Real-time interaction method, device and equipment based on digital human, medium and program

Abstract

The invention provides a real-time interaction method and device based on a digital human, equipment, a medium and a program, and relates to the technical field of data processing, in particular to artificial intelligence, natural voice processing, voice technology, deep learning, big data, virtual reality and big language model technology. The method comprises the following steps: acquiring real-time interaction data of each user in a current multi-user real-time interaction scene; classifying the users according to the real-time interaction data to obtain a plurality of classified user groups; screening a current classified user group from the classified user groups; determining target interaction data of the current classification user group; and generating digital human interaction data according to the target interaction data, thereby enabling the digital human to perform real-time interaction with the current classification user group according to the digital human interaction data. According to the embodiment of the invention, the interaction effect in a multi-user concurrent real-time interaction scene can be improved, so that the interaction experience is improved.

Description

Technical Field

[0001] The present disclosure relates to the field of data processing technology, and specifically to artificial intelligence, natural speech processing, speech technology, deep learning, big data, virtual reality, and large language model technology. Background Art

[0002] AI (Artificial Intelligence) can continuously accumulate knowledge and experience through learning, and use what it has learned to solve various problems. Different learning methods and techniques can be applied to different scenarios and problems to provide intelligent solutions. With the rapid development of AI technology, the application of AI in human-computer interaction has become more and more extensive. From the initial simple call center system to the current powerful intelligent assistant and even multi-user real-time interactive system, AI technology can not only interact with users in real time to answer questions, but also deeply understand users' needs and emotions. Summary of the invention

[0003] The embodiments of the present disclosure provide a real-time interaction method, device, equipment, medium and program based on digital humans, which can improve the interaction effect in a multi-user concurrent real-time interaction scenario, thereby improving the interaction experience.

[0004] In a first aspect, the present disclosure provides a real-time interaction method based on a digital human, comprising:

[0005] Obtain the real-time interaction data of each user in the current multi-user real-time interaction scenario;

[0006] Classifying the users according to the real-time interaction data to obtain multiple classified user groups;

[0007] Filter the current classified user group from each of the classified user groups;

[0008] Determining target interaction data of the currently classified user group;

[0009] Digital human interaction data is generated according to the target interaction data, so that the digital human can interact with the current classified user group in real time according to the digital human interaction data.

[0010] In a second aspect, the present disclosure provides a real-time interactive device based on a digital human, including:

[0011] A real-time interaction data acquisition module is used to acquire the real-time interaction data of each user in the current multi-user real-time interaction scenario;

[0012] A classified user group acquisition module is used to classify the users according to the real-time interaction data to obtain multiple classified user groups;

[0013] A current classified user group screening module, used for screening the current classified user group from each of the classified user groups;

[0014] A target interaction data determination module, used to determine the target interaction data of the currently classified user group;

[0015] The group real-time interaction module is used to generate digital human interaction data according to the target interaction data, so that the digital human can interact with the current classified user group in real time according to the digital human interaction data.

[0016] In a third aspect, an embodiment of the present disclosure provides an electronic device, including:

[0017] at least one processor; and

[0018] a memory communicatively connected to the at least one processor; wherein,

[0019] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the real-time interaction method based on digital humans provided in the embodiment of the first aspect.

[0020] In a fourth aspect, an embodiment of the present disclosure further provides a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable the computer to execute the real-time interaction method based on digital humans provided in the embodiment of the first aspect.

[0021] In a fifth aspect, an embodiment of the present disclosure further provides a computer program product, including a computer program, which, when executed by a processor, implements the real-time interaction method based on a digital human provided in the embodiment of the first aspect.

[0022] The disclosed embodiment obtains the real-time interaction data of each user in the current multi-user real-time interaction scenario, classifies the users according to the real-time interaction data, obtains multiple classified user groups, and screens the current classified user group from the classified user groups to determine the target interaction data of the current classified user group, thereby generating digital human interaction data according to the target interaction data, so that the digital human can interact with the current classified user group in real time according to the digital human interaction data, solves the problem of unsatisfactory real-time interaction effect in the existing multi-user concurrent real-time interaction scenario, can realize real-time interaction of batch users, thereby improving the interaction effect in the multi-user concurrent real-time interaction scenario, and further improving the interaction experience.

[0023] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure.

