features are extracted and normalized and weighted to eliminate data bias between platforms and reflect the importance of features, where interactive behaviors include but are not limited to likes, comments and shares, and social information includes but is not limited to follow lists, number of fans and music communities; machine learning algorithms are applied to process the normalized features, and combined with time dimension analysis and regional music preference analysis, the user's music preferences, behavior patterns and potential needs are identified to build a cross-platform user portrait with personalized, social and time dimensions. The machine learning algorithm includes but is not limited to cluster analysis, collaborative filtering or deep learning models, and time dimension analysis, such as the differences in music consumption behaviors between weekdays and weekends, daytime and nighttime; a dynamic update mechanism for user portraits is provided, and user portraits are regularly adjusted according to the user's newly generated music behavior data to maintain their timeliness and accuracy, and users are allowed to view and edit their own portrait information through a visual interface to improve user engagement and satisfaction.

[0043] By eliminating data bias between platforms, extracting key features and normalizing and weighting them, and combining machine learning algorithms, time dimension analysis, and regional music preference analysis, we have constructed a user portrait with personalization, social dimensions, and time dimensions. This not only improves the accuracy and timeliness of the user portrait, but also enhances user engagement through a visual interface, allowing users to view and edit their own portrait information, thereby improving user satisfaction and experience.

[0044] 104. Perform personalized recommendation algorithm processing based on the constructed cross-platform user portrait to obtain a target recommended music list.

[0045] Taking cross-platform user portraits as input, using collaborative filtering algorithms and deep learning models, combined with matrix decomposition technology, deep learning and analysis of users' historical behavior data and key features are carried out, and users' preferences and ratings for new music are calculated to capture users' complex preferences and behavior patterns; using time series analysis technology and regional music preference analysis, the recommendation list is dynamically adjusted according to users' music consumption behaviors in different time periods and music trends in their geographical locations to adapt to changes in users' music preferences and local music tastes; the musicians or music communities that users follow, as well as users' social interaction behaviors, are analyzed to recommend relevant music content to users, and a diversity enhancement strategy is adopted to introduce music that is different from users' known preferences into the recommendation list to increase the diversity and novelty of the recommendations; users are allowed to rate, like or comment on the recommended music to collect users' feedback on the recommendation results, which is used to optimize the recommendation algorithm and realize dynamic updating of the recommendation list. The recommendation list is regularly adjusted according to users' newly generated music behavior data to ensure the timeliness and accuracy of the recommendations.

[0046] In an embodiment of the present invention, by collecting and integrating the user's music data across platforms and after metadata completion processing, a detailed cross-platform user portrait is constructed, thereby generating a target recommended music list, which fully reflects the user's music preferences and immediate needs, and significantly improves the personalization and accuracy of music recommendations.

[0047] See also Figure 2 Another embodiment of the multi-platform music recommendation method in the embodiment of the present invention includes: 201. Determine the target music platform for which music data needs to be collected; Identify the popular music platforms with a large user base in the current market, which may include online music streaming services, music modules on social media platforms, music download stores, and music sharing and discovery platforms; based on whether these platforms provide public API interfaces, divide them into two categories: platforms that support API interfaces and platforms that do not support API interfaces.

[0048] 202. Extracting music data from each target music platform according to the type of each target music platform and the corresponding extraction strategy to obtain music data for multiple platforms, where the music data for multiple platforms includes the user's music listening history collected from at least two different music platforms; When the type of the target music platform is the first music platform that supports API, an API request is sent to the first music platform, and the response data returned by the first music platform is parsed to obtain the music data of the first music platform; when the type of the target music platform is the second music platform that does not support API, the corresponding data is captured by a crawler program to obtain the music data of the second music platform; when the target music platform is the third music platform where the local database is located, the corresponding data is extracted from the local database to obtain the music data of the third music platform.

[0049] For the first music platform that supports API interface, the music data is obtained by directly sending API requests. According to the platform's API documentation, design and build API requests that meet data requirements. These requests usually contain specific parameters and request headers to specify the type and format of the data to be obtained. Send these requests to the server of the first music platform programmatically. Once the request is successfully received and processed, the first music platform will return a response containing the requested music data. This response may be a JSON object, XML document or other format of data. Use the corresponding parsing technology to parse this response and extract the required music data, such as song information, artist details, playlists, etc., and store this data in a local database or cloud storage solution for subsequent analysis and processing.

