Method, device and storage medium for recommending audio-video content

Abstract

The application discloses a recommendation method and device of audio and video content and a storage medium, and relates to the technical field of computers and the Internet. The method comprises the following steps: in the playing process of the audio and video content, acquiring a real-time voice instruction input by a user, the real-time voice instruction being used for indicating the preference of the user for the audio and video content; based on the real-time voice instruction, adjusting a user portrait of the user to obtain an adjusted user portrait, the user portrait being used for indicating the explicit preference and the implicit preference of the user for the audio and video content; based on the adjusted user portrait, determining a content recommendation result, the content recommendation result comprising at least one recommended audio and video content provided to the user; and based on the content recommendation result, playing the recommended audio and video content. The method realizes dynamic adjustment of the user portrait, so that the recommended audio and video content can timely respond to the preference change of the user for the audio and video content, thereby improving the accuracy of the audio and video content recommendation.

Description

Technical Field

[0001] This application relates to the fields of computer and internet technology, and in particular to a method, device, and storage medium for recommending audio and video content. Background Technology

[0002] An audio and video playback platform is a web platform or application that provides streaming music or video services, allowing users to play audio and video content on the internet. The platform can recommend and push suitable audio and video content to users.

[0003] In related technologies, audio and video playback platforms select recommended audio and video content for users based on their user profiles. These user profiles are built upon historical behavioral data, which refers to the collection of behavioral data recorded during the operation of the audio and video playback platform, documenting user interactions with audio and video content (such as completing playback, skipping, and clicking "like"). For example, historical behavioral data can be used as training data to train a recommendation model, resulting in the user's profile. Subsequent use of historical behavioral data from mobile phone users can then be used to construct new training data, continuously updating the user profile.

[0004] However, before updating user profiles, it takes time to collect users' historical behavior data, which delays the response of audio and video playback platforms to user preference updates, resulting in low accuracy of the audio and video content recommended by the platforms. Summary of the Invention

[0005] This application provides a method, device, and storage medium for recommending audio and video content. The technical solution provided by this application is as follows: According to one aspect of the embodiments of this application, a method for recommending audio and video content is provided, the method comprising: During the playback of audio and video content, real-time voice commands input by the user are obtained, and the real-time voice commands are used to indicate the user's preferences for audio and video content; Based on the real-time voice command, the user profile of the user is adjusted to obtain the adjusted user profile, which is used to indicate the user's explicit and implicit preferences for audio and video content. Based on the adjusted user profile, a content recommendation result is determined, which includes at least one recommended audio or video content to be provided to the user. Based on the content recommendation results, the recommended audio and video content is played.

[0006] According to one aspect of the embodiments of this application, an audio and video content recommendation device is provided, the device comprising: The acquisition module is used to acquire real-time voice commands input by the user during the playback of audio and video content. The real-time voice commands are used to indicate the user's preferences for audio and video content. An adjustment module is used to adjust the user profile of the user based on the real-time voice command to obtain an adjusted user profile, wherein the user profile is used to indicate the user's explicit and implicit preferences for audio and video content. The determination module is used to determine the content recommendation result based on the adjusted user profile, wherein the content recommendation result includes at least one recommended audio or video content to be provided to the user; The playback module is used to play the recommended audio and video content based on the content recommendation results.

[0007] According to one aspect of the embodiments of this application, a computer device is provided, the computer device including a processor and a memory, the memory storing a computer program, the computer program being loaded and executed by the processor to implement the above-described method for recommending audio and video content.

[0008] According to one aspect of the embodiments of this application, a computer-readable storage medium is provided, wherein a computer program is stored in the storage medium, the computer program being loaded and executed by a processor to implement the above-described recommended method for audio and video content.

[0009] According to one aspect of the embodiments of this application, a computer program product is provided, the computer program product including a computer program, the computer program being loaded and executed by a processor to implement the above-described method for recommending audio and video content.

[0010] The technical solutions provided in this application have at least the following beneficial effects: By acquiring users' real-time voice commands during audio and video content playback and dynamically adjusting user profiles based on these commands, the adjusted user profiles can reflect users' real-time explicit preferences. Then, based on the adjusted user profiles, audio and video content is determined to be recommended to users. This enables the recommended audio and video content to respond promptly to changes in users' preferences for audio and video content, thereby improving the accuracy of audio and video content recommendations. Attached Figure Description

[0011] Figure 1 This is a schematic diagram of a playback system provided in one embodiment of this application; Figure 2 This is a flowchart of an embodiment of the method for recommending audio and video content provided in this application; Figure 3 This is a timing diagram of temporary preference settings provided in one embodiment of this application; Figure 4 This is an architecture diagram of dynamically adjusting user profiles provided in one embodiment of this application; Figure 5 This is a block diagram of an audio / video content recommendation device provided in one embodiment of this application; Figure 6 This is a structural block diagram of a computer device provided in one embodiment of this application. Detailed Implementation

[0012] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0013] Please refer to Figure 1 The diagram illustrates a playback system provided in one embodiment of this application. The implementation environment of this solution may include: a terminal device 10 and a server 20.

[0014] Terminal device 10 is an electronic device with data computing, processing, and storage functions. In some embodiments, terminal device 10 is used as a client to run a target application that requires the recommendation and playback of audio and video content. Optionally, the target application may be an application that needs to be downloaded and installed, or it may be in the form of a webpage or an app; this application does not limit this. Optionally, the target application is an application that requires the recommendation and playback of audio and video content.

[0015] In some embodiments, the terminal device 10 may include, but is not limited to, at least one of the following: mobile phone, tablet computer, personal computer, vehicle terminal, smart wearable device, smart TV, smart voice interaction device, multimedia playback device, etc., and may also include other electronic devices, which are not limited in this application embodiment.

[0016] In this application embodiment, the target application may include, but is not limited to, at least one of the following: music playback application, live streaming application, video playback application, remote conferencing application, game application, educational application, instant messaging application, social application, shopping application, radio playback application, audio reading application, screen reader application, etc., and may also be other applications that have the need to play media content. This application embodiment does not limit this.

[0017] Server 20 is used to provide background services for clients of the target application running on terminal device 10. In some embodiments, the background service may include a recommendation service and a broadcast service for determining and pushing audio and video content recommended to the user corresponding to terminal device 10.

[0018] In some embodiments, server 20 may include, but is not limited to, at least one of the following: physical server, cloud server, edge server, server cluster, etc., and may also include other types of servers, which are not limited in this application embodiment.

[0019] Terminal device 10 and server 20 communicate with each other via a network. This network can be a wired network or a wireless network.

[0020] Please refer to Figure 2 The diagram illustrates a flowchart of a method for recommending audio and video content according to an embodiment of this application. The execution entity for each step of this method can be a computer device; for example, the computer device could be... Figure 1 The terminal device 10 in the playback system shown can also be Figure 1 Server 20 in the playback system shown. The method may include at least one of the following steps (210-240): Step 210: During the playback of audio and video content, obtain real-time voice commands input by the user. The real-time voice commands are used to indicate the user's preferences for audio and video content.

[0021] Audio and video content refers to media content in audio or video format. In this application embodiment, audio and video content is used as an object recommended to the user by the playback system. In some embodiments, audio and video content may include media content in audio format and / or media content in video format. That is, audio and video content may include the following three cases: audio and video content only includes media content in audio format; audio and video content only includes media content in video format; audio and video content includes both media content in audio format and media content in video format.

[0022] Media content in audio format refers to media content stored in the form of audio signals. Optionally, media content in audio format may include, but is not limited to, at least one of the following: music, podcasts, audiobooks, radio programs, etc., and may also include other audio media content, which are not limited in this application embodiment.

[0023] Media content in video format refers to media content stored in the form of video signals. Optionally, media content in video format may include, but is not limited to, at least one of the following: movies and TV dramas, variety shows, short videos, live recordings, instructional videos, animations, cartoons, game videos, user-generated content, news, etc. It may also include other media content in video format, which is not limited in this application embodiment.