[0025] Figure 1 is a flow chart of a real-time interaction method based on digital human provided by an embodiment of the present disclosure;

[0026] Figure 2 is a flow chart of another real-time interaction method based on digital human provided by an embodiment of the present disclosure;

[0027] Figure 3 It is a schematic diagram of the interaction flow between functional modules in a real-time interactive system based on digital human provided by an embodiment of the present disclosure;

[0028] Figure 4 is another interactive flow diagram between functional modules in a real-time interactive system based on digital human provided by an embodiment of the present disclosure;

[0029] Figure 5 is a structural diagram of a real-time interactive device based on a digital human provided by an embodiment of the present disclosure;

[0030] Figure 6 It is a structural schematic diagram of an electronic device used to implement the real-time interaction method based on digital human in the embodiment of the present disclosure. DETAILED DESCRIPTION

[0031] The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0032] Digital human technology is a technology that combines digital technology, artificial intelligence, voice technology, deep learning, computer graphics, and large language models to create realistic and interactive virtual characters. These virtual characters are not only similar to humans in appearance, but also have certain intelligence and expressiveness in behavior, language, and expression. The core of digital human technology is artificial intelligence technology, including speech recognition, natural language processing, machine learning, and big data. Through these technologies, digital humans can recognize and understand human language and actions, and communicate and interact with humans. In addition, digital human technology also requires technical support from computer graphics, animation design, and virtual reality to create realistic virtual characters and scenes. These technologies can help digital humans present real facial expressions, body language, and actions, thereby enhancing the interactive experience with humans. The application field of digital human technology is very wide. For example, in various multi-person real-time interactive scenarios such as live broadcasts, online learning, and online news releases, digital human technology can be used to create virtual characters to interact with users online in real time.

[0033] At present, when digital human technology is applied to multi-person real-time interaction scenarios, multi-user concurrency often occurs, that is, a large number of users have interaction needs at the same time, such as multiple users asking questions concurrently in live broadcast scenarios and online news releases. At present, digital humans in multi-person real-time interaction scenarios can only interact with each user in a one-to-one interaction mode, such as answering each question in order according to the time when the question was asked or interacting with one user, etc. This one-to-one online interaction mode lacks group information insight and interrelated parallel interaction effects, and the interaction effect it brings is not ideal, and it is difficult to effectively improve the quality and atmosphere of interaction.

[0034] In one example, Figure 1 This is a flowchart of a real-time interaction method based on digital humans provided by an embodiment of the present disclosure. This embodiment is applicable to situations where a digital human interacts with a batch of similar users in a multi-person real-time interaction scenario. The method can be executed by a real-time interaction device based on digital humans. The device can be implemented by software and / or hardware and can generally be integrated into an electronic device. The electronic device can be a terminal device or a server device. As long as the real-time interaction method based on digital humans can be executed, the embodiment of the present disclosure does not limit the specific device type of the electronic device. Accordingly, Figure 1 As shown, the method includes the following operations:

[0035] S110: Acquire real-time interaction data of each user in the current multi-user real-time interaction scenario.

[0036] Among them, the current multi-user real-time interaction scene can be a scene based on real-time interaction between a digital human and multiple online users, for example, it can include but not limited to live broadcast scenes, online learning, and online news releases, etc. As long as there is a demand for multi-person real-time online interaction, the disclosed embodiment does not limit the specific online interaction scene type of the current multi-user real-time interaction scene. The user can be an actual online user in the current multi-user real-time interaction scene. For example, in a live broadcast scene, the user can be a user who interacts with the anchor in the live broadcast room. In an online learning scene, the user can be a user who studies online. The real-time interaction data can be any type of interaction data input by the user in real time in the current multi-user real-time interaction scene.

[0037] In the current multi-user real-time interaction scenario, the real-time interaction system based on the digital human can first obtain the real-time interaction data input by each user in the current multi-user real-time interaction scenario through the group interaction data collection module before the digital human interacts with the user in real time. Optionally, the real-time interaction data can be any type of interaction data, such as text data, voice data, like data, sharing data, and collection data, etc., as long as it can reflect the user's interactive operation. The embodiment of the present disclosure does not limit the data type and data content of the real-time interaction data.

[0038] Exemplarily, taking the live broadcast scenario as an example, the group interaction data collection module can monitor, capture and store the user's interactive behavior data in the live broadcast in real time, including but not limited to user interactive behaviors such as comment content, likes, sharing and collection. It can also consider user viewing time, access to product details and question frequency and other interactive behavior data to provide the data foundation required for the real-time interactive system based on digital humans.

[0039] S120: Classify the users according to the real-time interaction data to obtain multiple classified user groups.

[0040] The classified user group may be a plurality of types of user groups obtained after classifying the users. It is understandable that a classified user group may include at least one user.

[0041] Optionally, the real-time interactive system based on digital humans can also be configured with a group analysis and opinion decision module. The group analysis and opinion decision module can rely on distributed computing and group intelligence technology to analyze the group behavior data of users, including but not limited to sentiment analysis, keyword extraction, topic identification, and opinion aggregation. At the same time, it can be combined with large model technology processing to achieve deep cognition and decision support based on the overall user's wishes, providing a basis for subsequent personalized output.