[0050] For the second music platform that does not support API interface, web crawler technology is used to collect data. Programming languages ​​​​such as Python can be used in combination with crawler frameworks such as Selenium, BeautifulSoup or Scrapy to design and develop web crawler programs. These programs can simulate user behavior, parse web page structure, and extract the required music data. In the process of data crawling, special attention is paid to the response of anti-crawler mechanisms, such as using proxy IPs, simulating user operations and other strategies to ensure the success rate and efficiency of data crawling. Strict data quality control measures can also be implemented, including data verification, outlier processing and data deduplication, to ensure the accuracy and consistency of the collected data.

[0051] For the third music platform where the local database is located, directly extract the corresponding data from the local database. Make sure that the local database already contains the required data and that the data is up-to-date and complete. Use the database query language to write query statements, specify the data type and conditions you want to extract, and by executing these query statements, you can retrieve the corresponding music data from the local database and load it into memory or store it in another location for subsequent processing. The third music platform is the music platform currently used by the user.

[0052] 203. Extracting audio features of each music track from music data on multiple platforms; Preprocess the music tracks in the music data of multiple platforms to obtain target music information, and use audio analysis tools to process the target music information to extract audio features of each music track. The audio analysis tools include but are not limited to Librosa, Essentia or Melodia.

[0053] The steps of preprocessing the music tracks in the music data of multiple platforms to obtain the target music information are as follows: extract the music tracks in the music data of multiple platforms, divide each music track into multiple continuous audio clips, calculate the energy value or volume level of each audio clip, and use a preset threshold to determine whether each audio clip is a silent clip based on the energy value or volume level of each audio clip, remove the audio clips determined to be silent clips to obtain the target music information, or extract the audio clips determined to be non-silent clips to obtain the target music information. This method can effectively remove the silent parts in the music tracks, reduce the interference of audio feature extraction, and improve processing efficiency and accuracy.

[0054] The steps of using an audio analysis tool to process the target music information to extract the audio features of each music track are as follows: select an audio analysis tool, which has the ability to parse and convert the target music information, including but not limited to spectrum analysis, waveform analysis, pitch detection, rhythm analysis, etc.; input the target music information into the selected audio analysis tool, ensuring that the format of the input music information matches the tool requirements, which may involve format conversion or preprocessing steps; configure the setting parameters of the audio analysis tool to optimize the audio feature extraction process, which parameters include but are not limited to analysis accuracy, frequency range and time resolution; start the audio analysis tool, conduct an in-depth analysis of the target music information, and extract the audio features of each music track, which include but are not limited to pitch, timbre, rhythm, volume, sound intensity, and harmonic structure.

[0055] 204. Perform similarity matching in a preset database based on the audio features of each music track, and remove duplicate music tracks to obtain a plurality of first music tracks; In a preset music fingerprint database, a local sensitive hashing algorithm is used to perform similarity matching according to the audio features of each music track, and duplicate music tracks are removed to obtain multiple first music tracks.

[0056] A music fingerprint database containing a large number of music fingerprints is preset. Each music fingerprint is a unique identifier obtained by processing the original music data through a specific algorithm. It is used to quickly retrieve and match similar music content. The music fingerprint database should be updated regularly to include the latest music tracks.

[0057] The audio features of each music track are converted into music fingerprints, and the locality-sensitive hashing (LSH) algorithm is used to efficiently reduce the dimension and hash these music fingerprints. The LSH algorithm can map music fingerprints that are close to each other in the music feature space to the same or similar hash buckets with a high probability through a set of carefully designed hash functions, while maintaining the relative distance relationship between different music fingerprints, thereby achieving efficient similarity-preserving hashing. This process greatly reduces the amount of calculation and storage space required for subsequent similarity calculations. Within each hash bucket, similarity measurement methods such as Euclidean distance, cosine similarity or Manhattan distance are further used to calculate the similarity between music fingerprints. These similarity measurement methods can accurately reflect the similarities and differences between music fingerprints, thereby helping to identify repeated or highly similar music tracks. By setting a reasonable similarity threshold, potential repeated tracks can be automatically screened out. Once potential repeated tracks are identified, they are immediately marked, and further measures may be taken to verify the accuracy of these repeated identifications. This may include verifying the accuracy of recognition and eliminating false positives through manual review, audio comparison algorithms or more complex machine learning models. In addition, in order to improve the efficiency and accuracy of duplicate track recognition, the key elements such as the number of hash functions of the LSH algorithm, the hash bucket division strategy, the threshold of the similarity measurement, and the parameters of feature extraction can be carefully adjusted and optimized according to the needs of actual applications and effect feedback.

