A playlist generation method, electronic device, storage medium, program product

Abstract

The application provides a playlist generation method, an electronic device, a storage medium and a program product, and relates to the field of song recommendation.The method comprises the following steps: obtaining a listening behavior time sequence of a user; segmenting the listening behavior time sequence according to the change of listening behavior characteristics in the listening behavior time sequence, obtaining at least one behavior period, and determining a corresponding behavior mode according to the listening behavior characteristics corresponding to each behavior period; analyzing the music preference law of the user in each preset time range according to the listening behavior characteristics corresponding to different preset time ranges in the listening behavior time sequence, and obtaining interest clock information of the user; extracting a continuous listening sequence in the listening behavior time sequence, and determining listening intention information of the user according to the listening behavior characteristics corresponding to the continuous listening sequence; and generating a user playlist according to the behavior mode, the interest clock information and the listening intention information.

Description

Technical Field

[0001] This invention relates to the field of song recommendation, and in particular to a song list generation method, electronic device, storage medium, and program product. Background Technology

[0002] When generating playlists for users to recommend songs, the relevant technologies often recommend songs based on user preferences or simple scenario rules, resulting in a mismatch between the recommended results and the user's actual listening needs. Summary of the Invention

[0003] The purpose of this invention is to provide a playlist generation method, electronic device, storage medium, and program product that can determine the user's actual listening needs based on the user's listening behavior and generate a playlist that matches the user's current needs, thereby improving the personalization of the playlist and the user experience.

[0004] To solve the above technical problems, the present invention provides a playlist generation method, comprising: Obtain the time series of users' music listening behavior; Based on the changes in the listening behavior characteristics in the listening behavior time series, the listening behavior time series is segmented to obtain at least one behavior cycle, and the corresponding behavior pattern is determined based on the listening behavior characteristics corresponding to each behavior cycle. Based on the listening behavior characteristics corresponding to different preset time ranges in the listening behavior time series, the user's music preference patterns in each preset time range are analyzed to obtain the user's interest clock information. Extract continuous listening sequences from the listening behavior time series, and determine the user's listening intent information based on the listening behavior characteristics corresponding to the continuous listening sequences; A user playlist is generated based on the behavioral pattern, the interest clock information, and the listening intention information.

[0005] Optionally, based on the changes in the listening behavior characteristics in the listening behavior time series, the listening behavior time series is segmented to obtain at least one behavior cycle, including: The listening behavior time sequence is segmented according to a preset time window to obtain at least two sequence segments; Based on the listening behavior characteristics in each sequence segment, the feature data change value between adjacent sequence segments is determined, and it is determined whether the feature data change value is greater than a first preset threshold. If the change value of the feature data is greater than the first preset threshold, then the adjacent sequence segments are determined to be behavioral periodic boundaries; Based on the behavioral cycle boundary, the time sequence of the music listening behavior is segmented to obtain the behavioral cycle.

[0006] Optionally, based on the listening behavior characteristics corresponding to different preset time ranges in the listening behavior time series, the user's music preference patterns in each preset time range are analyzed to obtain the user's interest clock information, including: The listening behavior time sequence is segmented according to different preset time ranges to obtain listening behavior sub-sequences corresponding to each preset time range. Based on the song style information contained in the subsequence of listening behavior, the main song style information and style consistency information corresponding to each preset time range of the user are determined as the interest clock information.

[0007] Optionally, extracting the continuous listening sequence from the listening behavior time series includes: According to preset segmentation conditions, the boundaries of continuous listening sequences in the listening behavior time series are determined; wherein, the preset segmentation conditions include at least one of the following: the time interval between adjacent listening behaviors is greater than a second preset threshold, the style difference between adjacent played songs is greater than a third preset threshold, and the playback source of adjacent played songs is switched. The continuous listening sequence is extracted from the listening behavior time series based on the boundary of the continuous listening sequence.

[0008] Optionally, the listening behavior characteristics include at least one of the following: song playback source distribution, content repetition rate, search density, style jump, and proportion of new songs.

[0009] Optionally, it also includes: Based on the user's song collection information and playback history, the user's listening preference baseline information is set; Using the baseline information of listening preferences as reference information, the steps of determining the corresponding behavior pattern based on the listening behavior characteristics corresponding to each of the behavior cycles, analyzing the music preference patterns of the user in each of the preset time ranges to obtain the user's interest clock information, and determining the user's listening intention information based on the listening behavior characteristics corresponding to the continuous listening sequence are executed.

[0010] Optionally, the behavior pattern, the interest clock information, and the music listening intention information are determined using multiple agents, and the output of each agent is transmitted between the agents along the path defined by the directed acyclic graph.

[0011] Optionally, generating a user playlist based on the behavior pattern, the interest clock information, and the listening intent information includes: Based on the user's current behavior pattern, interest clock information within the current preset time range, and listening intent information within the current continuous listening sequence, candidate songs are determined from the candidate song set. The candidate songs are sorted according to the song parameter information of the candidate songs and a preset parameter change curve to obtain the user playlist.

[0012] Optionally, it also includes: Pre-generate user playlists and cache them in the database; Upon receiving a user request, the system selects a matching user playlist from the database and pushes it to the user based on the current time, the user's current behavior pattern during their current behavior cycle, their current interest clock information, and their listening intent information. If there is no matching user playlist in the database, then the step of generating a user playlist based on the behavior pattern, the interest clock information, and the listening intention information is executed.

[0013] The present invention also provides an electronic device, comprising: Memory, used to store computer programs; A processor is used to implement the playlist generation method described above when executing the computer program.

[0014] The present invention also provides a computer program product, including a computer program or instructions, which, when executed by a processor, implement the above-described playlist generation method.

[0015] The present invention also provides a computer-readable storage medium storing computer-executable instructions, which, when loaded and executed by a processor, implement the above-described playlist generation method.

[0016] This invention provides a playlist generation method, comprising: acquiring a user's listening behavior time series; segmenting the listening behavior time series according to the changes in listening behavior characteristics in the listening behavior time series to obtain at least one behavior cycle, and determining a corresponding behavior pattern according to the listening behavior characteristics corresponding to each behavior cycle; analyzing the user's music preference patterns in each preset time range according to the listening behavior characteristics corresponding to different preset time ranges in the listening behavior time series to obtain the user's interest clock information; extracting continuous listening sequences from the listening behavior time series, and determining the user's listening intent information according to the listening behavior characteristics corresponding to the continuous listening sequences; and generating a user playlist based on the behavior pattern, the interest clock information, and the listening intent information.