[0042] Specifically, the group analysis and opinion decision module can be used to classify all users according to the real-time interaction data obtained by the system, and obtain various types of classified user groups. It is understandable that each classified user group can have a typical feature. Optionally, when classifying users, a variety of available classification criteria can be used. For example, the stickiness between the user and the current multi-user real-time interaction scene can be determined based on the real-time interaction data to classify, and obtain new user groups, old user groups, and VIP (Very Important Person) user groups and other group types; the user's activity in the current multi-user real-time interaction scene can also be determined based on the real-time interaction data for classification, such as users with high viewing frequency, long viewing time, and high enthusiasm for interactive activities such as comments and likes can be classified as active user groups; users with relatively low viewing frequency, but long viewing time and interest in specific types of online display content can be classified as potential user groups, and users with relatively low viewing frequency and long viewing time can be classified as lost user groups, etc. In addition, users can also be classified based on their attributes such as gender, age, and geographic location, and based on their focus in the current multi-user real-time interaction scenario. The embodiments of the present disclosure do not limit the classification factors used to classify users.

[0043] It is understandable that, since there is a certain degree of user mobility in the current multi-user real-time interaction scenario, after the users are classified to obtain a plurality of classified user groups, the classified user groups can also be updated according to certain strategies. For example, each time a new user is added, the user is promptly classified into the corresponding classified user group. Alternatively, after the new user is added, the interaction data of the new user within a period of time can be obtained, and the user can be promptly classified into the corresponding classified user group based on the interaction data of the new user within a period of time. When the user leaves the current multi-user real-time interaction scenario, the user can be deleted from the corresponding classified user group.

[0044] S130: Filter the current classified user group from among the classified user groups.

[0045] The current classified user group may be a classified user group with which the digital human needs to interact in real time in the current multi-user real-time interaction scenario.

[0046] Correspondingly, after the real-time interaction system based on digital humans completes the classification of each user in the current multi-user real-time interaction scenario, the corresponding interaction strategy can be used to filter the current classified user group from each classified user group, thereby performing batch interaction on multiple users included in the entire current classified user group in units of user groups.

[0047] S140: Determine target interaction data of the currently classified user group.

[0048] The target interaction data may be interaction data that matches the current classified user group, and may reflect the current interaction characteristics of the current classified user group.

[0049] In the embodiment of the present disclosure, the target interaction data of the current classified user group may be determined according to the real-time interaction data of each user included in the current classified user group.

[0050] Optionally, if the real-time interaction data of each user included in the current classified user group is essentially the same, for example, they are all about product delivery issues or have triggered collection operations, any one of the real-time interaction data can be selected as the target interaction data for the current classified user group.

[0051] For example, if the questions asked by a certain classified user group in the live broadcast room are all about delivery, such as "when will the goods be shipped" or "what is the specific delivery time", then any one of the questions about delivery, such as "what is the specific delivery time", can be selected as the target interaction data of the current classified user group.

[0052] For example, if the questions raised by a certain classified user group in an online press conference are all about the release time, such as "when will it be released" or "the specific release time", then any one of the questions about delivery, such as "when will it be released", can be selected as the target interaction data of the current classified user group.

[0053] S150: Generate digital human interaction data according to the target interaction data, so that the digital human can interact with the currently classified user group in real time according to the digital human interaction data.

[0054] The digital human interaction data may be feedback data generated by a real-time interaction system based on the digital human for target interaction data, and is used for real-time interaction with all users included in the current classified user group based on the digital human image.

[0055] Correspondingly, after the real-time interaction system based on digital humans generates digital human interaction data for the target interaction data, various associated virtual animation data such as the digital human's audio, anthropomorphic movements, and facial expressions can be determined based on the digital human interaction data, so that the digital human can simulate real human characters and interact in real time with all users included in the current classified user group.

[0056] For example, in a live broadcast scenario, a digital human can act as a virtual anchor to interact with multiple live broadcast room users in real time, such as batch interaction with a user group that pays more attention to a hot-selling product, so as to achieve precise promotion of the product. In an online learning scenario, a digital human can act as a virtual teacher or learning partner to explain key knowledge points that attract more attention, so as to help most students learn by reference and interactively. In an online news release scenario, a digital human can act as a virtual news publisher to release news consultations, interact with online users, and answer questions raised by online users.

[0057] Optionally, the digital human technology in the real-time interactive system based on digital human can be implemented based on the large model technology. The large model is also called a large language model or a large language model (LLM), which refers to a deep learning model that can process text sequences using a large amount of relevant data (such as text data, voice data, or text and picture data, etc.), which can generate natural language text or understand the meaning of language text. Such models usually have a parameter scale of more than billions. The large model can handle a variety of natural language tasks, such as text classification, question and answer, and dialogue, and is widely used. The input of the large model is data, such as text data, voice data, or text and picture data, etc. The large model can obtain the corresponding word vector representation by encoding the input data, and further decodes the word vector obtained by the encoding, thereby automatically completing the processing of the input data and obtaining the output data corresponding to the input data. Exemplarily, the text can be input to the large model, and the large model processes and predicts the input text, and outputs the answer text corresponding to the text.