[0058] The music fingerprints are reduced in dimension and hashed by the local sensitive hashing algorithm, which significantly reduces the computational complexity and storage space of similarity calculation. At the same time, by using precise similarity measurement methods and reasonable threshold settings, it can automatically and accurately identify repeated or highly similar music tracks, and further improve the efficiency and accuracy of recognition through optimization and adjustment.

[0059] 205. Use the data source, text analysis technology, and the user's music listening history to complete the metadata of each first music track to obtain target music data; The first metadata of each first music track is obtained from multiple online data sources; the lyrics of each first music track are deeply analyzed by natural language processing technology to extract the second metadata of each first music track; based on each first music track, analysis is performed according to the music listening history of the user to obtain the third metadata of each first music track; metadata completion processing is performed on each first music track according to the first metadata, the second metadata and the third metadata to obtain target music data.

[0060] The detailed information of each first music track is captured from the music streaming platform to obtain the first metadata of each first music track, and the first metadata includes but is not limited to the album name, artist, release year and genre; the lyrics text of the first music track is obtained from the lyrics website, and each lyrics text is preprocessed, and the preprocessing includes word segmentation, removal of stop words and punctuation marks, etc., and a sentiment analysis algorithm is used to determine the sentiment tendency of each lyrics text, and the sentiment tendency includes positive, negative or neutral; the core theme category in each lyrics text is identified through a topic modeling algorithm, and the core theme categories include love, inspiration, friendship, etc.; the keyword extraction technology is used to screen out high-frequency keywords in each lyrics text, and these keywords can summarize the main content and characteristics of the lyrics, and the second metadata is constructed according to the extracted sentiment tendency, core theme category and keywords; the user's music listening history includes the user's comments on each first music track Purpose: playback behavior data, playback behavior data includes but is not limited to the number of plays, playback duration, playback time period, whether it is played repeatedly and whether it is skipped. Based on the playback behavior data of each first music track, data analysis technology is used to analyze the user's music preference to obtain third metadata, and the third metadata includes but is not limited to user preference, user interaction and user emotional resonance. Among them, user preference is calculated based on data such as the number of user plays and playback duration, reflecting the user's overall preference for a certain music track; user interaction is measured based on whether the user plays repeatedly, shares or comments on the track, etc., reflecting the user's active participation in the music track; user emotional resonance is evaluated by analyzing the user's emotional response when playing the track, such as whether it is played frequently under a specific emotional state. The degree of fit between the music track and the user's emotions.

[0061] 206. Construct cross-platform user portraits based on target music data; 207. Perform personalized recommendation algorithm processing based on the constructed cross-platform user portrait to obtain a target recommended music list.

[0062] In the embodiment of the present invention, a targeted extraction strategy is adopted to obtain multi-platform music data, and audio features are extracted for similarity matching to identify music tracks. The data source and text analysis technology are combined to complete the metadata, and finally a cross-platform user portrait is constructed and personalized recommendations are realized. This not only improves the efficiency and accuracy of music data collection, but also ensures the accuracy and data integrity of music identification through audio feature matching and metadata completion, thereby enabling a deeper understanding of user music preferences, achieving more accurate and personalized cross-platform music recommendations, and improving the consistency and satisfaction of user experience.

[0063] See also Figure 3 Another embodiment of the multi-platform music recommendation method in the embodiment of the present invention includes: 301. Acquire music data of multiple platforms, where the music data of multiple platforms includes music listening history of a user collected from at least two different music platforms; 302. Perform metadata completion processing on the music data of multiple platforms to obtain target music data; 303. Extracting music information and user behavior information from target music data to obtain target information; Music information and user behavior information are extracted from the target music data to obtain the target information; the music information covers the detailed attributes of the song, including but not limited to the title, artist, genre affiliation, release year and audio features. This information helps to gain a deeper understanding of the essential characteristics of music; while the user behavior information focuses on the specific behaviors of users interacting with music, including but not limited to playing, pausing, skipping, liking, sharing, etc., as well as the time, frequency and context of these behaviors. This information can reveal users' preferences and consumption habits for music.

[0064] 304. Analyze the corresponding behavior feature vector, preference model and situation perception model according to the target information; The target information is cleaned and formatted to remove duplicate, invalid or abnormal data and form a unified behavioral feature vector. The target information is deeply analyzed using a hybrid recommendation model based on collaborative filtering and content, as well as a topic model, to build a preference model that reflects the user's music preferences. Based on the target information, the conditional random field model is used to model the user's music needs in different situations, and a situational awareness model that can predict the user's situational music needs is obtained.