[0024] Optionally, the audio-visual content that includes both audio and video media content may include, but is not limited to, at least one of the following: music video (MV), animated video, dubbed video, etc., and may also include other audio content. This application embodiment does not limit this.

[0025] In some embodiments, the currently playing audio / video content is historically recommended audio / video content. During the playback of audio / video content, multiple audio / video content items may have already been recommended to the user. That is, the system-recommended audio / video content is currently being played in a specified order. The user refers to the user corresponding to the current terminal device. For example, the currently playing audio / video music is recommended by the recommendation module in the playback system. Here, the recommendation module refers to a functional unit in the playback system, used to filter, sort, or match candidate audio / video content according to a preset recommendation strategy, and output recommended audio / video content to the playback system.

[0026] Real-time voice commands refer to voice information issued by a user during the playback of audio and video content, expressing their current preferences for the content. Users adjust their preferences for the audio and video content recommended by the playback system by issuing real-time voice commands. In other words, the voice content of the real-time voice command is related to the user's preferences for the audio and video content. In some embodiments, the terminal device captures the real-time voice commands input by the user through a microphone. In some embodiments, the microphone can be built into the terminal device or a microphone connected to the terminal device via a wired or wireless means.

[0027] In some embodiments, the voice content of real-time voice commands may include, but is not limited to, at least one of the following: positive preference information, negative preference information, constraint information, scene setting information, control information, etc., and may also include other voice content, which is not limited in this application embodiment.

[0028] Positive preference information is used to indicate a user's specific preferences for audio and video content. For example, real-time voice commands may include positive preference information such as "Recommend more relaxing songs," "I like this style of audio and video content," or "Give me some faster-paced audio and video content."

[0029] Negative preference information is used to indicate to users a clear negative tendency towards audio and video content. For example, real-time voice commands may include negative preference information such as "Don't recommend this kind of noisy electronic music to me anymore," "Don't push this artist's music to me anymore," or "I don't like this style of music," etc.

[0030] Constraint information is used to limit the conditions for recommending audio and video content. For example, real-time voice commands may include constraints such as "Find me a playlist of rhythmic English songs suitable for listening to while cleaning on the weekend", "Find upbeat rock music to listen to while driving at night", or "Recommend videos of no more than ten minutes next".

[0031] Scene setting information is used to instruct users to specify recommended audio and video content for a given scene. For example, real-time voice commands may include scene setting information such as "Recommend similar playlists when I'm driving in the future," "Play instrumental music on rainy days like now," or "Recommend slow-paced videos to me in the early morning."

[0032] Control information is used to control the playback process or recommendation flow of audio and video content. For example, real-time voice commands may include control information such as "Recommend some similar ones", "Switch to the next song", or "Replay this audio and video content".

[0033] In some embodiments, the triggering methods for obtaining real-time voice commands may include active wake-up and passive listening.

[0034] Active wake-up refers to a method where the user actively triggers the input of real-time voice commands. Optionally, active wake-up methods may include at least one of the following: wake-up via a wake word, wake-up via physical operation commands, etc., and may also include other active wake-up methods, which are not limited in this embodiment. The wake word is the voice signal that triggers voice acquisition. It should be noted that different playback systems may use different wake words. Physical operation commands refer to input signals triggered by physical means, used to control the terminal device to execute commands for acquiring real-time voice commands. Optionally, physical operation commands can be input through at least one of the following: control operation, button operation, gesture operation, etc., and may also be input through other means, which are not limited in this embodiment.

[0035] Passive monitoring refers to the method by which the playback system continuously or periodically monitors and collects real-time voice commands issued by the user.

[0036] In other embodiments, other forms of preference information indicating a user's preferences for audio and video content may also be obtained. That is, real-time voice commands are preference information in the form of voice. In some embodiments, the form of preference information may include, but is not limited to, at least one of the following: text form, gesture form (via smart glasses or in-vehicle central control system), combination of shortcut options (such as multi-select labels for reasons for "dislike"), etc., and may also include other forms, which are not limited in this application embodiment.

[0037] Step 220: Based on real-time voice commands, adjust the user profile to obtain the adjusted user profile. The user profile is used to indicate the user's explicit and implicit preferences for audio and video content.

[0038] User profiles are multi-dimensional models constructed based on user behavior towards audio and video content, used to represent user preferences for such content. Explicit preferences for audio and video content refer to preferences directly expressed by the user. Implicit preferences for audio and video content refer to preferences inferred from the user's behavior or context during the playback or recommendation process. These behaviors or contexts may include, but are not limited to, at least one of the following: clicking, playing, pausing, fast forwarding, rewinding, stopping, switching, favorites, liking, commenting, sharing, playback duration, playback completion rate, number of repeated playbacks, skip rate, interruption position, current time, date, time period, holidays, user activity level, usage frequency, current operation stage, etc., and may also include other content, which is not limited in this embodiment.

[0039] In some embodiments, the user profile includes tag preference information for at least two content tags. This tag preference information indicates the user's degree of preference for audio and video content belonging to that content tag. Content tags refer to feature identification information used to classify or attribute audio and video content. Optionally, content tags include at least one dimension: theme, type, style, mood, scene, etc., and may also include other dimensions, which are not limited in this embodiment. Optionally, the tag preference information for content tags includes the tag weight of that content tag. The tag weight is a numerical parameter used to quantify the degree of user preference for audio and video content matching that content tag. The tag weight can reflect the user's interest intensity or inclination towards the audio and video content corresponding to that content tag, and the magnitude of the tag weight is positively or negatively correlated with the user's degree of preference.

[0040] Adjusting user profiles based on real-time voice commands refers to mapping explicit preferences for audio and video content expressed in real-time voice commands to user profiles, thereby updating the preferences represented by the user profiles. By performing semantic analysis on real-time voice commands, the explicit preferences for audio and video content directly expressed by the user are obtained. Based on these explicit preferences, the user profile is adjusted to better reflect the user's current preferences for audio and video content.

[0041] In some embodiments, speech recognition is performed on real-time voice commands to obtain the command text corresponding to the real-time voice commands; intent understanding is performed on the command text to obtain intent understanding results, which are used to indicate the user's explicit preferences for audio and video content; based on the intent understanding results, the user profile is adjusted to obtain the adjusted user profile.

[0042] Speech recognition of real-time voice commands refers to extracting the text content from the real-time voice commands to obtain the corresponding command text. In other words, the command text of a real-time voice command refers to the text content contained within the real-time voice command.

[0043] In some embodiments, an Automatic Speech Recognition (ASR) model is used to recognize real-time voice commands, obtaining the corresponding command text. The ASR model has the ability to convert speech signals into text information. In some embodiments, the ASR model can be designed and developed by relevant technical personnel, or it can be open-source.

[0044] Intent understanding of command text refers to semantic analysis of the command text to extract the user's intent and preferences directly expressed through real-time voice commands. It is understood that the process of intent understanding of command text is also a form of Natural Language Understanding. In some embodiments, intent understanding may include, but is not limited to, at least one of the following methods: intent recognition, entity extraction, semantic parsing, etc., and may also include other methods, which are not limited in this application embodiment. Intent recognition refers to analyzing and determining the user's core purpose from the command text. Entity extraction refers to extracting key objects from the command text. Semantic parsing refers to converting the command text into a semantic result that can fully express the user's intent.

[0045] The intent understanding result refers to the information obtained by understanding the intent of the instruction text, which is used to represent the information related to the user's intent. Optionally, the intent understanding result may include, but is not limited to, at least one of the following: intent recognition result, entity extraction result, semantic parsing result, etc., and may also include other content, which is not limited in this embodiment.

[0046] The intent recognition result refers to the result obtained after performing intent recognition on the instruction text. Optionally, the intent recognition result may include, but is not limited to, at least one of the following: preference modification, scene setting, temporary preference setting, playback control, search request, etc., and may also include other content, which is not limited in this embodiment. Preference modification refers to the user's intent to modify or correct existing preferences for audio and video content. Scene setting refers to the user's intent to set corresponding preferences for audio and video content for a specified scene. Temporary preference setting refers to the user's intent to express short-term preferences for audio and video content within the current session. Playback control refers to the user's intent to control the playback process of audio and video content. Search request refers to the user's intent to search for audio and video content.