[0017] The beneficial effects of this invention are as follows: First, this invention can obtain a user's music listening behavior time series, which represents the temporal changes in the user's music listening behavior and can be used to analyze changes in the user's music listening intention. Then, based on the changes in the music listening behavior characteristics in the time series, this invention can segment the music listening behavior time series to obtain at least one behavior cycle, and determine the corresponding behavior pattern based on the music listening behavior characteristics corresponding to the behavior cycle, thereby determining the user's music listening behavior pattern over a longer period. Subsequently, based on the music listening behavior characteristics corresponding to different preset time ranges in the music listening behavior time series, the invention can analyze the user's music preference patterns within each preset time range to obtain the user's interest clock information, thereby determining the user's interest preferences within different preset time ranges. Finally, the invention can extract continuous music listening sequences from the music listening behavior time series, and based on the music listening behavior characteristics corresponding to the continuous music listening sequences, determine the user's music listening intention information, thereby determining the user's actual music listening intention in short-term continuous music listening. Finally, user playlists can be generated based on behavioral patterns, interest clock information, and listening intent information. This allows for the creation of playlists based on users' listening behavior patterns over a longer period, their interests within a preset time range, or their actual listening intent during short-term continuous listening, ensuring that song recommendations better match users' actual listening needs.

[0018] The present invention also provides an electronic device, a computer-readable storage medium, and a computer program product, which have the above-mentioned beneficial effects. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0020] Figure 1 A flowchart of a playlist generation method provided in an embodiment of the present invention; Figure 2 A flowchart of data preprocessing provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of an intelligent agent workflow provided in an embodiment of the present invention; Figure 4 A schematic diagram of a multi-agent semantic analysis process provided in an embodiment of the present invention; Figure 5 A flowchart for generating a user playlist is provided as an embodiment of the present invention; Figure 6 This is a schematic diagram of a playlist push process provided in an embodiment of the present invention; Figure 7 This is a structural block diagram of a playlist generation device provided in an embodiment of the present invention; Figure 8 This is a structural block diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] In audio content platforms such as music, podcasts, and audiobooks, playlists are generally generated in the following ways: 1. Professional editors manually select songs based on themes, styles, or scenarios. 2. Playlists are generated using collaborative filtering, which recommends similar songs based on a user's playback history and the preferences of similar users, automatically creating a playlist. 3. Playlists are based on content or scenario rules, such as matching preset scenario rules based on external information like time, weather, and location, and filtering songs from the song library that match the scenario to generate a playlist with a fixed style.

[0023] However, the above methods have the following problems: 1. Manually editing playlists is costly and doesn't match users' listening needs well. 2. Collaborative filtering relies solely on users' listening interactions, making it difficult to understand users' motivations and intentions for listening to certain types of songs in specific scenarios. 3. Users' listening behavior often changes over time, and related technologies struggle to identify these long-term behavioral patterns, ignoring the cyclical nature of user listening behavior. 4. Users often have specific intentions and emotions during continuous listening, and related technologies struggle to capture these short-term listening intentions. In short, the playlists generated by these technologies have a low match with users' current actual needs, impacting the user experience.

[0024] In view of this, the present invention provides a playlist generation method to address the technical problem of how to generate playlists that match the user's actual listening needs. This method can determine the user's actual listening needs based on the user's listening behavior and generate playlists that match the user's current needs, thereby improving the personalization of the playlists and the user experience.

[0025] For easier understanding, please refer to Figure 1 , Figure 1 A flowchart of a playlist generation method provided in an embodiment of the present invention, the method may include: S10. Obtain the time series of the user's music listening behavior.

[0026] In this step, various music listening behaviors generated by the user over a period of time can be obtained and arranged in chronological order to obtain a music listening behavior time series that reflects the user's music listening process during that period. It is understandable that the music listening behavior time series can relatively completely record the changes in the user's music listening behavior during that period, thus providing a data foundation for subsequent analysis of the user's music listening needs and intentions from multiple perspectives.

[0027] In one scenario, this embodiment can retrieve user playback behavior (e.g., playback, search, favorites, song skipping, etc.) within a preset time window from the user behavior log system of the audio content platform, and arrange them by timestamp. The specific data collected in this embodiment may include: user playback behavior (song ID, playback duration, playback time), search behavior (search keywords, time), favorites behavior (favorited song ID, time), song skipping behavior (skip time, skip position), etc.

[0028] Furthermore, this embodiment can preprocess the acquired listening behavior information, such as associating the played song with its metadata to enrich the semantic information of the behavior. The metadata can include the artist, genre, release year, tags, etc. Additionally, the playback source for each playback can be labeled to help determine the user's exploration intent for the song. Playback sources can include user-initiated searches, playback from favorites, system recommendations, playlist playback, etc. Furthermore, the complete playback rate for each playback can be calculated to determine the user's level of preference for each song. For easier understanding, please refer to... Figure 2 , Figure 2 This is a flowchart of a data preprocessing method provided in an embodiment of the present invention.

[0029] It should be noted that this embodiment does not limit the duration covered by the music listening behavior time series. It can be set according to actual application needs, such as 1 to 3 months, in order to identify long-term user behavior patterns.

[0030] Furthermore, in addition to obtaining the user's listening behavior time series, this invention can also set the user's listening preference baseline information based on the user's song collection information and historical playback information. Then, in subsequent steps, the listening preference baseline information can be used as reference information to perform the following steps: determining the corresponding behavior pattern based on the listening behavior characteristics corresponding to each behavior cycle; analyzing the user's music preference patterns within each preset time range to obtain the user's interest clock information; and determining the user's listening intent information based on the listening behavior characteristics corresponding to a continuous listening sequence. Specific steps can be found in the descriptions in subsequent embodiments.

[0031] To establish baseline information for listening preferences, one could statistically analyze at least one of the following: the distribution of song styles, eras, and artists in a user's favorite songs. This data could serve as a baseline for long-term user preferences. Conversely, one could analyze at least one of the following historical playback records: frequently played songs, frequently played artists, and frequently played styles. This data could then be used as a reference for stable user preferences. Thus, song collection information and historical playback information can be used as baseline information for listening preferences. In subsequent analyses of user listening behavior, such as calculating the percentage of new songs, style jumps, or determining primary song style information, this baseline information can be used as a benchmark, thereby improving the accuracy of the analysis.

[0032] S20. Based on the changes in the listening behavior characteristics in the listening behavior time series, the listening behavior time series is segmented to obtain at least one behavior cycle, and the corresponding behavior pattern is determined based on the listening behavior characteristics corresponding to each behavior cycle.