[0065] Data cleaning techniques, such as deduplication, missing value processing, outlier detection and processing, are used to purify target information to ensure data accuracy and consistency. Through data formatting, the cleaned data is converted into a unified format to facilitate subsequent analysis. On this basis, key features are extracted, such as user behavior patterns, song attributes, etc., to form a unified behavioral feature vector.

[0066] Collaborative filtering algorithms are used to mine potential associations in user historical behavior information to find other users or items with similar interests to the user. Combined with content-based recommendation models, the characteristics of music content are analyzed and matched with the user's known preferences. The topic model is used to extract topics from large amounts of text data to gain a deeper understanding of the user's music preferences. These methods are combined to conduct in-depth analysis of the target information, and ultimately a preference model that can accurately reflect the user's music preferences is constructed.

[0067] The conditional random field model is used to conduct in-depth analysis of these target information. By training the conditional random field model, it can learn and capture the potential rules and patterns of users' music selection in different situations, and then build a situational awareness model. This model can intelligently predict and recommend music content that meets the user's situational music needs based on the user's current specific situation, including time, location, activity type, etc., thereby providing more accurate and personalized music recommendation services.

[0068] 305. Build cross-platform user portraits based on behavioral feature vectors, preference models, and context-aware models; Preprocess the behavioral feature vectors, preference models, and situational awareness models to ensure that they match each other in format, scale, and meaning. Use feature fusion algorithms, such as splicing, weighted summation, or deep learning models, to fuse these data at the feature level to form a comprehensive feature vector that includes user behavior, preference, and situational information. Use machine learning or deep learning technology to train and optimize the comprehensive feature vector to build a user portrait that can comprehensively reflect the comprehensive information of multiple platforms.

[0069] 306. Perform personalized recommendation algorithm processing based on the constructed cross-platform user portrait to obtain a target recommended music list.

[0070] In the embodiment of the present invention, by acquiring music data from multiple platforms and completing metadata, extracting music information and user behavior information, and then analyzing the behavioral feature vector, preference model and situational perception model, a cross-platform user portrait is constructed that comprehensively reflects the user's behavioral characteristics, music preferences and situational music needs on multiple music platforms, which significantly improves the accuracy of personalized recommendations and can be more in line with the user's immediate needs and preference changes, thereby providing users with more intimate and situational music recommendation services, greatly enhancing the personalization and satisfaction of the user experience.

[0071] See also Figure 4 Another embodiment of the multi-platform music recommendation method in the embodiment of the present invention includes: 401. Acquire music data of multiple platforms, where the music data of multiple platforms includes music listening history of a user collected from at least two different music platforms; 402. Perform metadata completion processing on the music data of multiple platforms to obtain target music data; 403. Construct a cross-platform user portrait based on the target music data; 404. Extracting a latent vector reflecting the user's cross-platform music taste based on the cross-platform user portrait; Features reflecting the user's music taste are extracted from the cross-platform user portraits and quantified to obtain the target feature vector; the target feature vector is matrix decomposed and the user's latent vector is extracted from the result of the matrix decomposition.

[0072] From the cross-platform user portraits, features closely related to the user's music taste are screened out, including but not limited to playback frequency, number of likes, number of shares, and preferred genres, and these features are quantified by mathematical methods to form a target feature vector with a unified metric standard. A target matrix decomposition technology is selected, and the target matrix decomposition technology includes singular value decomposition (SVD) or non-negative matrix factorization (NMF). The target feature vector is organized into a matrix form, in which rows represent users or user groups, and columns represent different music taste features. The target matrix decomposition technology is applied to decompose the organized matrix to obtain two or more low-dimensional matrices, which contain the main information of the original matrix. The part directly related to the user's music taste is selected from the decomposed low-dimensional matrix to construct a low-dimensional latent vector that can efficiently represent the user's music taste.

[0073] 405. Constructing music feature vectors of each second music track according to the music content in the local database, where the second music track is the music track included in the local database; For each second music track in the local database, a corresponding music feature vector is constructed according to its music attribute. These feature vectors are mathematical representations that can fully reflect the music characteristics of each track.

[0074] 406. Calculate the similarity between each second music track and the user preference according to the latent vector and each music feature vector set to obtain a music candidate set; According to the latent vector and each music feature vector, the cosine similarity algorithm is used to calculate the similarity between each second music track and the user's preference; all second music tracks are sorted according to the calculated similarity, and a music candidate set is generated based on the sorting result.