[0047] Entity extraction result refers to the result obtained after extracting entities from the instruction text. Optionally, the entity extraction result may include, but is not limited to, at least one of the following: content type, style attribute, language information, subject matter information, character information, emotional information, etc., which are not limited in this embodiment. Content type is used to characterize the basic category of audio and video content. Optionally, content type may include, but is not limited to, at least one of the following: movies, TV series, variety shows, animation, music, short videos, etc., and may also include other content types, which are not limited in this embodiment. Style attribute is used to describe the performance style or formal characteristics of audio and video content. Optionally, style attribute may include, but is not limited to, at least one of the following: relaxed, funny, tense, exciting, artistic, realistic, fast-paced, slow-paced, etc., and may also include other style attributes, which are not limited in this embodiment. Language information is used to indicate the language attribute of audio and video content. Optionally, language information may include, but is not limited to, at least one of the following: Chinese, English, Japanese, Korean, French, German, etc., and may also include other languages, which are not limited in this embodiment. Subject matter information is used to describe the theme of audio and video content. Optionally, the subject matter information may include, but is not limited to, at least one of the following: documentary, science fiction, suspense, romance, action, horror, etc., and may also include other subjects; this application embodiment does not limit this. Character information is used to indicate characters related to the audio-visual content. Optionally, character information may include, but is not limited to, at least one of the following: singer, actor, director, broadcaster, producer, composer, lyricist, etc., and may also include other characters; this application embodiment does not limit this. Emotional information is used to indicate the emotional characteristics expressed by the audio-visual content. Optionally, emotional information may include, but is not limited to, at least one of the following: healing, sadness, relaxation, excitement, celebration, happiness, etc., and may also include other emotions; this application embodiment does not limit this.

[0048] The semantic parsing result refers to the result obtained after semantic parsing the instruction text. Optionally, the semantic parsing result may include, but is not limited to, at least one of the following: the combined structure of intent recognition result and entity extraction result, constraints, preference intensity information, semantic relationships between various pieces of information, contextual association information, etc., and may also include other semantic information, which is not limited in this embodiment.

[0049] In some embodiments, the intent understanding result is implemented as structured data. Structured data refers to data that stores the intent understanding structure in a structured format, having a fixed format, clearly defined fields, and an organizational structure. Optionally, the structured format may include, but is not limited to, at least one of the following: JSON (JavaScript Object Notation), MsgPack (MessagePack), XML (Extensible Markup Language), YAML (YAML Ain't Markup Language), etc., and may also include other structured formats, which are not limited in this embodiment.

[0050] The above method, by recognizing and understanding the intent of real-time voice commands and dynamically adjusting the user profile based on the intent understanding results, achieves accurate acquisition and real-time reflection of users' explicit preferences, thereby significantly improving the accuracy of audio and video content recommendation and enabling real-time response to changes in user preferences.

[0051] In some embodiments, the instruction text is processed by a natural language understanding engine to understand the intent, resulting in an intent understanding outcome. The natural language understanding engine is a functional module within the playback system used for understanding the intent of instruction text.

[0052] In some embodiments, the natural language understanding engine includes an intent understanding model; the intent understanding model performs intent understanding on the instruction text to obtain the intent understanding result. An intent understanding model refers to an artificial intelligence (AI) model that has the ability to perform natural language understanding on instruction text to achieve intent understanding of the instruction text.

[0053] In some embodiments, the intent understanding model can be implemented based on at least one of the following: a classification model, a pre-trained language model, predefined rules, pattern matching, etc., and may also be based on other implementations, which are not limited in this application embodiment. The classification model may include, but is not limited to, at least one of the following: SVM (Support Vector Machine), logistic regression, random forest, neural network, etc., and may also include other classification models, which are not limited in this application embodiment. The pre-trained language model may include, but is not limited to, at least one of the following: an autoregressive EasyLanguage model, a masked speech model, etc., and may also include other pre-trained language models, which are not limited in this application embodiment. Predefined rules refer to using pre-set rules or logic to parse instruction text. Pattern matching refers to matching instruction text based on regular expressions or text templates to achieve intent understanding.

[0054] In some embodiments, understanding prompts are obtained, which are used to indicate the intent understanding task of the schematic diagram understanding model on the instruction text. The intent understanding model is implemented based on a large language model. The intent understanding model performs the intent understanding task on the instruction text based on the understanding prompts to obtain the intent understanding result.

[0055] Understanding prompts refers to descriptive text used to guide or constrain the understanding of intent from instruction text. Since intent understanding models are based on large language models, understanding prompts are the prompts for these models. The intent understanding task refers to the task of understanding the intent of instruction text.

[0056] In some embodiments, the understanding prompt information may include, but is not limited to, at least one of the following: task description information, intent recognition description information, entity extraction description information, output specification information, intent understanding examples, etc., and may also include other information, which are not limited in this application embodiment.

[0057] The task description information is used to describe the task objective that the intent understanding model is to perform. For example, the task description information could be "Help me understand the intent of this instruction text".

[0058] Intent recognition description information is used to guide the intent understanding model to extract the user's intent from the instruction text. Optionally, the intent recognition description information may include at least one intent tag. The intent tag is used to indicate the intent type to which the instruction text belongs. Optionally, the intent type may include, but is not limited to, at least one of the following: preference correction, scene setting, temporary preference setting, playback control, search request, etc., and may also include other intent types, which are not limited in this embodiment. In this case, the intent recognition description information is used to guide the intent understanding model to determine one or more intent tags that match the instruction text from at least one intent tag. For example, the intent recognition description information may be "Help me determine which of the following intent tags this instruction text belongs to: preference correction, scene setting, temporary preference setting, playback control, search request".

[0059] Entity extraction description information is used to guide the intent understanding model to extract specified entity content from the instruction text. In some embodiments, the entity extraction description information may include at least one extracted entity. Optionally, the extracted entity may include, but is not limited to, at least one of the following: content type, style attribute, language information, subject matter information, character information, emotion information, etc., and may also include other extracted entities, which are not limited in this embodiment. In this case, the entity extraction description information is used to guide the extraction of the content corresponding to the extracted entity from the instruction text. For example, the entity extraction description information may be "Help me extract the following extracted entity content from the instruction text: singer, style attribute, language, emotion, scene, and output the extracted entity and its content in key-value pair format."

[0060] Output specification information refers to the constraint information that standardizes the output format of the result of understanding the intent of the instruction text. Optionally, the content of the output specification information may include, but is not limited to, at least one of the following: output format, output field names, field order, data type, etc., and may also include other content, which is not limited in this embodiment. Optionally, the output format may be a structured format. It should be noted that the output format has been shown above, and will not be repeated here.

[0061] Intent understanding examples are used to guide intent understanding models to learn or infer the mapping relationship between input (instruction text) and output (intent understanding result).

[0062] For example, an example of intent understanding is as follows: { "intent": "Playback request", "entity": { Content Type: Movie Genre: Suspense Style: Relaxed } } The above approach guides the intent understanding model based on a large language model by introducing understanding prompts, enabling the intent understanding model to perform accurate intent understanding of the instruction text and generate accurate intent understanding results. This improves the accuracy and robustness of intent understanding and enhances the system's adaptability to complex natural language instructions.

[0063] Step 230: Based on the adjusted user profile, determine the content recommendation results, which include at least one recommended audio or video content to be provided to the user.

[0064] In other words, during the playback of audio and video content, the user profile is dynamically adjusted based on the real-time voice commands input by the user, and then new audio and video content is recommended to the user based on the adjusted audio and video content.