[0033] A behavioral cycle refers to a time period in which a user exhibits a stable pattern of music listening behavior. For example, if a user intensively searches for and explores new music during a certain period, this time period can be called a deep-exploration period; if a user mainly listens to and collects songs repeatedly during a certain period, this time period can be called an asset consumption period; if a user's music listening behavior is erratic and disorganized during a certain period, this time period can be called a chaotic transition period. Current methods cannot identify such long-term patterns, leading to a mismatch between song recommendations and user needs. For example, recommending too many new songs during the asset consumption period may not interest the user; recommending older songs during the deep-exploration period may miss out on the user's exploration needs.

[0034] Therefore, in this step, the changes in listening behavior characteristics over time can be examined based on the listening behavior time series. The boundaries of behavioral cycles are determined where these characteristics change, thus dividing the listening behavior time series into one or more behavioral cycles. Then, based on the listening tendencies reflected by the listening behavior characteristics within each cycle, the corresponding behavioral pattern is determined, and playlists are generated for the user based on this pattern. Since users' listening behavior often exhibits phased changes over time, dividing the listening behavior time series into behavioral cycles and determining their behavioral patterns allows us to identify the patterns of users' listening behavior over a longer time span.

[0035] Among them, music listening behavior characteristics refer to the feature quantities used to determine a user's music listening behavior, obtained statistically from a music listening behavior time series or a portion thereof. This embodiment can determine whether a user has a stable music listening behavior pattern within a certain time period based on music listening behavior characteristics, thereby dividing the music listening behavior time series into behavior periods.

[0036] It should be noted that this embodiment does not limit specific listening behavior characteristics and can be set according to actual application needs. For example, in one case, listening behavior characteristics may include at least one of the following: song playback source distribution, content repetition rate, search density, style jump, and new song ratio. Here, song playback source distribution refers to the proportion of playback times from each playback source (e.g., favorites, search, system recommendations, playlists, etc.) within a corresponding time period. Content repetition rate refers to the proportion of repeatedly played songs within a corresponding time period to the total number of plays; a higher content repetition rate indicates that users are more inclined to repeatedly listen to existing content rather than explore new content. Search density refers to the ratio of the number of search actions within a corresponding time period to the duration of that time period; a higher search density indicates that users are more inclined to explore new content. Style jump refers to the statistical value of the style difference between adjacent played songs within a corresponding time period; a higher style jump indicates that the user's listening behavior is more disordered. New song ratio refers to the proportion of songs that the user has not previously played within a corresponding time period; a higher new song ratio indicates that the user is more inclined to explore new content.

[0037] The following describes one method for segmenting behavioral cycles. In one implementation, the listening behavior time series is segmented based on changes in listening behavior characteristics to obtain at least one behavioral cycle, which may include: S21. Divide the time sequence of listening behavior into segments according to the preset time window to obtain at least two sequence segments.

[0038] In this step, the listening behavior time series can first be divided into multiple sequence segments according to a preset time window (e.g., 3 to 7 days) to statistically analyze the listening behavior characteristics in each sequence segment. There can be overlap between the sequence segments to improve the accuracy of determining the boundaries of the behavior cycle.

[0039] S22. Based on the listening behavior characteristics in each sequence segment, determine the feature data change value between adjacent sequence segments, and determine whether the feature data change value is greater than the first preset threshold.

[0040] In this step, for two adjacent sequence segments, the feature data change value between them is determined based on the listening behavior characteristics in each sequence segment. The feature data change value can be, for example, the absolute value or weighted sum of the differences between the corresponding listening behavior characteristics; and the feature data change value is compared with a first preset threshold.

[0041] It should be noted that different first preset thresholds can be set for different listening behavior characteristics to effectively adapt to actual situations.

[0042] S23. If the change value of the feature data is greater than the first preset threshold, then the boundary between adjacent sequence segments is determined to be the behavioral periodic boundary.

[0043] In this step, if the change value of the feature data is greater than the first preset threshold, it indicates that the listening behavior characteristics between adjacent sequence segments have changed significantly. For example, the search density may suddenly increase from low to high, or the proportion of new songs may suddenly increase from low to high. This indicates that the user's behavior pattern has changed significantly. Therefore, the adjacent sequence segments can be identified as the boundary of the behavior cycle.

[0044] Of course, if the change values ​​of the feature data between the above sequence segments are not greater than the first preset threshold, and it is impossible to determine the boundary of the behavior cycle between adjacent sequence segments, then the entire listening behavior time series can also be regarded as a behavior cycle.

[0045] S24. Based on the behavioral cycle boundary, the time series of listening behavior is segmented to obtain the behavioral cycle.

[0046] In this step, the listening behavior time series can be segmented into behavioral periods based on the determined boundaries of each behavioral period, so as to determine the behavioral pattern based on the listening behavior characteristics within each behavioral period.

[0047] For example: when the content repetition rate is greater than the corresponding threshold, it is determined to be an asset consumption mode mainly based on repeatedly playing the saved content; when the search density is greater than the corresponding threshold, it is determined to be a deep mining mode mainly based on intensive search to explore new content; when the proportion of new songs is greater than the corresponding threshold and involves a large number of styles, it is determined to be a breadth exploration mode; when the style jump is greater than the corresponding threshold and there is no obvious dominant style, it is determined to be a chaotic transition mode.

[0048] As can be seen, this embodiment can automatically determine the boundary of the behavior cycle based on the changes in listening behavior characteristics over time, thereby dividing the user's listening behavior time series into multiple behavior cycles and determining the behavior pattern of each behavior cycle. This allows for a more accurate determination of the user's listening behavior patterns over a longer time span, providing a basis for generating user playlists that match the user's current behavior stage.

[0049] S30. Based on the listening behavior characteristics corresponding to different preset time ranges in the listening behavior time series, analyze the user's music preference patterns in each preset time range to obtain the user's interest clock information.

[0050] Interest clock information represents the patterns of music preferences exhibited by users within different preset time ranges, and is used to determine what kind of content users tend to listen to at what times.

[0051] In this step, listening behaviors belonging to the same preset time range within a time series are grouped together. The music preferences reflected in these listening behaviors within that preset time range are then statistically analyzed to determine the user's interest clock information within that range. The preset time range can be set according to a day / night cycle, such as dividing a day into morning and afternoon, or further dividing a day into more time periods. The preset time range can also be set according to a weekday cycle, such as grouping by day of the week (e.g., Monday to Friday, or Saturday and Sunday) to statistically analyze differences in playback behavior between weekdays and weekends.