[0075] The latent vector is compared with the music feature vector of each second music track one by one to calculate the corresponding similarity. When calculating the similarity, the cosine formula is used to calculate the dot product of the two vectors and divide it by the product of their respective moduli. The value obtained is between -1 and 1, where 1 indicates complete similarity, -1 indicates complete oppositeness, and 0 indicates no correlation. After completing the similarity calculation of all second music tracks and latent vectors, the second music tracks are sorted in descending order according to the similarity value, and a screening threshold is preset, which represents the length of the recommendation list. The first preset number of tracks are selected from the sorted list as the music candidate set. The first preset number is determined according to the length of the recommendation list. The music candidate set accurately captures the user's music preferences and provides rich and high-quality choices for generating a personalized final recommendation list that meets the user's taste.

[0076] 407. Generate a target recommended music list based on the music candidate set and in combination with at least one music prediction model.

[0077] When a music prediction model exists, each music track in the music candidate set is input into the music prediction model. The model uses deep learning, collaborative filtering or other algorithms to calculate the degree to which each music track meets user preferences, i.e., the prediction score, based on audio features of the track, user historical behavior and other data. The music candidate set is sorted in descending order according to the prediction score, and the first second preset number of music tracks after sorting are selected as the target recommended music list to ensure the accuracy of the recommendation and user satisfaction.

[0078] When there are two or more music prediction models, weights are assigned to at least one music prediction model; the music candidate set is processed according to at least one music prediction model to obtain prediction results of each music prediction model; and the weights and prediction results of each music prediction model are combined for analysis to obtain a target recommended music list.

[0079] Different weights are assigned to the music prediction models according to their performance evaluation results, and these weights reflect the relative importance of the models in the final recommendation. Each music track in the music candidate set is input into each model respectively to obtain the prediction score of each music track under each model. The weights and prediction scores of each model are combined to calculate the comprehensive prediction score of each music track by weighted summation or weighted average. The music candidate set is sorted in descending order according to the comprehensive prediction score, and the first second preset number of music tracks after sorting are selected as the target recommended music list, thereby ensuring the accuracy and diversity of the recommendation.

[0080] In an embodiment of the present invention, a cross-platform user portrait is constructed by acquiring multi-platform music data and performing metadata completion, and a latent vector reflecting the user's cross-platform music taste is extracted therefrom. By calculating the similarity between the music feature vectors of the music tracks in the local database and the user's preference latent vectors, a music candidate set is screened out, and a target recommended music list is generated in combination with a music prediction model. This not only improves the personalization of music recommendations, but also comprehensively captures the user's music taste based on his or her cross-platform behavior, ensuring the accuracy and personalization of the recommended music, thereby significantly improving the depth and satisfaction of the user experience.

[0081] The above describes the music recommendation method based on multiple platforms in the embodiment of the present invention. The following describes the music recommendation device based on multiple platforms in the embodiment of the present invention. Figure 5 In one embodiment of the present invention, a multi-platform music recommendation device includes: An acquisition module 501 is used to acquire music data from multiple platforms, where the music data from multiple platforms includes a user's music listening history, preference information, and interaction records collected from at least two different music platforms; The first processing module 502 is used to perform metadata completion processing on the music data of multiple platforms to obtain target music data; A construction module 503 is used to construct a cross-platform user portrait based on the target music data; The second processing module 504 is used to perform personalized recommendation algorithm processing according to the constructed cross-platform user portrait to obtain a target recommended music list.

[0082] In an embodiment of the present invention, by collecting and integrating the user's music data across platforms and after metadata completion processing, a detailed cross-platform user portrait is constructed, thereby generating a target recommended music list, which fully reflects the user's music preferences and immediate needs, and significantly improves the personalization and accuracy of music recommendations.

[0083] See also Figure 6 Another embodiment of the multi-platform music recommendation device in the embodiment of the present invention includes: An acquisition module 501 is used to acquire music data from multiple platforms, where the music data from multiple platforms includes a user's music listening history, preference information, and interaction records collected from at least two different music platforms; The first processing module 502 is used to perform metadata completion processing on the music data of multiple platforms to obtain target music data; A construction module 503 is used to construct a cross-platform user portrait based on the target music data; The second processing module 504 is used to perform personalized recommendation algorithm processing according to the constructed cross-platform user portrait to obtain a target recommended music list.