[0065] In some embodiments, the content recommendation result may include, but is not limited to, at least one of the following: content identification information, resource location information, recommendation score, hit tag information, etc., and may also include other information, which is not limited in this application embodiment. Content identification information is used to uniquely identify an audio / video content. Optionally, content identification information may include, but is not limited to, at least one of the following: content ID (Identify), globally unique identifier (UUID), database primary key, content name, etc., and may also include other information, which is not limited in this application embodiment. Resource location information is used to indicate the storage location or access path of the audio / video content. Optionally, resource location information may include, but is not limited to, at least one of the following: local storage path, streaming media address, URL (Uniform Resource Locator), CDN (Content Delivery Network) address, etc., and may also include other information, which is not limited in this application embodiment. Recommendation score is used to indicate the degree of matching between the recommended audio / video content and the adjusted user profile. Hit tag information refers to the set of content tags that match the recommended audio / video content with the adjusted user profile.

[0066] In some embodiments, based on the adjusted user profile, at least one recommended audio / video content is determined from each candidate audio / video content to obtain a content recommendation result.

[0067] Candidate audio and video content refers to audio and video content that may be recommended to users. Optionally, candidate audio and video content is stored in the cloud. Cloud storage refers to a service that stores data on a remote server via the internet.

[0068] In some embodiments, a preset recommendation algorithm is used to determine at least one recommended audio / video content from each candidate audio / video content based on the adjusted user profile, thereby obtaining a content recommendation result.

[0069] Recommendation algorithms are used to instruct rules or logic for selecting recommended audio and video content based on adjusted user profiles. In some embodiments, recommendation algorithms can be implemented based on at least one of the following methods: collaborative filtering, content similarity, vector retrieval, scoring and ranking, etc., and may also include other methods, which are not limited in this application embodiment.

[0070] In some embodiments, during the process of determining content recommendation results based on the adjusted user profile, explicit preferences for audio and video content in the adjusted user profile have a higher priority than implicit preferences for audio and video content.

[0071] Step 240: Based on the content recommendation results, play the recommended audio and video content.

[0072] In other words, the audio and video content that the playback system will play next includes recommended audio and video content determined based on the adjusted user profile.

[0073] In some embodiments, the play-to-play list is updated based on at least one recommended audio / video content, resulting in an updated play-to-play list, which includes at least one audio / video content to be played; and each audio / video content in the updated play-to-play list is played sequentially according to a preset playback order. The audio / video content to be played refers to the audio / video content that is about to be played.

[0074] Optionally, at least one recommended audio / video content can be mixed with at least one audio / video content to be played to obtain an updated list of content to be played.

[0075] Optionally, at least one audio / video content to be played can be removed from the play list, and at least one recommended audio / video content can be added to the play list to obtain an updated play list.

[0076] In summary, the technical solution provided in this application obtains the user's real-time voice commands during the playback of audio and video content, and dynamically adjusts the user profile based on the real-time voice commands. This allows the adjusted user profile to reflect the user's real-time explicit preferences. Then, based on the adjusted user profile, the recommended audio and video content is determined, enabling the recommended audio and video content to respond promptly to changes in the user's preferences for audio and video content, thereby improving the accuracy of audio and video content recommendations.

[0077] The following describes the process of modifying user profiles to reflect preferences.

[0078] In some embodiments, the user profile includes preference rule information, which is used to indicate at least one preference rule of the user. The preference rule is constraint information of audio and video content used to describe the user's preferences. Based on the intent understanding result, a first preference rule is generated. Based on the first preference rule, the preference rule information is updated to obtain updated preference rule information. The adjusted user profile includes the updated preference rule information.

[0079] Preference rule information, also known as an active preference rule base, is used to indicate the active constraints on users' preferences for audio and video content. It is used to filter or adjust the ranking of candidate audio and video content during recommendation. Each preference rule corresponds to an explicit constraint on the audio and video content.

[0080] In some embodiments, the type of preference rule may include, but is not limited to, at least one of the following: weight adjustment rule, filtering rule, condition constraint rule, etc., and may also include other rules, which are not limited in this application embodiment.

[0081] In some embodiments, preference rules are implemented as structured data. For example, the filtering rules are as follows: { "Filter": ["Dynamic", "Electronic"] }

[0082] Weighting rules are used to adjust the priority of recommended specified audio and video content. Filtering rules are used to exclude user-specified audio and video content. Conditional constraint rules are used to limit the range of audio and video content preferred by the user. Optionally, the conditional constraints are related to at least one of the following: content type, subject matter, style attributes, language, duration, etc., and may also be related to other factors, which are not limited in this embodiment.

[0083] In some embodiments, a first preference rule is generated based on the intent understanding result by a user profile dynamic trimmer; based on the first preference rule, the preference rule information is updated to obtain the updated preference rule information. The user profile dynamic trimmer is a functional module in the playback system used to modify user profile preferences. The user profile dynamic trimmer implements the logic for modifying user profile preferences.

[0084] The above approach introduces preference rule information into user profiles and updates the preference rule information based on the intent understanding results, thereby adjusting the user profiles. This allows the adjusted user profiles to dynamically reflect users' explicit preferences for audio and video content, thus improving the granularity of user preference modeling. At the same time, by applying preference rules to the recommendation process, effective screening and constraints on candidate audio and video content are achieved, improving the accuracy and controllability of content recommendation results and enhancing the system's real-time response capability to changes in user preferences.

[0085] In some embodiments, the intent understanding result includes preference correction information, which is used to indicate the audio and video content to be adjusted in the user profile; the first preference rule includes at least one of the following: increasing the weight of audio and video content that matches the preference correction information; decreasing the weight of audio and video content that matches the preference correction information; and filtering audio and video content that matches the preference correction information.

[0086] Preference correction information is used to indicate the type of user intent in real-time voice commands as preference correction, i.e., correcting the user profile, and how to correct the user profile.

[0087] In some embodiments, the preference correction information includes a rule type label and a correction content label. The rule type label indicates the type to which the first preference rule belongs, and the correction content label indicates the content label to which the audio / video content to be adjusted belongs. Based on the rule type label and the correction content label, the first preference rule is generated. It should be noted that different rule type labels may correspond to different types of preference rules, and different types of preference rules are generated in different ways.

[0088] In some embodiments, the generation method corresponding to the rule type tag is adopted to generate the first preference rule based on the rule type tag and the correction content tag.

[0089] In some embodiments, when the rule type label is a weight adjustment rule, a weight correction amount is obtained. This weight correction amount indicates the weight to be adjusted for audio / video content matching the preference correction information. A first preference rule is generated based on the weight correction amount, the rule type label, and the correction content label. The weight correction amount indicates the amount of weight adjustment for audio / video content matching the correction content label. In this case, the first preference rule indicates that the weight of audio / video content matching the correction content label needs to be adjusted according to the weight correction amount. Optionally, the preference correction information includes the weight correction amount. That is, the weight correction amount is determined by performing intent analysis on the instruction text to adjust the weight of audio / video content matching the correction content label. Optionally, the weight correction amount is preset by a relevant technical person.

[0090] In some embodiments, the weight adjustment rules include weight increase rules and weight decrease rules. Weight increase rules instruct the increase of weight for audio and video content matching the preference correction information, while weight decrease rules instruct the decrease of weight for audio and video content matching the preference correction information. In this case, increasing the weight of audio and video content matching the preference correction information can also be understood as increasing the weight of audio and video content matching the correction content tags; conversely, decreasing the weight of audio and video content matching the preference correction information can also be understood as decreasing the weight of audio and video content matching the correction content tags.

[0091] In some embodiments, the preference correction information includes a correction level, which indicates that different correction levels correspond to a preset weight correction amount; a first preference rule is generated based on the weight correction amount corresponding to the correction level, the rule type label, and the correction content label. Optionally, the correction level is related to at least one of the following: matching degree, speech expression intensity, tone intensity, etc., and may also be related to other factors, which are not limited in this embodiment.

[0092] For example, the weight adjustment rules can be as follows: { Action: +0.2 Variety Shows: -0.3 }

[0093] In some embodiments, when the rule type label is a filtering rule, a first preference rule is generated based on the rule type label and the correction content label. Optionally, when the rule type label is a filtering rule, a weight decay factor is obtained, which indicates the degree of weight decay of audio and video content matching the correction content label; the first preference rule is generated based on the weight decay factor, the rule type label, and the correction content label. For example, the weight decay factor can be 0 or 0.1. It should be noted that the weight decay factor can also be any value. That is, filtering audio and video content that matches the preference correction information can also be understood as filtering audio and video content that matches the correction content label.