[0052] It is understandable that users often have different listening preferences at different times. By determining the user's interest clock information within each preset time range, it is possible to determine the pattern of how the user's preferences change over time.

[0053] The following describes a method for determining interest clock information. In one implementation, based on the listening behavior characteristics corresponding to different preset time ranges in the listening behavior time series, the user's music preference patterns within each preset time range are analyzed to obtain the user's interest clock information, which may include: S31. Divide the time sequence of listening behavior into segments according to different preset time ranges to obtain the subsequence of listening behavior corresponding to each preset time range.

[0054] In this step, the preset time range can be segmented according to different time periods within a day, or according to weekdays and rest days within a week, or according to other methods. For example, when segmenting according to different time periods within a day, it can be divided into several time periods such as 0:00 to 6:00, 6:00 to 9:00, 9:00 to 12:00, 12:00 to 14:00, 14:00 to 18:00, 18:00 to 22:00, and 22:00 to 24:00.

[0055] Subsequently, the listening behavior time sequence can be segmented according to different preset time ranges to obtain listening behavior sub-sequences corresponding to each preset time range, such as listening behavior sub-sequences corresponding to 0:00 to 6:00, listening behavior sub-sequences corresponding to 6:00 to 9:00, etc.

[0056] S32. Based on the song style information contained in the song listening behavior subsequence, determine the main song style information and style consistency information corresponding to each preset time range for the user, and use them as interest clock information.

[0057] In this step, the proportion of each style in the subsequence of listening behavior is counted, and the style with the highest proportion is determined as the main song style information corresponding to the preset time range. The style consistency information corresponding to the preset time range is determined according to the concentration of style distribution. The higher the style consistency, the more stable the user's preferences are within that time range.

[0058] As can be seen, this embodiment can determine the music preference patterns of users in different time dimensions, thereby providing a basis for pushing user playlists that conform to their preference patterns within a specific time range.

[0059] S40. Extract the continuous listening sequence from the listening behavior time series, and determine the user's listening intent information based on the listening behavior characteristics corresponding to the continuous listening sequence.

[0060] A continuous listening sequence refers to a segment of listening behavior that is coherent in terms of listening intent and emotion within a time series of listening actions; its time span is usually relatively short. Listening intent information refers to the user's purpose or intention in listening within the corresponding continuous listening sequence.

[0061] In this step, locations where listening intent changes can be identified within the time series of listening behavior, thereby extracting one or more consecutive listening sequences. Subsequently, based on the characteristics reflected in the listening behavior within each consecutive listening sequence, the listening intent information corresponding to that sequence is determined. It is understandable that continuous listening within a short period often carries specific intentions and emotions; by extracting consecutive listening sequences and determining their listening intent information, the user's listening intent over a short time span can be identified.

[0062] Furthermore, when determining listening intent information, listening behavior characteristics can be extracted for each continuous listening sequence. These characteristics include the duration of the sequence, the average complete playback rate of the songs, the stylistic consistency of the songs within the sequence, and the average energy level calculated based on the rhythm and energy of the songs. Subsequently, the listening intent information corresponding to the continuous listening sequence can be determined based on these listening behavior characteristics. For example, an average energy level greater than a certain threshold indicates an intent to release emotions; a duration greater than a certain threshold and high stylistic consistency indicate an intent to provide background music for extended periods; high stylistic consistency revolving around a specific style or artist indicates an intent to focus and explore; an average complete playback rate less than a certain threshold indicates an intent to quickly preview and filter content; and playing songs already saved by the user indicates an intent to revisit familiar material and express nostalgia. Corresponding emotional tags can also be added. Of course, other methods can also be used to determine the listening intent of a continuous listening sequence, and these can be set according to actual application needs.

[0063] It should be noted that this embodiment does not limit the method of dividing the continuous listening sequence. For example, it can be divided according to the interval between adjacent listening behaviors, the style change between adjacent listening behaviors, the source change between adjacent listening behaviors, etc. The following describes the method of dividing the continuous listening sequence. In one embodiment, extracting the continuous listening sequence from the listening behavior time series may include: S41. Determine the boundary of the continuous listening sequence in the listening behavior time series according to the preset segmentation conditions; wherein, the preset segmentation conditions include at least one of the following: the time interval between adjacent listening behaviors is greater than a second preset threshold, the style difference between adjacent played songs is greater than a third preset threshold, and the playback source of adjacent played songs is switched.

[0064] In this step, the preset segmentation conditions include at least one of the following: the time interval between adjacent listening behaviors is greater than a second preset threshold (e.g., the time interval between two adjacent playbacks exceeds 30 minutes); the style difference between adjacent played songs is greater than a third preset threshold (e.g., switching from instrumental to rock); and the playback source of adjacent played songs changes (e.g., switching from playing from favorites to playing from search). When adjacent listening behaviors meet at least one of the above preset segmentation conditions, the interval is determined as the boundary of a continuous listening sequence.

[0065] S42. Extract the continuous listening sequence from the listening behavior time series based on the boundary of the continuous listening sequence.

[0066] In this step, one or more consecutive listening sequences can be extracted from the listening behavior time series based on the determined boundaries of the consecutive listening sequences.

[0067] Furthermore, it should be noted that this embodiment does not limit the execution order of the above steps S20, S30 and S40. All three are based on the same listening behavior time sequence and do not have any dependency relationship with each other. They can be executed sequentially or in parallel.

[0068] Furthermore, when determining behavioral patterns, interest clock information, and listening intent information, this embodiment can further integrate more contextual signals, such as the user's geographical location information, social relationship data (e.g., the content that the user's friends are listening to), and at least one of the physiological signals such as heart rate and cadence collected by the wearable device, in order to more accurately determine the user's current state and intent.

[0069] In another implementation, cross-domain behavioral data of users on other content platforms can also be integrated, such as behavioral data generated by users on platforms such as film and television, books, and games, to build a more complete user interest profile, thereby improving the matching degree between the generated user playlist and the user's actual needs.

[0070] S50 generates user playlists based on behavioral patterns, interest clock information, and listening intent information.

[0071] In this step, the user's current behavioral pattern, interest clock information corresponding to the user's current preset time range, and listening intent information corresponding to the user's current continuous listening sequence can be combined to determine matching songs and organize them in a certain order, thereby generating a user playlist that matches the user's current needs.

[0072] This embodiment does not limit the specific method of generating user playlists; please refer to the embodiments described later.