[0084] Optionally, the acquisition module 501 includes: A determination unit 5011, used to determine a target music platform for which music data needs to be collected; The first extraction unit 5012 is used to extract music data from each target music platform according to the type of each target music platform and the corresponding extraction strategy to obtain music data for multiple platforms.

[0085] Optionally, the first extraction unit 5012 may be specifically used for: When the type of the target music platform is the first music platform that supports API, an API request is sent to the first music platform, and the response data returned by the first music platform is parsed to obtain the music data of the first music platform; when the type of the target music platform is the second music platform that does not support API, the corresponding data is captured by a crawler program to obtain the music data of the second music platform; when the target music platform is the third music platform where the local database is located, the corresponding data is extracted from the local database to obtain the music data of the third music platform.

[0086] Optionally, the first processing module 502 includes: The second extraction unit 5021 is used to extract the audio features of each music track from the music data of multiple platforms; A matching unit 5022 is used to perform similarity matching in a preset database based on the audio features of each music track, and remove duplicate music tracks to obtain multiple first music tracks, where the first music tracks are music tracks included in the music data of multiple platforms; The completion unit 5023 is used to complete the metadata of each first music track by using the data source, text analysis technology and the user's music listening history to obtain the target music data.

[0087] Optionally, the matching unit 5022 may be specifically configured to: In a preset music fingerprint database, a local sensitive hashing algorithm is used to perform similarity matching according to the audio features of each music track, and duplicate music tracks are removed to obtain multiple first music tracks. Optionally, the completion unit 5023 may be specifically used for: The first metadata of each first music track is obtained from multiple online data sources; the lyrics of each first music track are deeply analyzed by natural language processing technology to extract the second metadata of each first music track; based on each first music track, analysis is performed according to the music listening history of the user to obtain the third metadata of each first music track; metadata completion processing is performed on each first music track according to the first metadata, the second metadata and the third metadata to obtain target music data.

[0088] Optionally, the building block 503 includes: The third extraction unit 5031 is used to extract music information and user behavior information from the target music data to obtain target information; An analysis unit 5032 is used to analyze the corresponding behavior feature vector, preference model and situation perception model according to the target information; The first construction unit 5033 is used to construct a cross-platform user portrait based on the behavioral feature vector, preference model and situational awareness model. The cross-platform user portrait comprehensively reflects the user's behavioral characteristics, music preferences and situational music needs on multiple music platforms.

[0089] Optionally, the analysis unit 5032 may be specifically configured to: The target information is cleaned and formatted to remove duplicate, invalid or abnormal data and form a unified behavioral feature vector. The target information is deeply analyzed using a hybrid recommendation model based on collaborative filtering and content, as well as a topic model, to build a preference model that reflects the user's music preferences. Based on the target information, the conditional random field model is used to model the user's music needs in different situations, and a situational awareness model that can predict the user's situational music needs is obtained.

[0090] Optionally, the second processing module 504 includes: A fourth extraction unit 5041 is used to extract a latent vector reflecting the user's cross-platform music taste according to the cross-platform user portrait; A second construction unit 5042 is used to construct a music feature vector of each second music track according to the music content in the local database, where the second music track is the music track included in the local database; A calculation unit 5043 is used to calculate the similarity between each second music track and the user preference according to the latent vector and each music feature vector set to obtain a music candidate set; The generating unit 5044 is used to generate a target recommended music list according to the music candidate set in combination with at least one music prediction model.

[0091] Optionally, the fourth extraction unit 5041 may be specifically used for: Features reflecting the user's music taste are extracted from the cross-platform user portraits and quantified to obtain the target feature vector; the target feature vector is matrix decomposed and the user's latent vector is extracted from the result of the matrix decomposition.

[0092] Optionally, the calculation unit 5043 may be specifically configured to: According to the latent vector and each music feature vector, the cosine similarity algorithm is used to calculate the similarity between each second music track and the user's preference; all second music tracks are sorted according to the calculated similarity, and a music candidate set is generated based on the sorting result.

[0093] Optionally, the generating unit 5044 may be specifically configured to: When there are two or more music prediction models, weights are assigned to at least one music prediction model; the music candidate set is processed according to at least one music prediction model to obtain prediction results of each music prediction model; and the weights and prediction results of each music prediction model are combined for analysis to obtain a target recommended music list.