[0094] For example, the filtering rules can be as follows: { Filtered: ["Quiet", "Sad"] } or, { "Quiet": "×0.1", "Sadness": "×0.1" }

[0095] In some embodiments, when the rule type label is a conditional constraint rule, the constraint range is obtained, which is used to limit the range of conditions that the audio and video content must meet; based on the constraint range, the rule type label, and the modified content label, a first preference rule is generated.

[0096] For example, the condition constraint rule can be as follows: { Duration: <60 minutes }

[0097] The above method introduces preference correction information into the intent understanding result and generates a corresponding first preference rule based on the preference correction information. It then performs weight increase, weight decrease, or filtering operations on the audio and video content that matches the preference correction information, thereby achieving fine-grained adjustment of the recommended audio and video content, improving the accuracy and controllability of the content recommendation results, and enhancing the responsiveness to real-time changes in user preferences.

[0098] In some embodiments, the user profile includes a long-term profile, which indicates the user's implicit preferences for audio and video content; based on the updated preference rule information, each candidate audio and video content is labeled to obtain a preference label for each candidate audio and video content, which indicates the weight adjustment amount of the candidate audio and video content or whether the candidate audio and video content is filtered; based on the long-term profile and the preference labels of each candidate audio and video content, at least one recommended audio and video content is determined from each candidate audio and video content to obtain the content recommendation result.

[0099] Long-term profiles are a part of user profiles, used to represent implicit preference models built from historical behavioral data, and are characterized by their relative stability. In some embodiments, historical behavioral data may include, but is not limited to, at least one of the following: historical viewing records, click behavior, dwell time, search records, collection behavior, liking behavior, etc., and may also include other historical behaviors, which are not limited in this application embodiment.

[0100] In some embodiments, long-term profiles and preference rule information are stored independently. By storing stable long-term profiles and real-time preference rule information separately, the stability of long-term profiles can be ensured while enabling rapid updates and flexible application of preference rules.

[0101] Tagging each candidate audio and video content refers to matching and judging each candidate audio and video content based on the preference rule information in the adjusted user profile, and marking whether the candidate audio and video content needs to be corrected or filtered.

[0102] In some embodiments, preference tags may include at least one of the following: hit preference rules, filter tags, etc., and may also include other content, which is not limited in this application embodiment.

[0103] In some embodiments, for each candidate audio / video content, at least one content tag of the candidate audio / video content is obtained; based on the at least one content tag of the candidate audio / video content, it is matched with the modified content tag corresponding to each preference rule to obtain the preference tag of the candidate audio / video content.

[0104] In some embodiments, each candidate audio and video content is filtered based on its preference tag to obtain at least one remaining candidate audio and video content; based on the preference tag and long-term profile of the at least one remaining candidate audio and video content, a recommendation score is determined for each remaining candidate audio and video content, the recommendation score of the candidate audio and video content being used to indicate the degree of matching between the candidate audio and video content and the user's preferences; based on the recommendation scores of each remaining candidate audio and video content, at least one recommended audio and video content is determined to obtain a content recommendation result.

[0105] It should be noted that the recommendation algorithm for determining recommended audio and video content based on long-term profiles and preference rule information can also be other recommendation algorithms, which can be designed by relevant technical personnel. This application embodiment does not limit this.

[0106] In some embodiments, an initial policy model is trained based on a first preference rule to obtain a trained policy model. Optionally, the initial policy model can be a policy model built based on user profiles. The trained policy model is constrained by the first preference rule. In some embodiments, content recommendation results are generated using the trained policy model.

[0107] The above method introduces long-term profiles to represent users' implicit preferences and tags candidate audio and video content based on updated preference rule information to obtain preference tags used to indicate weight adjustment or content filtering. Furthermore, by combining stable long-term profiles and real-time preference tags, candidate audio and video content is sorted and filtered to determine recommended audio and video content, thereby improving the accuracy, real-time performance and controllability of the recommendation results.

[0108] The following describes the process of setting scenarios for user profiles.

[0109] In some embodiments, the user profile includes scene preference information, which indicates the user's preference for audio and video content in different scenarios; a first environmental feature is obtained, which describes the first scenario in which the user is at a first moment; a first scene preference feature is generated based on the first environmental feature and the intent understanding result; the scene preference information is updated based on the first scene preference feature to obtain the updated scene preference information, and the adjusted user profile includes the updated scene preference information.

[0110] Scene preference information, also known as a scene preference mapping library, stores the mapping relationship between the environmental features of a scene and the content features of the recommended audio and video content. This indicates the user's different preferences for audio and video content in different scenes. This allows the playback system to recommend audio and video content corresponding to similar scenes when it detects them.

[0111] The first moment is the instant the playback system receives the real-time voice command. The first scene is the scene the user is in at that first moment.

[0112] In some embodiments, the first environmental feature may include, but is not limited to, at least one of the following: time, location, device status, network status, user activity status, vehicle status, weather status, etc., and may also include other information, which is not limited in this application embodiment. Time refers to the first moment. Location refers to the geographical location of the first scene. Device status is used to indicate the operating status of the terminal device running the playback system. Optionally, device status may include, but is not limited to, at least one of the following: device battery status, device performance status, device volume status, device screen status, whether the device is locked, whether the device is in power saving mode, device storage status, etc., and may also include other information, which is not limited in this application embodiment. Network status is used to indicate the quality of the network of the terminal device running the playback system. User activity status is used to indicate the user's activity at the first moment. Optionally, user activity status may include, but is not limited to, at least one of the following: walking status, stationary status, running status, driving status, working status, leisure status, commuting status, exercising status, sleeping status, watching status, etc., and may also include other statuses, which is not limited in this application embodiment. Vehicle status refers to the status of the vehicle when the user is in the vehicle and the vehicle is in motion. Weather status refers to the weather conditions of the first scene at the first moment. Optionally, the weather conditions may include, but are not limited to, at least one of the following: weather type, humidity, temperature, wind force level, precipitation, air quality, etc., and may also include other information, which are not limited in this application embodiment.

[0113] In some embodiments, the source of the first environmental feature may include, but is not limited to, at least one of the following: data collected by terminal devices, data collected by sensors, data provided by the operating system, data obtained by applications, data obtained by external service interfaces, user input, speed sensor, vehicle CAN bus data (such as windshield wiper activation), mobile phone biosensors (such as heart rate to determine exercise status), smart home devices (such as dimming lights to determine nighttime rest), etc., and may also include other sources, which are not limited in this application embodiment.

[0114] In some embodiments, a first scene preference feature is generated by a scene preference memory manager based on a first environmental feature and intent understanding results; based on the first scene preference feature, the scene preference information is updated to obtain updated scene preference information. The scene preference memory manager is a functional module in the playback system used to establish and manage the mapping relationship between environmental features and content features of different scenes. The scene preference memory manager implements the logic for establishing and managing the mapping relationship between environmental features and content features of different scenes.

[0115] The above method obtains first environmental features and generates first scene preference features by combining intent understanding results. It models the user's preferences for audio and video content in specific scenes and updates the scene preference information in the user profile based on the first scene features to obtain updated scene preference information. This allows the adjusted user profile to reflect the user's differentiated preferences in different scenes, realizes the contextual expression and dynamic updating of user preferences, and improves the matching degree between content recommendation results and the user's current scene.

[0116] In some embodiments, the intent understanding result includes scene content features, which are used to indicate the attributes of audio and video content preferred by the user in a first scene; feature extraction is performed on each audio and video content to be played to obtain content features of each audio and video content to be played, which are used to indicate the attributes of the audio and video content to be played; a first content feature is generated based on the scene content features and the content features of each audio and video content to be played; and a first scene preference feature is generated based on the first environment features and the first content features.

[0117] Scene content features are used to indicate the type of user intent in real-time voice commands as a scene setting, that is, the mapping relationship between the environmental features and content features of the scene.