[0073] Based on the above embodiments, the present invention first obtains a user's music listening behavior time series, which characterizes the temporal changes in the user's music listening behavior and can be used to analyze changes in the user's music listening intention. Subsequently, the present invention can segment the music listening behavior time series according to the changes in the music listening behavior characteristics in the time series, obtaining at least one behavior period, and determine the corresponding behavior pattern based on the music listening behavior characteristics corresponding to the behavior period, thereby determining the user's music listening behavior pattern over a longer period. Then, based on the music listening behavior characteristics corresponding to different preset time ranges in the music listening behavior time series, the present invention can analyze the user's music preference patterns in each preset time range, obtaining the user's interest clock information, thereby determining the user's interest preferences in different preset time ranges. Finally, continuous music listening sequences can be extracted from the music listening behavior time series, and based on the music listening behavior characteristics corresponding to the continuous music listening sequences, the user's music listening intention information can be determined, thereby determining the user's actual music listening intention in short-term continuous music listening. Finally, user playlists can be generated based on behavioral patterns, interest clock information, and listening intent information. This allows for the creation of playlists based on users' listening behavior patterns over a longer period, their interests within a preset time range, or their actual listening intent during short-term continuous listening, ensuring that song recommendations better match users' actual listening needs.

[0074] Based on the above embodiments, to achieve automatic analysis of behavior patterns, interest clock information, and listening intent information, in this embodiment, behavior patterns, interest clock information, and listening intent information can be determined using multiple agents. The output of each agent is transmitted between agents along the path defined by the directed acyclic graph. For ease of understanding, please refer to... Figure 3 , Figure 3 This is a schematic diagram of an intelligent agent workflow provided in an embodiment of the present invention.

[0075] In this context, an agent refers to a processing unit capable of performing specific analytical tasks; a directed acyclic graph (DAG) is a graph structure consisting of nodes and directed edges that does not contain cycles, where nodes represent agents or processing steps, and directed edges define the direction of data flow.

[0076] For example, a first agent can determine behavioral patterns, a second agent can determine interest clock information, and a third agent can determine listening intent information. The entire analysis process can be modeled as a directed acyclic graph (DAG), where each node corresponds to an agent or a processing step, and the directed edges between nodes define data flow and control flow. In this way, agent workflows can be orchestrated to achieve automatic analysis of behavioral patterns, interest clock information, and listening intent information.

[0077] Furthermore, in one feasible implementation, when each agent determines the behavior pattern, interest clock information, or listening intention information, it can input the corresponding listening behavior features, along with the corresponding segments in the listening behavior time series and optional listening preference baseline information, into a pre-trained language processing model. The pre-trained language processing model then infers and outputs the corresponding behavior pattern, interest clock information, or listening intention information. Here, the pre-trained language processing model refers to a language processing model that has been pre-trained and possesses semantic understanding and reasoning capabilities. This invention does not limit its specific type.

[0078] For easier understanding, please refer to Figure 4 , Figure 4 This is a schematic diagram of a multi-agent semantic analysis process provided in an embodiment of the present invention. The processing flow of each agent is described in detail below: 1. Behavioral Cycle Segmentation Agent (First Intelligent Agent): Used to divide a user's long-term music listening trajectory into multiple behavioral cycles and identify the behavioral patterns in each cycle.

[0079] The inputs to the agent include: the user's complete music listening timeline (usually spanning 1-3 months), the user's bookmarks, and knowledge base.

[0080] The agent's output includes: a list of behavior cycles, each cycle containing: cycle ID, start and end time; cycle type (asset consumption period, deep mining period, breadth exploration period, chaotic transition period); cycle summary (natural language description within 100 words); and confidence level (High / Medium / Low).

[0081] The core processing flow of this intelligent agent includes: 1) Feature extraction: Playback source distribution: Statistics on the percentage of various playback sources (favorites, search, recommendations, playlists, etc.) within this time period.

[0082] Content duplication rate: Calculates the percentage of songs played repeatedly. A high duplication rate indicates asset consumption behavior.

[0083] Search density: The frequency of search behavior within a given time period. High search density indicates in-depth data mining behavior.

[0084] Style jump: Calculates the style difference between adjacent songs played. A high jump indicates chaotic transition behavior.

[0085] New song percentage: This counts the percentage of new songs that users have never heard before among the songs played. A high percentage of new songs indicates exploratory behavior.

[0086] 2) LLM (Pre-trained Language Processing Model) Inference: The extracted features and timeline segments are input into the LLM (via the Prompt project) to allow the LLM to infer behavioral patterns for that time period.

[0087] 3) Periodic boundary identification: Behavioral features are calculated using a sliding window (window size 3-7 days). When a feature changes significantly (e.g., search density suddenly increases from low to high, or the proportion of new songs suddenly increases from low to high), it is determined to be a periodic boundary.

[0088] LLM is used to perform secondary confirmation of the boundary to avoid misjudgment.

[0089] 2. Interest Clock Analysis Agent (Secondary Intelligent Agent): Used to discover the patterns of users' music interest preferences across different time dimensions (circadian rhythm, weekly rhythm).

[0090] The inputs to the agent include: the user's complete music listening timeline and the user's knowledge base.

[0091] The agent's output includes: an interest clock JSON structure containing: user profile summary (dual profile, behavioral features), circadian rhythm pattern (preference features at different times), and weekly rhythm pattern (preference differences between weekdays and weekends).

[0092] The core processing flow of this intelligent agent includes: 1) Time-based bucket aggregation: Day and night dimension: Divide 24 hours into multiple time periods (such as 0-6, 6-9, 9-12, 12-14, 14-18, 18-22, 22-24) and count the playback behavior in each time period.

[0093] Weekly dimension: Grouped by day of the week (Monday to Friday, Saturday and Sunday), and the differences in playback behavior between weekdays and weekends are statistically analyzed.

[0094] 2) Style Preference Analysis: The songs played in each time period are clustered by style to identify the dominant style (e.g., preference for instrumental music and piano pieces in the afternoon on weekdays; preference for sad pop and rock in the late night).

[0095] Calculate the style consistency score for each time period. High consistency indicates that the preferences are stable during that time period.

[0096] 3) LLM Summary: By inputting the aggregated features of the time dimension into the LLM, the LLM generates a natural language summary of interest preferences.

[0097] 3. Semantic Session Segmentation Agent (Third Agent): Identify short-term listening intent sessions (i.e., continuous listening sequences) from the user's listening behavior time series, and infer the emotion and motivation of each session.

[0098] The inputs to the agent include: the user's complete listening behavior time series or recent segments, the user's knowledge base, and interest clock.