[0058] Optionally, the real-time interactive system based on digital human can be configured with natural language generation and understanding module, multimodal perception and sentiment analysis module, multitask execution management module, recommendation system and cross-selling module, etc. Among them, the natural language generation and understanding module can include NLG (Natural Language Generation) and NLU (Natural Language Understanding) modules, etc., using the large model as the dialogue generation engine between the digital human and the user, processing the natural language query input by the user and generating relevant responses. The natural language generation and understanding module can support context tracking and reasoning capabilities, automatically distinguish and understand the details and levels in the product inquiry, and maintain interaction by answering or asking questions; it can also support real-time dialogue response, and maintain consistent tone, expression style and logical coherence in the simultaneous interaction of large-scale users. The multimodal perception and sentiment analysis module can perceive the user's emotions by combining multimodal data such as text, images, videos and audio, and analyze the user's attitude in the interaction process in real time. For example, understand the user's satisfaction through facial expression analysis, or perceive the user's current emotions through voice tone, and adjust the digital human's tone, posture and response style accordingly. The multimodal perception and sentiment analysis module can support active acquisition and perception of user expressions, tones, and usage environments, and adjust the dialogue and feedback methods with users in real time, thereby optimizing the interactive experience. The multi-task execution management module can be responsible for handling the dialogue between multiple users (i.e., classified user groups) and digital humans at the same time, dynamically adjust the interactive tasks of each target interactive data according to its urgency and importance, and allocate interaction and feedback priorities in real time to ensure the naturalness and stability of digital humans in multiple interactive sessions. When multiple users ask complex questions, the multi-task execution management module can use large models to support segmentation and multi-threaded parallel operations, so that digital humans have multiple "minds" to simultaneously respond to multiple task scenarios. For multi-person real-time interactive scenarios involving product sales, the recommendation system and cross-selling module can use real-time data and context understanding technology to accurately recommend products that are highly consistent with user queries and needs, and perform related recommendations and cross-selling. For example, by analyzing the user's preferences for a certain type of product in real time, intelligently adjust the details of the product they care about, and recommend other similar or derivative products that may meet their expectations. The recommendation system and cross-selling module can support personalized push, and tailor exclusive recommendations for users based on their consumption habits, historical records and purchasing needs, and accompany them with sales suggestions to enhance the user experience.

[0059] It can be seen that the real-time interactive system based on digital humans can achieve rapid response and real-time interaction for batch users by using digital human technology based on user groups, thereby improving the interaction efficiency in multi-user concurrent real-time interaction scenarios, and improving the system response efficiency as a whole, thereby improving the interaction effect in multi-user concurrent real-time interaction scenarios, and further improving the interactive experience.

[0060] The disclosed embodiment obtains the real-time interaction data of each user in the current multi-user real-time interaction scenario, classifies the users according to the real-time interaction data, obtains multiple classified user groups, and screens the current classified user group from the classified user groups to determine the target interaction data of the current classified user group, thereby generating digital human interaction data according to the target interaction data, so that the digital human can interact with the current classified user group in real time according to the digital human interaction data, solves the problem of unsatisfactory real-time interaction effect in the existing multi-user concurrent real-time interaction scenario, can realize real-time interaction of batch users, thereby improving the interaction effect in the multi-user concurrent real-time interaction scenario, and further improving the interaction experience.

[0061] In one example, Figure 2 It is a flow chart of another real-time interaction method based on digital humans provided by an embodiment of the present disclosure. The embodiment of the present disclosure has been optimized and improved on the basis of the technical solutions of the above embodiments, and provides a variety of specific optional implementation methods for screening the current classified user group and generating digital human interaction data according to the target interaction data, and provides improvement methods for other functions of multi-person real-time interaction scenarios.

[0062] like Figure 2 A real-time interaction method based on a digital human is shown, comprising:

[0063] S210: Acquire real-time interaction data of each user in the current multi-user real-time interaction scenario.

[0064] S220: Classify the users according to the real-time interaction data to obtain multiple classified user groups.

[0065] S230: Filter the current classified user group from among the classified user groups.

[0066] Optionally, step S230 may include the following operations:

[0067] S231a, determining the number of users in each of the classified user groups.

[0068] The number of users is the number of users included in the classified user group.

[0069] S232a. Sort the classified user groups in descending order according to the number of users.

[0070] S233a: Use the classified user group with the largest number of users as the current classified user group.