[0094] In an embodiment of the present invention, by targeted extraction of multi-platform music data and metadata completion, combined with audio feature matching and text analysis technology, a comprehensive and accurate cross-platform user portrait can be constructed, thereby generating a target recommended music list, which not only significantly improves the accuracy of music recommendations, but also through the analysis of user behavior characteristics, preference models and situational awareness models, it is possible to more deeply understand the user's music preferences and immediate needs, and based on accurate user portraits and personalized recommendation algorithms, it is possible to provide users with more intimate, contextual and personalized music recommendation services, thereby improving the depth, consistency and satisfaction of the user experience.

[0095] above Figure 5 and Figure 6 The multi-platform music recommendation apparatus in the embodiment of the present invention is described in detail from the perspective of modular functional entities. The multi-platform music recommendation device in the embodiment of the present invention is described in detail from the perspective of hardware processing.

[0096] See also Figure 7 As shown, the multi-platform-based music recommendation device includes a processor 700 and a memory 701 , wherein the memory 701 stores machine executable instructions that can be executed by the processor 700 , and the processor 700 executes the machine executable instructions to implement the above-mentioned multi-platform-based music recommendation method.

[0097] Further, Figure 7 The multi-platform-based music recommendation device shown further includes a bus 702 and a communication interface 703 , and the processor 700 , the communication interface 703 and the memory 701 are connected via the bus 702 .

[0098] Among them, the memory 701 may include a high-speed random access memory (RAM), and may also include a non-volatile memory (non-volatile memory), for example, at least one disk storage. The communication connection between the system network element and at least one other network element is realized through at least one communication interface 703 (which can be wired or wireless), and the Internet, wide area network, local area network, metropolitan area network, etc. can be used. The bus 702 can be an ISA bus, a PCI bus or an EISA bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 7 Only one bidirectional arrow is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0099] The processor 700 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the hardware integrated logic circuit or software instructions in the processor 700. The above processor 700 may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The disclosed methods, steps and logic block diagrams in the embodiments of the present disclosure can be implemented or executed. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in conjunction with the embodiments of the present disclosure can be directly embodied as a hardware decoding processor to be executed, or a combination of hardware and software modules in the decoding processor can be executed. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory 701 , and the processor 700 reads the information in the memory 701 and completes the method steps of the above-mentioned embodiment in combination with its hardware.

[0100] The present invention also provides a multi-platform music recommendation device, wherein the computer device includes a memory and a processor, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor executes the steps of the multi-platform music recommendation method in the above-mentioned embodiments. The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions, and when the instructions are executed on a computer, the computer executes the steps of the multi-platform music recommendation method.

[0101] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0102] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the whole or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0103] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A music recommendation method based on multiple platforms, characterized in that: The method comprises: Acquire music data of multiple platforms, wherein the music data of multiple platforms includes music listening history of a user collected from at least two different music platforms; Performing metadata completion processing on the music data of the multiple platforms to obtain target music data; Constructing a cross-platform user portrait based on the target music data; A personalized recommendation algorithm is processed based on the constructed cross-platform user portrait to obtain a target recommended music list.

2. The multi-platform music recommendation method according to claim 1, characterized in that: The obtaining of music data from multiple platforms includes: Determine the target music platforms where music data needs to be collected; Music data of each target music platform is extracted according to the type of each target music platform and the corresponding extraction strategy to obtain music data of multiple platforms.

3. The multi-platform music recommendation method according to claim 2, characterized in that: The method of extracting music data from each target music platform according to the type of each target music platform and the corresponding extraction strategy to obtain music data for multiple platforms includes: When the type of the target music platform is a first music platform that supports the API, an API request is sent to the first music platform, and response data returned by the first music platform is parsed to obtain music data of the first music platform; When the type of the target music platform is a second music platform that does not support the API, the corresponding data is captured using a crawler program to obtain the music data of the second music platform; When the target music platform is a third music platform where the local database is located, corresponding data is extracted from the local database to obtain the music data of the third music platform.

4. The multi-platform music recommendation method according to claim 1, characterized in that: The performing metadata completion processing on the music data of the multiple platforms to obtain target music data includes: Extracting audio features of each music track from the music data of the multiple platforms; Based on the audio features of each music track, similarity matching is performed in a preset database, and duplicate music tracks are removed to obtain a plurality of first music tracks, where the first music tracks are music tracks included in the music data of multiple platforms; The metadata of each first music track is completed by using the data source, text analysis technology and the music listening history of the user to obtain the target music data.

5. The multi-platform music recommendation method according to claim 4, characterized in that: The similarity matching is performed in a preset database based on the audio features of each music track, and duplicate music tracks are removed to obtain a plurality of first music tracks, including: In a preset music fingerprint database, a local sensitive hashing algorithm is used to perform similarity matching according to the audio features of each music track, and duplicate music tracks are removed to obtain multiple first music tracks.