[0099] The output of the agent includes: a list of semantic sessions, each session containing: session ID, start and end time, session mode (venting, background noise, studying, shopping, nostalgia, etc.), intent description (natural language description of why the user listens to these songs), key evidence (behavioral features that support the judgment), and emotion tags (excitement / release, focus, curiosity / exploration, trying something new, calm / nostalgia, etc.).

[0100] The core processing flow of this intelligent agent includes: 1) Session boundary recognition: Time interval threshold: If the time interval between two consecutive playbacks exceeds 30 minutes, it is considered a session boundary.

[0101] Style abrupt change detection: When the style difference between adjacent songs exceeds a threshold (such as suddenly switching from instrumental to rock), it is determined to be a session boundary.

[0102] Playback source switching: When the playback source changes (such as switching from favorites to search), it is determined to be a session boundary.

[0103] 2) Session feature extraction: Playback duration: Calculates the total duration of the session. Long sessions (>1 hour) are usually in background sound mode.

[0104] Song completeness rate: Calculates the average complete playback rate of songs within a session. A low completeness rate indicates the purchase mode (quick listening and filtering).

[0105] Style consistency: Calculate the style consistency of songs within a session. High consistency indicates a research mode (focused on exploring a particular style).

[0106] Energy Level: The average energy level of a session is calculated based on the audio characteristics of the song (rhythm, volume). High energy indicates a cathartic mode.

[0107] 3) LLM Intent Reasoning: Input the session's behavioral characteristics and song list into the LLM, allowing the LLM to infer the user's listening intentions and emotional state.

[0108] In this way, each independent agent is responsible for a type of analysis task and collaborates based on a directed acyclic graph. This makes the processing logic of each type of analysis task relatively independent, easy to expand and maintain, thereby improving the flexibility of the processing flow while ensuring the accuracy of the analysis results.

[0109] In addition to using directed acyclic graphs for orchestration, the aforementioned multiple intelligent agents can also be implemented using other orchestration methods such as event-driven architecture and custom state machines. The determination of the aforementioned behavioral patterns, interest clock information, and music listening intention information can be achieved not only by using pre-trained language processing models, but also by combining clustering, time series analysis, and other methods. This embodiment of the invention does not limit this.

[0110] Based on the above embodiments, the specific method for generating user playlists is described below. For ease of understanding, please refer to... Figure 5 , Figure 5 This invention provides a flowchart for generating a user playlist. In one embodiment, generating a user playlist based on behavioral patterns, interest clock information, and listening intent information may include: S51. Based on the user's current behavior pattern, interest clock information within the current preset time range, and listening intent information within the current continuous listening sequence, determine candidate songs from the candidate song set.

[0111] In this step, the candidate song set may include, for example, the user's favorites library and the platform's recommendation library. It is understood that the song selection strategies corresponding to behavioral patterns, interest clock information, and listening intent information are not the same. For example, when the current behavioral pattern is asset consumption, frequently collected songs from the user's favorites library are selected as candidate songs; when the current listening intent is catharsis, high-energy songs are selected as candidate songs.

[0112] Here are some possible song selection strategies: Strategy 1: Periodic Perception Playlist 1. Asset Consumption Period: Generate a collection and review playlist, select the user's frequently played collection songs, and sort them according to the emotional curve (from soothing to climax to calm).

[0113] Playlist title example: Your Nostalgia Tribute: Classic Hits Retrospective.

[0114] Playlist Description: During this time, you seem to prefer returning to familiar melodies. These classic old songs by these artists will accompany you through every moment when your emotions need comfort.

[0115] 2. In-depth exploration phase: Generate playlists for exploring new worlds, recommending new songs and artists of the same type based on the user's recent search keywords and playback style.

[0116] Example playlist title: AI Cover Song Exploration: A Singer's Musical Experiment.

[0117] Playlist Description: You've recently developed a strong interest in AI cover singers. This playlist features a selection of more AI cover songs, along with comparisons to the original versions, taking you deeper into this cutting-edge music field.

[0118] 3. Chaotic Transition Period: Generates a random roaming playlist, recommending songs of diverse styles and a wide range to satisfy users' desire for exploration.

[0119] Example playlist title: Musical Journey: A Leap from Nostalgia to Exotic Lands.

[0120] Playlist Description: Your recent listening preferences have been unpredictable, ranging from nostalgia to Thai rock and K-Pop hits. This playlist will take you on another musical adventure, with each song a surprise.

[0121] Strategy 2: Time-Aware Playlist Based on the user's interest clock, playlists that match their rhythm are pushed at specific times: 1. 8-10 AM: Morning wake-up playlist, recommending hit songs, nostalgic oldies and other music suitable for commuting.

[0122] 2. Weekday afternoons from 4 PM to 6 PM: Focus on instrumental music playlists, recommending instrumental music suitable for background music during work.

[0123] 3. Late at night 10 PM - 1 AM: Playlist for releasing emotions late at night, recommending sad pop, rock and other music suitable for emotional catharsis.

[0124] Strategy 3: Intent-Aware Playlists Generate a matching playlist based on the user's recent session intent: 1. Catharsis mode: We recommend high-energy rock, electronic music, metal and other music suitable for releasing stress.

[0125] 2. Background sound mode: We recommend long-duration instrumental music, movie soundtracks, ambient sounds, and other music suitable for work and study.

[0126] 3. In-depth mode: Recommends all works of artists that users are currently exploring, as well as in-depth content such as similar recommendations.

[0127] 4. Procurement Model: Recommends the latest released songs, popular charts, and other content suitable for quick listening.

[0128] S52. Sort the selected candidate songs according to the preset parameter change curve based on the song parameter information of the candidate songs to obtain the user playlist.

[0129] In this step, the song parameter information may include at least one of the song's rhythm and energy. The preset parameter change curve may specify a change in energy that first remains flat, then gradually increases to a climax, and finally falls back to calm. This allows the song parameters in the generated user playlist to present the preset parameter change curve as the playback order changes after the candidate songs are sorted, so that the user playlist presents a coherent change during listening, thus closely resembling the effect of a manually generated playlist.

[0130] Furthermore, based on at least one of behavioral patterns, interest clock information, and listening intent information, corresponding titles and descriptions can be generated for the generated user playlists.

[0131] As can be seen, this embodiment can sort candidate songs according to a preset parameter change curve, so that the generated user playlist presents a coherent change during the listening process, thereby improving the coherence of the user playlist and the user's listening experience.