[0118] In some embodiments, a first content feature is generated based on scene content features and / or each audio and video content to be played.

[0119] In some embodiments, the scene content features include at least one scene content tag, which indicates the audio and video content preferred by the user in a first scene. The content features of the audio and video content to be played include at least one content tag. Based on the content features of each audio and video content to be played, a content feature to be played is determined, which indicates the tag distribution of each audio and video content to be played across various content tags. Based on the content feature to be played and / or the scene content features, a first content feature is generated. The content feature to be played includes at least one content tag, which indicates the degree of association between each audio and video content to be played and the content tag.

[0120] In other embodiments, a first scene preference feature is generated based on scene content features and a first content feature.

[0121] In some embodiments, the first scene preference feature is implemented as a first environment feature and a first content feature in the form of key-value pairs.

[0122] The above method extracts scene content features from intent understanding results and extracts content features from the audio and video content to be broadcast. It aligns user preferences and content attributes in a unified feature space, generates first content features based on scene content features and content features, and generates first scene preference features by combining first environment features. This achieves multi-dimensional fusion modeling of user preferences and content attributes, improving the scene adaptability and content matching accuracy of recommendation results.

[0123] In some embodiments, the user profile includes a long-term profile, which indicates the user's implicit preference for audio and video content. The scene preference information includes at least one scene preference feature, which includes environmental features and content features. The latest environmental feature is obtained, which indicates the second scene in which the user is at a second moment. The first moment is earlier than the second moment. The latest environmental feature is matched with the environmental features in each scene preference feature to obtain a scene matching result. The scene matching result indicates whether the second scene matches the corresponding scene preference feature. If the scene matching result indicates that the second scene matches the corresponding scene preference feature, based on the scene preference feature corresponding to the second scene and the long-term profile, at least one recommended audio and video content is determined from each candidate audio and video content to obtain a content recommendation result.

[0124] The second moment refers to the current moment. In other words, the second scenario is the scenario in which the user is currently situated. The latest environmental features are the environmental features used to describe the scenario in which the user is currently situated.

[0125] The system compares the environmental features of each scene preference feature with the latest environmental features and scene preference information to determine whether the second scene matches the set scene. If a matching scene is found, the system uses the content features of the scene and the long-term profile to select recommended audio and video content from each candidate audio and video content, so as to accurately recommend the preferred audio and video content to the user in the current scene.

[0126] In some embodiments, the matching method between the latest environmental features and the environmental features in the scene preference features may include, but is not limited to, at least one of the following: exact matching, tolerance matching, similarity matching, etc., and matching may also be performed in other ways. This application embodiment does not limit this.

[0127] The above method constructs a user profile that includes long-term profiles and scene preference information, and matches the latest environmental features with the environmental features in the scene preference features. When a match is successful, the candidate audio and video content is filtered by combining the corresponding scene preference features with the long-term profile, thereby realizing scene-based personalized recommendations.

[0128] The process of setting temporary preferences is described below.

[0129] In some embodiments, the intent understanding result is recognized to obtain constraint conditions in at least one dimension. The constraint conditions refer to the conditions that the recommended audio-visual content should satisfy. Based on the at least one constraint condition, a content query instruction is generated, and the content query instruction is used to trigger the determination of the content recommendation result in combination with the adjusted user profile. Based on the content query instruction and the adjusted user profile, the content recommendation result is determined.

[0130] In the embodiments of the present application, the constraint condition is short-term preference information. That is to say, the life cycle of the restriction of the constraint condition on the audio-visual content recommendation process is non-persistent relative to the long-term profile, preference rule information or scenario preference information. In some embodiments, the life cycle of the restriction of the constraint condition on the audio-visual content recommendation process can be the life cycle of a session or a preset effective duration (such as one hour), and the embodiments of the present application do not limit this.

[0131] In some embodiments, the content query instruction is in the form of a database query statement. In some embodiments, the constraint condition is implemented as structured data.

[0132] In some embodiments, a query instruction template is obtained. The query instruction template refers to a formatted template for describing the content query structure. Based on the at least one constraint condition and the query instruction template, a content query instruction is generated.

[0133] In some embodiments, at least one constraint condition is converted into a content query instruction through a large language model.

[0134] It should be noted that the method for determining the content recommendation result has been shown above, and the embodiments of the present application will not elaborate here.

[0135] In some embodiments, based on the adjusted user profile, the content query instruction is executed to obtain the content recommendation result.

[0136] For example, the intent understanding result is: "Style: Popular ancient style, Emotion: Exciting", and the generated content query instruction is "SELECT FROM songs WHERE genre LIKE ‘%ancient style%’ AND tags CONTAINS ‘popular’ AND emotion_score>0.7".

[0137] In other embodiments, the intent understanding result is identified to obtain at least one dimension of constraints, which refer to the conditions that the recommended audio and video content must meet. Based on at least one constraint, a content query instruction is generated, which triggers the determination of content recommendation results in conjunction with the adjusted user profile. Based on the content query instruction and the user profile, the content recommendation result is determined. That is to say, even without adjusting the user profile, the playback system can be controlled to recommend corresponding audio and video content based on temporary preferences through real-time voice commands.

[0138] In some embodiments, semantic analysis of the intent understanding results is performed using a semantic understanding model to obtain constraints in at least one dimension; wherein the dimensions of the constraints include at least one of the following: scene, language, and style attributes.

[0139] In some embodiments, the dimensions of the constraints may include, but are not limited to, at least one of the following: content type, subject matter information, character information, emotional information, etc., and may also include other dimensions, which are not limited in this application embodiment.

[0140] A semantic understanding model is an AI model with the ability to understand and analyze semantics. In some embodiments, a semantic understanding model can be implemented based on at least one of the following: a classification model, a pre-trained language model, predefined rules, pattern matching, etc., and may also be based on other implementations, which are not limited in this application embodiment.

[0141] In some embodiments, the semantic understanding model and the intent understanding model can be the same model or different models.

[0142] For example, please refer to Figure 3 The diagram illustrates a timing diagram of a temporary preference setting provided in an embodiment of this application. The process of setting the temporary preference setting includes the following steps.

[0143] 1. The terminal device sends a real-time voice command to the server: "Find some upbeat rock music to listen to while driving at night."

[0144] 2. The server performs intent understanding on real-time voice commands and obtains the intent understanding results. Intent understanding results: Scene = Night + In-vehicle, Emotion = Excitement, Genre = Rock.

[0145] 3. Based on the intent understanding results, the server generates content query instructions.

[0146] 4. The server sends a content query command to the audio and video resource library.

[0147] 5. The audio and video resource library provides the server with at least one candidate audio and video content.

[0148] 6. Based on the (adjusted) user profile, the server determines the content recommendation result from at least one candidate audio and video content.

[0149] 7. The server sends content recommendation results to the terminal device.

[0150] 8. Based on the content recommendation results, the terminal device plays recommended audio and video content. Voice announcement: "Playing a rock energy playlist for night driving."

[0151] The above method identifies the intent understanding results to generate at least one constraint to indicate temporary preferences, and generates a content query instruction based on the constraint to drive the determination of content recommendation results in combination with the adjusted user profile. This realizes the structuring of user intent and its transformation into an executable query, so that the audio and video content recommended by the playback system can meet the user's temporary needs.

[0152] The following is an architecture diagram and complete process for dynamically adjusting user profiles.

[0153] Please refer to Figure 4 This illustrates an architecture diagram of dynamically adjusting user profiles provided in one embodiment of this application.

[0154] The playback system is deployed on a client-server architecture (i.e., terminal device). The client (mobile phone, in-vehicle system) is responsible for voice capture, wake-up, simple front-end NLU, and command uploading. The server is responsible for deep NLP, user profile calculation, recommendation engine, music library management, and other functions.

[0155] The voice acquisition module of the terminal device is responsible for collecting the user's real-time voice commands and serves as the entry point for user interaction with the system.