6. The multi-platform music recommendation method according to claim 4, characterized in that: The method of using the data source, text analysis technology and the user's music listening history to complete the metadata of each first music track to obtain target music data includes: Acquire first metadata for each first music track from a plurality of online data sources; Performing deep text analysis on the lyrics of each first music track by using natural language processing technology to extract second metadata of each first music track; Based on each first music track, analyzing is performed according to the music listening history of the user to obtain third metadata of each first music track; Metadata completion processing is performed on each first music track according to the first metadata, the second metadata and the third metadata to obtain target music data.

7. The multi-platform music recommendation method according to claim 1, characterized in that: The constructing a cross-platform user portrait according to the target music data includes: Extracting music information and user behavior information from the target music data to obtain target information; Analyze the corresponding behavior feature vector, preference model and situation perception model according to the target information; A cross-platform user portrait is constructed based on the behavioral feature vector, the preference model and the context-aware model, and the cross-platform user portrait comprehensively reflects the user's behavioral characteristics, music preferences and contextualized music needs on multiple music platforms.

8. The multi-platform music recommendation method according to claim 7, characterized in that: The step of analyzing the corresponding behavior feature vector, preference model and situation awareness model according to the target information includes: Performing data cleaning and formatting processing on the target information to remove duplicate, invalid or abnormal data and form a unified behavior feature vector; Using a hybrid recommendation model of collaborative filtering and content-based, and a topic model to deeply analyze the target information, and construct a preference model reflecting the user's music preference; Based on the target information, a conditional random field model is used to model the user's music needs in different situations, thereby obtaining a situational awareness model that can predict the user's situational music needs.

9. The multi-platform music recommendation method according to claim 1, characterized in that: The personalized recommendation algorithm is processed according to the constructed cross-platform user portrait to obtain a target recommended music list, including: Extracting a latent vector reflecting the user's cross-platform music taste according to the cross-platform user portrait; constructing a music feature vector of each second music track according to the music content in the local database, wherein the second music track is a music track contained in the local database; Calculating the similarity between each second music track and the user's preference based on the latent vector and each music feature vector set to obtain a music candidate set; A target recommended music list is generated based on the music candidate set and combined with at least one music prediction model.

10. The multi-platform music recommendation method according to claim 9, characterized in that: The step of extracting a latent vector reflecting the user's cross-platform music taste according to the cross-platform user portrait includes: Extracting features reflecting the user's music taste from the cross-platform user portrait, and performing quantization processing to obtain a target feature vector; The target feature vector is subjected to matrix decomposition, and a latent vector of the user is extracted from the result of the matrix decomposition.

11. The multi-platform music recommendation method according to claim 9, characterized in that: The similarity between each second music track and the user preference is calculated according to the latent vector and each music feature vector to obtain a music candidate set, including: The similarity between each two music tracks and the user's preference is calculated using a cosine similarity algorithm based on the latent vector and each music feature vector; All second music tracks are sorted according to the calculated similarities, and a music candidate set is generated based on the sorting results.

12. The multi-platform music recommendation method according to claim 9, characterized in that: When there are two or more music prediction models, generating a target recommended music list according to the music candidate set in combination with at least one music prediction model includes: assigning weights to at least one music prediction model; Processing the music candidate set according to the at least one music prediction model to obtain prediction results of each music prediction model; The weights and prediction results of each music prediction model are analyzed to obtain the target recommended music list.

13. A music recommendation device based on multiple platforms, characterized in that: The multi-platform-based music recommendation device comprises: An acquisition module, configured to acquire music data from multiple platforms, wherein the music data from multiple platforms includes a user's music listening history, preference information, and interaction records collected from at least two different music platforms; A first processing module, configured to perform metadata completion processing on the music data of the multi-platforms to obtain target music data; A construction module, used for constructing a cross-platform user portrait based on the target music data; The second processing module is used to perform personalized recommendation algorithm processing based on the constructed cross-platform user portrait to obtain a target recommended music list.

14. A music recommendation device based on multiple platforms, characterized in that: The multi-platform-based music recommendation device comprises: a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instructions in the memory so that the multi-platform-based music recommendation device executes the multi-platform-based music recommendation method as described in any one of claims 1-12.

15. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the multi-platform based music recommendation method as described in any one of claims 1-12 is implemented.

Images