[0132] In addition, for new users with limited behavioral data, their basic behavioral patterns can be identified through short-term behavioral observation (e.g., 3 to 7 days), and an initial playlist can be generated for them accordingly. As the user's behavioral data is gradually accumulated, the aforementioned behavioral patterns, interest clock information, and listening intent information can be further refined to improve the cold start effect for new users.

[0133] Furthermore, to improve the efficiency of pushing playlists to users, this embodiment can also pre-generate playlists for users and push the pre-generated playlists to users when they request them. The process of pre-generating playlists is described below. For ease of understanding, please refer to... Figure 6 , Figure 6 This is a schematic diagram of a playlist push process provided by an embodiment of the present invention. In one embodiment, the method may further include: S71. Pre-generate user playlists and cache them in the database.

[0134] In this embodiment, one or more user playlists can be pre-generated for the user during their idle time and cached in the database. The database can also cache the user's behavior patterns, interest clock information, and listening intention information. In a feasible implementation, the database can be an in-memory database (such as Redis) to reduce read latency.

[0135] In addition, this embodiment can also pre-acquire and store the user's behavioral semantic profile, such as the behavioral pattern of the current behavioral cycle, the interest clock information in each preset time range, and the listening intention in the current continuous listening session.

[0136] S72. Upon receiving a user request, based on the current time, the user's current behavior pattern within their current behavior cycle, current interest clock information, and listening intent information, select a matching user playlist from the database and push it to the user.

[0137] For example, when a user starts listening to music, the system can obtain the current time and the user's recent playback behavior, match them with the user's behavioral semantic profile, and then select a suitable user playlist from the database based on the matching results.

[0138] S73. If there is no matching user playlist in the database, then execute the step of generating a user playlist based on behavior patterns, interest clock information, and listening intent information.

[0139] If a matching user playlist exists in the database, it is directly selected and pushed. If no matching playlist exists, the playlist generation process described above is executed in real time to generate a new user playlist. By pre-generating and caching user playlists, it is possible to quickly select and push a playlist that matches the user's current state when a user request is received, thereby improving real-time response speed while maintaining personalization.

[0140] Furthermore, it can also obtain user feedback on the pushed user playlists (such as skipping, liking, and saving), and dynamically adjust the sorting and content of the user playlists based on the feedback information. The feedback information is used as new listening behavior data to continuously optimize the subsequently determined behavior patterns, interest clock information, and listening intent information.

[0141] The following describes the playlist generation device, electronic device, computer-readable storage medium, and computer program product provided in the embodiments of the present invention. The playlist generation device, electronic device, computer-readable storage medium, and computer program product described below can be referred to in correspondence with the playlist generation method described above.

[0142] Please refer to Figure 7 , Figure 7 This is a structural block diagram of a playlist generation device provided in an embodiment of the present invention. The device may include: Data acquisition module 701 is used to acquire the time series of users' music listening behavior; The behavior cycle analysis module 702 is used to segment the listening behavior time series according to the changes in the listening behavior characteristics in the listening behavior time series to obtain at least one behavior cycle, and to determine the corresponding behavior pattern according to the listening behavior characteristics corresponding to each behavior cycle. The preference analysis module 703 is used to analyze the user's music preference pattern in each preset time range based on the music listening behavior characteristics corresponding to different preset time ranges in the music listening behavior time series, and obtain the user's interest clock information. The intent analysis module 704 is used to extract the continuous listening sequence in the listening behavior time series, and determine the user's listening intent information based on the listening behavior characteristics corresponding to the continuous listening sequence; The playlist generation module 705 is used to generate a user playlist based on the behavior pattern, the interest clock information, and the listening intention information.

[0143] Optionally, the behavior cycle analysis module 702 includes: The segmentation submodule is used to segment the time sequence of the listening behavior according to a preset time window to obtain at least two sequence segments; The judgment submodule is used to determine the feature data change value between adjacent sequence segments based on the listening behavior characteristics in each sequence segment, and to determine whether the feature data change value is greater than a first preset threshold. The behavior period boundary determination submodule is used to determine the behavior period boundary between adjacent sequence segments if the change value of the feature data is greater than a first preset threshold. The behavior cycle segmentation submodule is used to segment the time sequence of the music listening behavior according to the behavior cycle boundary to obtain the behavior cycle.

[0144] Optionally, the preference analysis module 703 includes: The subsequence division submodule is used to segment the music listening behavior time sequence according to different preset time ranges to obtain music listening behavior subsequences corresponding to each preset time range. The style preference analysis submodule is used to determine the main song style information and style consistency information corresponding to each preset time range of the user based on the song style information contained in the song listening behavior subsequence, as interest clock information.

[0145] Optionally, the intent analysis module 704 includes: The continuous listening sequence boundary determination submodule is used to determine the boundary of the continuous listening sequence in the listening behavior time series according to preset segmentation conditions; wherein, the preset segmentation conditions include at least one of the following: the time interval between adjacent listening behaviors is greater than a second preset threshold, the style difference between adjacent played songs is greater than a third preset threshold, and the playback source of adjacent played songs is switched. The continuous listening sequence segmentation submodule is used to extract continuous listening sequences from the listening behavior time series based on the boundaries of the continuous listening sequence.

[0146] Optionally, listening behavior characteristics include at least one of the following: distribution of song playback sources, content repetition rate, search density, style jump, and proportion of new songs.

[0147] Optionally, the device may further include: The preference baseline information setting module is used to set the user's listening preference baseline information based on the user's song collection information and historical playback information; The reference module is used to use the music preference baseline information as reference information to perform the following steps: determining the corresponding behavior pattern based on the music behavior characteristics corresponding to each of the aforementioned behavior cycles; analyzing the music preference patterns of the user in each of the aforementioned preset time ranges to obtain the user's interest clock information; and determining the user's music listening intention information based on the music behavior characteristics corresponding to the continuous music listening sequence.

[0148] Optionally, behavioral patterns, interest clock information, and music listening intention information are determined using multiple agents, and the outputs of each agent are transmitted between agents along the path defined by the directed acyclic graph.

[0149] Optionally, the playlist generation module 705 includes: The candidate song determination submodule is used to determine candidate songs from the candidate song set based on the user's current behavior pattern, interest clock information within the current preset time range, and listening intent information of the current continuous listening sequence. The candidate song sorting submodule is used to sort the selected candidate songs according to the song parameter information and a preset parameter change curve to obtain the user's playlist.

[0150] Optionally, the device may further include: The playlist pre-generation module is used to pre-generate user playlists and cache them in the database; The playlist matching module is used to select a matching user playlist from the database and push it to the user when a user request is received, based on the current time, the user's current behavior pattern during the current behavior cycle, current interest clock information, and listening intent information. The playlist generation module 705 is also used to generate a user playlist based on behavioral patterns, interest clock information, and listening intent information if there is no matching user playlist in the database.