[0156] Command receiving and preprocessing module: performs preliminary processing on the acquired real-time voice commands (such as noise reduction, format conversion, frame segmentation, etc.) to prepare for subsequent scene recognition and semantic understanding.

[0157] Scene recognition module: Analyzes preprocessed real-time voice (or combines device status and environmental information) to identify the user's current scene (such as in-vehicle, home, office, sports, etc.).

[0158] Network transmission module: Transmits the identified environmental features and preprocessed real-time voice commands to the server via the network.

[0159] The server's natural language understanding engine: performs intent analysis on the transmitted real-time voice commands to understand the user's true intentions.

[0160] Intent classifier: Accurately classifies intents based on their understanding, determining the direction of subsequent services. (1) Scene setting: Supported by a scene-based preference memory manager and a scene preference mapping library. Based on user historical behavior data and content feature library, a mapping relationship between environmental features and content features of different scenes is constructed.

[0161] (2) Preference Adjustment: Supported by a user profile dynamic adjustment adjuster and an active preference rule base. Based on the intent understanding results, user preferences are dynamically adjusted to ensure the timeliness and accuracy of recommendations.

[0162] (3) Application function instruction converter: Converts the categorized user intent into specific application function instructions (such as content query instructions).

[0163] Audio and video resource library: Stores various types of audio and video content and calls up relevant resources according to instructions.

[0164] Historical behavior data: used to store users' past operation records (such as playing, favorite, skipping) and preference settings (such as favorite singers, scene tags).

[0165] Content Feature Library: Used to store content tags for audio and video content, which are used for content filtering and matching.

[0166] The following are embodiments of the apparatus described in this application, which can be used to execute the embodiments of the method described in this application. For details not disclosed in the apparatus embodiments of this application, please refer to the embodiments of the method described in this application.

[0167] Please refer to Figure 5 This diagram illustrates a block diagram of an audio / video content recommendation device according to an embodiment of this application. The device has the functions described above, which can be implemented in hardware or by hardware executing corresponding software. The device can be the terminal device 10 described above, or it can be located within the terminal device 10; alternatively, the device can be the server 20 described above, or it can be located within the server 20. Figure 5 As shown, the device 500 may include an acquisition module 510, an adjustment module 520, a determination module 530, and a playback module 540.

[0168] The acquisition module 510 is used to acquire real-time voice commands input by the user during the playback of audio and video content. The real-time voice commands are used to indicate the user's preferences for audio and video content.

[0169] The adjustment module 520 is used to adjust the user profile of the user based on the real-time voice command to obtain an adjusted user profile, wherein the user profile is used to indicate the user's explicit and implicit preferences for audio and video content.

[0170] The determining module 530 is used to determine content recommendation results based on the adjusted user profile, the content recommendation results including at least one recommended audio or video content to be provided to the user.

[0171] The playback module 540 is used to play the recommended audio and video content based on the content recommendation results.

[0172] In some embodiments, the adjustment module 520 includes an identification submodule, an understanding submodule, and an adjustment submodule (in... Figure 5 (Not shown in the image).

[0173] The recognition submodule is used to perform speech recognition on the real-time voice command to obtain the command text corresponding to the real-time voice command.

[0174] The understanding submodule is used to perform intent understanding on the instruction text to obtain intent understanding results, which are used to indicate the user's explicit preferences for audio and video content.

[0175] The adjustment submodule is used to adjust the user profile of the user based on the intent understanding result, so as to obtain the adjusted user profile.

[0176] In some embodiments, the user profile includes preference rule information, which is used to indicate at least one preference rule of the user. The preference rule is constraint information for describing the audio and video content preferred by the user. The adjustment submodule is used to generate a first preference rule based on the intent understanding result. Based on the first preference rule, the preference rule information is updated to obtain updated preference rule information. The adjusted user profile includes the updated preference rule information.

[0177] In some embodiments, the intent understanding result includes preference correction information, which is used to indicate the audio and video content to be adjusted in the user profile; the first preference rule includes at least one of the following: increasing the weight of audio and video content that matches the preference correction information; decreasing the weight of audio and video content that matches the preference correction information; and filtering audio and video content that matches the preference correction information.

[0178] In some embodiments, the user profile includes a long-term profile, which indicates the user's implicit preference for audio and video content; the determining submodule is used to mark each candidate audio and video content based on the updated preference rule information to obtain a preference tag for each candidate audio and video content, the preference tag indicating the weight adjustment amount of the candidate audio and video content or the candidate audio and video content being filtered; based on the long-term profile and the preference tags of each candidate audio and video content, at least one recommended audio and video content is determined from each candidate audio and video content to obtain the content recommendation result.

[0179] In some embodiments, the user profile includes scene preference information, which indicates the user's preference for audio and video content in different scenarios; the adjustment submodule is further configured to obtain a first environmental feature, which describes the first scenario in which the user is located at a first moment; generate a first scene preference feature based on the first environmental feature and the intent understanding result; update the scene preference information based on the first scene preference feature to obtain updated scene preference information, and the adjusted user profile includes the updated scene preference information.

[0180] In some embodiments, the intent understanding result includes scene content features, which are used to indicate the attributes of the audio and video content preferred by the user in the first scene; the adjustment submodule is further configured to extract features from each audio and video content to be played to obtain content features of each audio and video content to be played, which are used to indicate the attributes of the audio and video content to be played; generate a first content feature based on the scene content features and the content features of each audio and video content to be played; and generate a first scene preference feature based on the first environment features and the first content features.

[0181] In some embodiments, the user profile includes a long-term profile, which indicates the user's implicit preference for audio and video content. The scene preference information includes at least one scene preference feature, which includes environmental features and content features. The adjustment submodule is further configured to obtain the latest environmental features, which indicate the second scene in which the user is located at a second time. The first time is earlier than the second time. The latest environmental features are matched with the environmental features in each of the scene preference features to obtain a scene matching result. The scene matching result indicates whether the second scene matches the corresponding scene preference feature. If the scene matching result indicates that the second scene matches the corresponding scene preference feature, based on the scene preference feature corresponding to the second scene and the long-term profile, at least one recommended audio and video content is determined from each of the candidate audio and video content to obtain the content recommendation result.

[0182] In some embodiments, the apparatus 500 further includes a generation module (in Figure 5 (Not shown in the image) is used to identify the intent understanding result and obtain at least one dimension of constraint conditions, which refer to the conditions that the recommended audio and video content must meet; based on the at least one constraint condition, a content query instruction is generated, which is used to trigger the determination of the content recommendation result in combination with the adjusted user profile; the determination module 530 is also used to determine the content recommendation result based on the content query instruction and the adjusted user profile.

[0183] In some embodiments, the generation module is further configured to perform semantic analysis on the intent understanding result through a semantic understanding model to obtain the constraint conditions of the at least one dimension; wherein the dimension of the constraint conditions includes at least one of the following: scene, language, and style attributes.

[0184] In some embodiments, the understanding submodule is further configured to obtain understanding prompt information, which indicates the intent understanding task of the schematic diagram understanding model for the instruction text, the intent understanding model being implemented based on a large language model; and to perform the intent understanding task on the instruction text based on the understanding prompt information by the intent understanding model to obtain the intent understanding result.

[0185] In summary, the technical solution provided in this application obtains the user's real-time voice commands during the playback of audio and video content, and dynamically adjusts the user profile based on the real-time voice commands. This allows the adjusted user profile to reflect the user's real-time explicit preferences. Then, based on the adjusted user profile, the recommended audio and video content is determined, enabling the recommended audio and video content to respond promptly to changes in the user's preferences for audio and video content, thereby improving the accuracy of audio and video content recommendations.

[0186] It should be noted that the apparatus provided in the above embodiments is only illustrated by the division of the above functional modules when implementing its functions. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the apparatus and method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process can be found in the method embodiments, which will not be repeated here.

[0187] Please refer to Figure 6 This diagram illustrates a structural block diagram of a computer device 600 provided in one embodiment of this application. The computer device 600 may be... Figure 1 The terminal device 10 in the playback system shown can also be Figure 1 The server 20 in the playback system shown is used to implement the audio and video content recommendation method provided in the above embodiments. Specifically: Typically, computer device 600 includes a processor 610 and a memory 620.