[0071] Optionally, when the real-time interactive system based on digital humans selects the current classified user group from each classified user group, the classified user group with the largest number of users can be selected from each classified user group in turn based on the number of users as the current classified user group. The advantage of this setting is that batch user interaction based on the classified user group with the largest number of users can ensure that real-time and parallel interactive services are provided to as many users as possible at the same time, thereby improving the efficiency of user interaction.

[0072] S231b: Determine a group label for each of the classified user groups according to the user association data of the user.

[0073] Among them, user-related data may be relevant data that can reflect user attributes, such as, but not limited to, user tags, age, occupation, and geographic location. Group tags may be tags for classified user groups, used to identify the characteristics of each classified user group. Exemplarily, group tags may include, but are not limited to, new user groups, old user groups, user groups of different levels, groups that follow different products, and user groups of different activity levels. The embodiments of the present disclosure do not limit the specific tag type and tag content of group tags.

[0074] S232b: Calculate the weight of the group label of each of the classified user groups, and sort the classified user groups in descending order of the weight of the group label.

[0075] The weight of the group label can reflect the importance of the corresponding classified user group. It can be understood that the greater the weight of the group label, the more important the classified user group is; on the contrary, the smaller the weight of the group label, the less important the classified user group is.

[0076] It is understandable that the method of calculating the weight of the group label of the classified user group may be different depending on the type of the current multi-user real-time interaction scenario. For example, in a live broadcast scenario, the weight of the group label of the classified user group can be calculated based on the user's willingness to buy and the user's willingness to interact as reference factors. For example, the stronger the purchase intention reflected by the online user, and the more frequent the historical interaction data of the online user, the greater the weight value of the group label of the corresponding classified user group. In an online learning scenario, the weight of the group label of the classified user group can be calculated based on key knowledge points and user labels as reference factors. For example, the more relevant the questions asked by online users are to key knowledge points, and the more frequent the historical interaction data of online users, the greater the weight value of the group label of the corresponding classified user group.

[0077] S233b: taking the classified user group with the largest weight of the group label as the current classified user group.

[0078] Optionally, when the real-time interactive system based on digital humans selects the current classified user group from each classified user group, it can also calculate the corresponding weight of the group label of each classified user group based on the user's importance, and sort each classified user group in descending order of the weight of the group label, and then select the classified user group with the largest group label weight from each classified user group in turn as the current classified user group. The advantage of this setting is that batch user interaction is performed based on the classified user group with the largest weight of the user's corresponding group label, which can ensure that real-time and parallel interactive services are provided to as many and more important users as possible at the same time, thereby improving user stickiness and further improving the interaction effect.

[0079] It should be noted that steps S231a-S233a and steps S231b-S233b may be in a parallel relationship, and one may choose to execute steps S231a-S233a or steps S231b-S233b.

[0080] In an optional embodiment of the present disclosure, after taking the classified user group with the largest number of users as the current classified user group, it may also include: when it is determined that the target interaction data of the current classified user group completes the interaction, selecting the classified user group with the largest number of users from the remaining classified user groups in turn according to the sorting results of the classified user groups as the target replacement classified user group; and updating the current classified user group according to the target replacement classified user group.

[0081] The remaining classified user groups may be other classified user groups except the current classified user group. The target replacement classified user group may be a classified user group that can replace the current classified user group and become a new current classified user group.

[0082] Correspondingly, if it is determined that all target interaction data of the current classified user group have been processed, the number of users included in the remaining classified user groups can be updated in real time and dynamically, and then the remaining classified user groups can be sorted in order from most to least according to the number of users, and then the classified user groups with the largest number of users are selected from the remaining classified user groups in turn as the target replacement classified user groups, and the current classified user groups are updated according to the target replacement classified user groups to ensure that the number of users included in the current classified user group is always the maximum value of the current number of users. The advantage of this setting is that it can always ensure that the current multi-user real-time interaction scenario can always provide real-time and parallel interaction services to as many users as possible at the same time, thereby improving the efficiency of user interaction.

[0083] In an optional embodiment of the present disclosure, after taking the classified user group with the largest number of users as the current classified user group, the method may also include: when it is determined that the target interaction data of the current classified user group has completed the interaction, calculating the similarity between the remaining classified user groups and the current classified user group; selecting the remaining classified user groups with the largest similarity in turn as target replacement classified user groups according to the similarity between the remaining classified user groups and the current classified user group; and updating the current classified user group according to the target replacement classified user group.

[0084] Correspondingly, if it is determined that all target interaction data of the current classified user group have been processed, the users included in the remaining classified user groups can be first dynamically updated in real time, and then the similarity between the remaining classified user groups and the current classified user group can be calculated in real time, and the remaining classified user groups with the greatest similarity can be selected in turn according to the similarity between the remaining classified user groups and the current classified user groups as the target replacement classified user groups, so as to update the current classified user group according to the target replacement classified user group to ensure that the updated current classified user group is most similar to the current classified user group before the update.