[0151] Please refer to Figure 8 , Figure 8 This is a structural block diagram of an electronic device provided in an embodiment of the present invention. The electronic device 40 provided in this embodiment includes a processor 41 and a memory 42; wherein, the memory 42 is used to store a computer program; and the processor 41 is used to execute the playlist generation method provided in the foregoing embodiment when executing the computer program.

[0152] For details on the specific process of the above playlist generation method, please refer to the corresponding content provided in the foregoing embodiments, which will not be repeated here.

[0153] Furthermore, the memory 42, as a carrier for resource storage, can be a read-only memory, random access memory, disk, or optical disk, and the storage method can be temporary storage or permanent storage.

[0154] In addition, the electronic device 40 also includes a power supply 43, a communication interface 44, an input / output interface 45, and a communication bus 46; wherein, the power supply 43 is used to provide operating voltage for the various hardware devices on the electronic device 40; the communication interface 44 can create a data transmission channel between the electronic device 40 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this invention, and is not specifically limited here; the input / output interface 45 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.

[0155] This invention also provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the playlist generation method described in the above embodiments.

[0156] Since the embodiments of the computer program product part correspond to the embodiments of the playlist generation method part, please refer to the description of the embodiments of the playlist generation method part for the embodiments of the computer program product part, and will not be repeated here.

[0157] This invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the playlist generation method described in the above embodiments.

[0158] Since the embodiments of the computer-readable storage medium portion correspond to the embodiments of the playlist generation method portion, the embodiments of the storage medium portion are described in the description of the embodiments of the playlist generation method portion, and will not be repeated here.

[0159] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.

[0160] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0161] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0162] The above provides a detailed description of the playlist generation method, electronic device, storage medium, and program product provided by this invention. Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and core ideas of this invention. It should be noted that those skilled in the art can make various improvements and modifications to this invention without departing from its principles, and these improvements and modifications also fall within the protection scope of this invention.

Claims

1. A method for generating playlists, characterized in that, include: Obtain the time series of users' music listening behavior; Based on the changes in the listening behavior characteristics in the listening behavior time series, the listening behavior time series is segmented to obtain at least one behavior cycle, and the corresponding behavior pattern is determined based on the listening behavior characteristics corresponding to each behavior cycle. Based on the listening behavior characteristics corresponding to different preset time ranges in the listening behavior time series, the user's music preference patterns in each preset time range are analyzed to obtain the user's interest clock information. Extract continuous listening sequences from the listening behavior time series, and determine the user's listening intent information based on the listening behavior characteristics corresponding to the continuous listening sequences; A user playlist is generated based on the behavioral pattern, the interest clock information, and the listening intention information.

2. The playlist generation method according to claim 1, characterized in that, Based on the changes in the listening behavior characteristics in the listening behavior time series, the listening behavior time series is segmented to obtain at least one behavior cycle, including: The listening behavior time sequence is segmented according to a preset time window to obtain at least two sequence segments; Based on the listening behavior characteristics in each sequence segment, the feature data change value between adjacent sequence segments is determined, and it is determined whether the feature data change value is greater than a first preset threshold. If the change value of the feature data is greater than the first preset threshold, then the adjacent sequence segments are determined to be behavioral periodic boundaries; Based on the behavioral cycle boundary, the time sequence of the music listening behavior is segmented to obtain the behavioral cycle.

3. The playlist generation method according to claim 1, characterized in that, Based on the listening behavior characteristics corresponding to different preset time ranges in the listening behavior time series, the user's music preference patterns in each preset time range are analyzed to obtain the user's interest clock information, including: The listening behavior time sequence is segmented according to different preset time ranges to obtain listening behavior sub-sequences corresponding to each preset time range. Based on the song style information contained in the subsequence of listening behavior, the main song style information and style consistency information corresponding to each preset time range of the user are determined as the interest clock information.

4. The playlist generation method according to claim 1, characterized in that, Extracting the continuous listening sequence from the time series of the listening behavior includes: According to preset segmentation conditions, the boundaries of continuous listening sequences in the listening behavior time series are determined; wherein, the preset segmentation conditions include at least one of the following: the time interval between adjacent listening behaviors is greater than a second preset threshold, the style difference between adjacent played songs is greater than a third preset threshold, and the playback source of adjacent played songs is switched. The continuous listening sequence is extracted from the listening behavior time series based on the boundary of the continuous listening sequence.

5. The playlist generation method according to claim 1, characterized in that, The listening behavior characteristics include at least one of the following: song playback source distribution, content repetition rate, search density, style jump, and proportion of new songs.

6. The playlist generation method according to claim 1, characterized in that, Also includes: Based on the user's song collection information and playback history, the user's listening preference baseline information is set; Using the baseline information of listening preferences as reference information, the steps of determining the corresponding behavior pattern based on the listening behavior characteristics corresponding to each of the behavior cycles, analyzing the music preference patterns of the user in each of the preset time ranges to obtain the user's interest clock information, and determining the user's listening intention information based on the listening behavior characteristics corresponding to the continuous listening sequence are executed.

7. The playlist generation method according to claim 1, characterized in that, The behavioral pattern, the interest clock information, and the music listening intention information are determined by multiple agents, and the output of each agent is transmitted between the agents along the path defined by the directed acyclic graph.

8. The playlist generation method according to claim 1, characterized in that, The step of generating a user playlist based on the behavior pattern, the interest clock information, and the listening intent information includes: Based on the user's current behavior pattern, interest clock information within the current preset time range, and listening intent information within the current continuous listening sequence, candidate songs are determined from the candidate song set. The candidate songs are sorted according to the song parameter information of the candidate songs and a preset parameter change curve to obtain the user playlist.

9. The playlist generation method according to claim 1, characterized in that, Also includes: Pre-generate user playlists and cache them in the database; Upon receiving a user request, the system selects a matching user playlist from the database and pushes it to the user based on the current time, the user's current behavior pattern during the current behavior cycle, the current interest clock information, and the user's listening intent information. If there is no matching user playlist in the database, then the step of generating a user playlist based on the behavior pattern, the interest clock information, and the listening intention information is executed.

10. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the playlist generation method as described in any one of claims 1 to 9.

11. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or instructions are executed by the processor, they implement the playlist generation method as described in any one of claims 1 to 9.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when loaded and executed by a processor, implement the playlist generation method as described in any one of claims 1 to 9.

Images