[0188] Processor 610 may include one or more processing cores, such as a quad-core processor or an octa-core processor. Processor 610 may be implemented using at least one hardware form selected from Digital Signal Processing (DSP), Field Programmable Gate Array (FPGA), and Programmable Logic Array (PLA). Processor 610 may also include a main processor and a coprocessor. The main processor, also known as the Central Processing Unit (CPU), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, processor 610 may integrate a Graphics Processing Unit (GPU), which is responsible for rendering and drawing the content to be displayed on the screen. In some embodiments, processor 610 may also include an AI processor for handling computational operations related to machine learning.

[0189] The memory 620 may include one or more computer-readable storage media, which may be non-transitory. The memory 620 may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices or flash memory devices. In some embodiments, the non-transitory computer-readable storage media in the memory 620 are used to store a computer program configured to be executed by one or more processors to implement the recommended method for the above-described audio and video content.

[0190] Those skilled in the art will understand that Figure 6 The structure shown does not constitute a limitation on the computer device 600, and may include more or fewer components than shown, or combine certain components, or use different component arrangements.

[0191] In an exemplary embodiment, a computer-readable storage medium is also provided, wherein a computer program is stored therein, which, when executed by a processor, implements the recommended method for the aforementioned audio and video content. Optionally, the computer-readable storage medium may include: read-only memory (ROM), random access memory (RAM), solid-state drives (SSDs), or optical discs, etc. The random access memory may include resistive random access memory (ReRAM) and dynamic random access memory (DRAM).

[0192] In an exemplary embodiment, a computer program product is also provided, the computer program product including a computer program stored in a computer-readable storage medium. A processor of a computer device reads the computer program from the computer-readable storage medium, and the processor executes the computer program, causing the computer device to perform the recommended method for the audio and video content described above.

[0193] It should be understood that "multiple" as used herein refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. Furthermore, the step numbers described herein are merely illustrative of one possible execution order. In some other embodiments, the steps may not be executed in numerical order, such as two steps with different numbers being executed simultaneously, or two steps with different numbers being executed in the reverse order of the illustration. This application does not limit this.

[0194] The above description is merely an optional embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for recommending audio and video content, characterized in that, The method includes: During the playback of audio and video content, real-time voice commands input by the user are obtained, and the real-time voice commands are used to indicate the user's preferences for audio and video content; Based on the real-time voice command, the user profile of the user is adjusted to obtain the adjusted user profile, which is used to indicate the user's explicit and implicit preferences for audio and video content. Based on the adjusted user profile, a content recommendation result is determined, which includes at least one recommended audio or video content to be provided to the user. Based on the content recommendation results, the recommended audio and video content is played.

2. The method according to claim 1, characterized in that, The step of adjusting the user profile based on the real-time voice command to obtain the adjusted user profile includes: The real-time voice command is subjected to speech recognition to obtain the command text corresponding to the real-time voice command; The instruction text is subjected to intent understanding to obtain intent understanding results, which are used to indicate the user's explicit preferences for audio and video content; Based on the intent understanding results, the user profile of the user is adjusted to obtain the adjusted user profile.

3. The method according to claim 2, characterized in that, The user profile includes preference rule information, which is used to indicate at least one preference rule of the user. The preference rule is constraint information used to describe the audio and video content preferred by the user. The step of adjusting the user profile based on the intent understanding result to obtain the adjusted user profile includes: Based on the intent understanding results, a first preference rule is generated; Based on the first preference rule, the preference rule information is updated to obtain updated preference rule information, and the adjusted user profile includes the updated preference rule information.

4. The method according to claim 3, characterized in that, The intent understanding result includes preference correction information, which is used to indicate the audio and video content to be adjusted in the user profile; The first preference rule includes at least one of the following: Increase the weight of audio and video content that matches the preference correction information; Reduce the weight of audio and video content that matches the preference correction information; Filter audio and video content that matches the preference correction information.

5. The method according to claim 3, characterized in that, The user profile includes a long-term profile, which is used to indicate the user's implicit preferences for audio and video content; The determination of content recommendation results based on the adjusted user profile includes: Based on the updated preference rule information, each candidate audio and video content is labeled to obtain a preference label for each candidate audio and video content. The preference label of the candidate audio and video content is used to indicate the weight adjustment amount of the candidate audio and video content or the candidate audio and video content is filtered. Based on the long-term profile and the preference tags of each candidate audio and video content, at least one recommended audio and video content is determined from each candidate audio and video content to obtain the content recommendation result.

6. The method according to claim 2, characterized in that, The user profile includes scene preference information, which is used to indicate the user's preference for audio and video content in different scenes; The step of adjusting the user profile based on the intent understanding result to obtain the adjusted user profile includes: Obtain a first environmental feature, which describes the first scenario in which the user is located at a first moment; Based on the first environmental features and the intent understanding result, a first scene preference feature is generated; Based on the first scenario preference feature, the scenario preference information is updated to obtain updated scenario preference information, and the adjusted user profile includes the updated scenario preference information.

7. The method according to claim 6, characterized in that, The intent understanding result includes scene content features, which are used to indicate the attributes of the audio and video content preferred by the user in the first scene; The step of generating first scene preference features based on the first environmental features and the intent understanding result includes: Feature extraction is performed on each video content to be played to obtain the content features of each video content to be played. The content features of the video content to be played are used to indicate the attributes of the video content to be played. Based on the scene content features and the content features of each of the audio and video contents to be played, a first content feature is generated; Based on the first environmental features and the first content features, the first scene preference features are generated.

8. The method according to claim 7, characterized in that, The user profile includes a long-term profile, which indicates the user's implicit preference for audio and video content. The scene preference information includes at least one scene preference feature, which includes environmental features and content features. The step of adjusting the user profile based on the intent understanding result to obtain the adjusted user profile includes: The latest environmental features are obtained, which are used to indicate the second scenario in which the user is located at a second time; wherein the first time is earlier than the second time. The latest environmental features are matched with the environmental features in each of the scene preference features to obtain a scene matching result. The scene matching result is used to indicate whether the second scene matches the corresponding scene preference feature. If the scene matching result indicates that the second scene matches the corresponding scene preference feature, based on the scene preference feature corresponding to the second scene and the long-term profile, at least one recommended audio and video content is determined from each of the candidate audio and video content to obtain the content recommendation result.

9. The method according to claim 2, characterized in that, The method further includes: The intent understanding result is identified to obtain at least one dimension of constraint conditions, which refer to the conditions that the recommended audio and video content must meet; Based on the at least one constraint, a content query instruction is generated, which is used to trigger the determination of the content recommendation result in combination with the adjusted user profile; The process of determining content recommendation results based on the adjusted user profile includes: Based on the content query command and the adjusted user profile, the content recommendation result is determined.

10. The method according to claim 9, characterized in that, The process of identifying the intent understanding result to obtain at least one constraint includes: The semantic understanding results are semantically analyzed using a semantic understanding model to obtain the constraints of at least one dimension. The dimensions of the constraints include at least one of the following: scene, language, and style attributes.

11. The method according to claim 2, characterized in that, The process of performing intent understanding on the instruction text to obtain the intent understanding result includes: Obtaining understanding prompts, wherein the understanding prompts are used to indicate the intent understanding task of the schematic diagram understanding model for the instruction text, and the intent understanding model is implemented based on a large language model; The intent understanding model performs the intent understanding task on the instruction text based on the understanding prompt information to obtain the intent understanding result.

12. A computer device, characterized in that, The computer device includes a processor and a memory, the memory storing a computer program that is loaded and executed by the processor to implement the method as described in any one of claims 1 to 11.

13. A computer-readable storage medium, characterized in that, The storage medium stores a computer program that is executed by a processor to implement the method as described in any one of claims 1 to 11.

14. A computer program product, characterized in that, The computer program product includes a computer program that is loaded and executed by a processor to implement the method as described in any one of claims 1 to 11.

Images