[0085] The above technical solution can avoid the digital human from interacting with users in a jumpy manner by performing real-time interaction between the most similar classified user groups, thereby ensuring the smoothness and rationality of the current interactive operation and further improving the quality of user interaction.

[0086] In an optional embodiment of the present disclosure, the calculating the similarity between the remaining classified user groups and the current classified user group may include: obtaining real-time group interaction data between the remaining classified user groups and the current classified user group; and calculating the group similarity between the remaining classified user groups and the current classified user group based on the real-time group interaction data.

[0087] The group real-time interaction data may be all real-time interaction data of each user included in the classified user group. Exemplarily, the group real-time interaction data may include but is not limited to the attribute characteristics, main interaction data content, and main interaction action types of users in the classified user group. The group similarity may be the similarity calculated from the perspective of the classified user group.

[0088] Optionally, when calculating the similarity between the remaining classified user groups and the current classified user group, the group real-time interaction data of the remaining classified user groups and the group real-time interaction data of the current classified user group can be obtained respectively. Further, the similarity between the group real-time interaction data of each remaining classified user group and the group real-time interaction data of the current classified user group is calculated to obtain the group similarity, and the group similarity is used as the similarity between the remaining classified user group and the current classified user group. The advantage of this setting is that the similarity between each remaining classified user group and the current classified user group can be calculated from the overall perspective of the classified user group, thereby calculating the overall similarity between the remaining classified user group and the current classified user group.

[0089] In an optional embodiment of the present disclosure, the calculating the similarity between the remaining classified user groups and the current classified user group may include: obtaining real-time group interaction data between the remaining classified user groups and the current classified user group; determining the current group interaction problems between the remaining classified user groups and the current classified user group based on the real-time group interaction data; and calculating the interaction problem similarity between the current group interaction problems of the remaining classified user groups and the current classified user group.

[0090] The current group interaction problem may be an interaction problem of the classified user group that is currently pending or has been processed. The interaction problem similarity may be the similarity between the current group interaction problem of the remaining classified user groups and the current group interaction problem of the current classified user group.

[0091] Optionally, when calculating the similarity between the remaining classified user groups and the current classified user group, the group real-time interaction data of the remaining classified user groups and the group real-time interaction data of the current classified user group can be obtained respectively. Further, the current group interaction problems of the remaining classified user groups and the current classified user group can be determined respectively from the group real-time interaction data of each classified user group. It can be understood that, for the remaining classified user groups, the current group interaction problem can be the interaction problem currently to be processed by the remaining classified user groups; for the current classified user group, the current group interaction problem can be the interaction problem that has been processed by the current classified user group. Accordingly, the similarity of the interaction problem between the current group interaction problem of the remaining classified user groups and the current classified user group can be calculated, and the interaction problem similarity can be used as the similarity between the remaining classified user groups and the current classified user groups. The advantage of such a setting is that according to the similarity between the current group interaction problems of each user group, the similarity between the interaction data between the remaining classified user groups and the current classified user groups can be calculated, thereby ensuring the fluency and rationality of the interaction process of the real-time interaction system based on digital people.

[0092] In an optional embodiment of the present disclosure, the target interaction data includes user question data, and the number of the user question data is multiple.

[0093] Optionally, the user question data may be question data in text form, such as user comment data and barrage data in a live broadcast room, or may be question data in voice form, such as knowledge point questions in voice form asked by users to digital humans in online learning scenarios. The embodiments of the present disclosure do not limit the specific data type of the user question data.

[0094] Accordingly, the above method may further include the following operations:

[0095] S240: Classify and process the multiple user question data of the current classified user group to obtain classified question data.

[0096] The classified question data may be question data of each type obtained after classifying multiple user question data of the current classified user group.

[0097] It is understandable that since each classified user group may include multiple users, each classified user group may also include multiple user question data. In order to better interact with the current classified user group in real time, the multiple user question data of the current classified user group can be classified and processed to obtain multiple types of classified question data. It is understandable that the classified question data of the current classified user group can be question data of different vertical categories, such as question data about shipment, question data about after-sales, and question data about pre-sales. The classified question data of the current classified user group can also be different detailed question data of the same vertical category, such as question data about shipment time, question data about shipment location, and question data about shipment time appointment.

[0098] S250: Filter the current classified question data from the classified question data.

[0099] The current classified question data may be classified question data that needs to be processed currently for the current classified user group.

[0100] When there are multiple classified question data for the current classified user group, one of the classified question data can be selected from the multiple classified question data of the previous classified user group as the current classified question data for processing.

[0101] In an optional embodiment of the present disclosure, filtering the current classified question data from each of the classified question data may include: sorting each of the classified question data according to a priority sorting reference factor; wherein the priority sorting reference factor includes at least one of the number of user questions, the frequency of question occurrence, and the user's urgency; and filtering the current classified question data according to the sorting result of the classified question data.

[0102] The priority sorting reference factor may be a sorting factor referenced when sorting the classified question data.

[0103] Optionally, when filtering the current classified question data from each classified question data, each classified question data may be first sorted according to at least one priority sorting reference factor among the number of user questions, the frequency of question occurrence, and the user's urgency. Optionally, each classified question data may be sorted directly according to a single-dimensional priority sorting reference factor, such as sorting each classified question data according to the number of user questions. Alternatively, each classified question data may also be sorted according to a multi-dimensional priority sorting reference factor, such as setting weights corresponding to the number of user questions, the frequency of question occurrence, and the user's urgency, and comprehensively calculating the score value of each classified question data according to the number of user questions, the frequency of question occurrence, the user's urgency, and the weights corresponding to each priority sorting reference factor, thereby sorting the classified question data according to the score value of each classified question data.

[0104] The above technical solution can ensure that the classified question data with higher importance are processed first for the current classified user group by sorting the classified question data according to multiple types of priority sorting reference factors, thereby improving the stickiness of the current classified user group and further improving the interaction quality and interaction effect.

[0105] S260: Generate answer data for the current classified question data as the digital human interaction data.

[0106] Correspondingly, after determining the current classified question data of the current classified user group, the natural language generation and understanding module can be called to process the current classified question data, process the current classified question data and generate relevant answer data as digital human interaction data.

[0107] It is understandable that after the current classified question data is processed, the current classified question data can be further screened from the remaining classified question data of the current classified user group for processing until all classified question data of the current classified user group are processed. At the same time, since the users included in the current classified user group may be updated in real time, the classified question data corresponding to the current classified user group can also be updated in real time.

[0108] The above technical solution can realize batch interaction problem processing of the current classified user group by classifying and processing multiple user question data of the current classified user group, thereby improving the real-time interaction efficiency and interaction quality of the current classified user group, and further improving the interaction experience.

[0109] S270: Generate digital human interaction data according to the target interaction data, so that the digital human can interact with the current classified user group in real time according to the digital human interaction data.

[0110] In an optional embodiment of the present disclosure, the above method may also include: determining the interface attributes of the current real-time interaction interface of the current multi-user real-time interaction scene; wherein the interface attributes include the current layout format and / or the current digital human image; and dynamically adjusting the interface attributes of the current real-time interaction interface in real time according to the current classified user group.

[0111] The current real-time interaction interface may be the interface currently displayed in the current multi-user real-time interaction scene. The interface attribute may be the relevant attribute of the current real-time interaction interface. The layout format may be the layout format of the current real-time interaction interface. The digital human image may be the image of the digital human in the current real-time interaction interface.

[0112] In the current multi-user real-time interaction scenario, users can interact with digital humans in real time based on the current real-time interaction interface. In order to further improve the interaction effect, the layout format of the current real-time interaction interface and / or interface attributes such as the digital human image can be dynamically adjusted in real time based on the relevant data of the current classified user group, so that the adjusted current real-time interaction interface is more suitable for the current interaction atmosphere requirements of the current classified user group.

[0113] In an optional embodiment of the present disclosure, the real-time dynamic adjustment of the interface attributes of the current real-time interaction interface according to the current classified user group may include: real-time dynamic adjustment of the interface attributes of the current real-time interaction interface according to the target interaction data of the current classified user group; or real-time dynamic adjustment of the interface attributes of the current real-time interaction interface according to the group label of the current classified user group.

[0114] Optionally, the interface attributes of the current real-time interaction interface can be adjusted dynamically in real time based on the target interaction data of the current classified user group or group labels and other group-related data. For example, the layout format of the current real-time interaction interface can be adjusted based on the target interaction data of the current classified user group. Specifically, the layout of the product introduction area can be adjusted based on the type of the current classified question data; the digital human image of the current real-time interaction interface can be adjusted based on the group labels of the current classified user group. For example, for the current classified user group of young users, a young and energetic virtual image can be used as the current digital human image.

[0115] The above technical solution can improve the adaptability of the current real-time interaction interface to the current classified user group by dynamically adjusting the interface attributes of the current real-time interaction interface in real time according to the target interaction data of the current classified user group or group labels and other group-related data.

[0116] In an optional embodiment of the present disclosure, the above method may also include: filtering the target real-time interaction data of the target user from the real-time interaction data of each of the users; generating independent interaction data based on the data content of the target real-time interaction data; and sending the independent interaction data to the target user through an independent channel for real-time interaction.

[0117] The target user may be a user with independent interaction needs. The target real-time interaction data may be the real-time interaction data of the target user in the current multi-user real-time interaction scenario. The independent interaction data may be the interaction feedback data independently generated by the real-time interaction system based on the digital human for the target user. The independent channel may be a channel that can send the independent interaction data to the target user separately.

[0118] Considering that some users have the need for additional individual interactions, for such target users, the real-time interaction system based on digital humans can filter out the corresponding target real-time interaction data from various real-time interaction data, and generate matching independent interaction data for the specific data content included in each target real-time interaction data, so as to send the independent interaction data to the target user through an independent channel for real-time interaction. For example, in a live broadcast scenario, for a user's inquiry about the logistics of an ordered product, the real-time logistics data of the corresponding product can be fed back to the user through a private message.

[0119] The above technical solution can meet the personalized interaction needs of independent users by performing independent interactions on users independent of the group, thereby improving the response speed of the real-time interaction system based on digital humans.

[0120] After obtaining the real-time interaction data of each user in the current multi-user real-time interaction scene, the disclosed embodiment classifies the users according to the real-time interaction data to obtain a plurality of classified user groups. Further, by screening the current classified user group from each classified user group according to a plurality of optional methods, the target interaction data of the current classified user group is classified at the same time, so as to generate digital human interaction data according to the classification result, and finally the digital human interacts with the current classified user group in real time according to the digital human interaction data. During the interaction process, the current layout format of the current real-time interaction interface and / or the interface attributes such as the current digital human image can also be adjusted in real time and dynamically according to the target interaction data or group label of the current classified user group. For users who need independent interaction, their independent interaction data can also be automatically generated and sent to the corresponding target user through an independent channel for real-time interaction. It can be seen that the above-mentioned real-time interaction system based on digital human has a complete interaction function, which can improve the quality and effect of multi-person real-time interaction.

[0121] In one example, Figure 3is a schematic diagram of an interaction flow between functional modules in a real-time interactive system based on a digital human provided by an embodiment of the present disclosure, Figure 4 FIG. 1 is another interactive flow diagram of functional modules in a real-time interactive system based on a digital human provided by an embodiment of the present disclosure. In a specific example, a live broadcast scene is used as an example. Figure 3 and Figure 4 As shown, the real-time interactive system based on digital humans applied in live broadcast scenarios can be configured with a group interaction data collection module, a group analysis and opinion decision-making module, a personalized group dialogue generation module, a content adjustment and customized presentation module, a personal exclusive intelligent assistant module, and a concurrent management and response coordination module. Among them:

[0122] The group interaction data collection module can be used to monitor, capture and store the behavior data of users (hereinafter referred to as users) in real time during live broadcasts, including but not limited to user interaction activities such as comment content, likes, sharing and collections, and can also consider group behavior data such as user viewing time, access to product details and frequency of asking questions. This module is mainly responsible for providing the data foundation required by the system.

[0123] The implementation of the group interaction data collection module can take into account the efficient processing requirements in a high-concurrency user environment, adopt efficient data stream capture technology, and coordinate with distributed stream processing tools to orderly schedule and distribute data. This module can process different types of data such as interactive behavior logs and text comments, and transmit them to subsequent modules in a structured form.

[0124] Specifically, the group interaction data collection module can continuously monitor user behavior events during the live broadcast process by means of event monitoring, and monitor users' real-time interactions in various ways such as text comments, likes, forwarding, browsing behaviors, and sharing of goods. The monitored user behavior events can be classified according to different types. The main classification types may include but are not limited to text behaviors (such as comments) and non-text behaviors (such as likes, viewing time, and click sharing, etc.), so that different processing mechanisms can be selected based on different types of behaviors to provide more targeted data optimization and response. The group interaction data collection module can call the natural language processing model for data processing, such as part-of-speech tagging, text segmentation, etc. Then it is further passed to the subsequent submodules for detailed text content analysis according to the predefined logic. The group interaction data collection module can also use the sentiment analysis model in NLP to identify positive, neutral, and negative emotions for the user's comment information. This information will be used in subsequent steps to infer the user's emotional attitude towards the live interactive content as an important reference for deriving content adjustment. By analyzing the text comments, core keywords are extracted. For example, if a large number of users raise questions related to "power consumption" during a live broadcast, this keyword will be extracted at a high frequency and used as an object of focus for subsequent analysis. Non-textual interactive behaviors can be recorded as log information, including numerical data such as likes, forwarding, and page dwell time. By monitoring the behavior, the user's significant behavior during the live broadcast can be directly recorded, such as the total number of likes, the user's dwell time, etc. All analysis and processing of the group interaction data collection module will be stored as structured data, including but not limited to the record data of user behavior (such as the number of likes and the time when the comment was initiated, etc.), the emotional results of the comments, the structured information of keywords, etc. These structured data will be used as feedback and pushed to the subsequent group analysis and opinion decision-making module in real time through the distributed stream processing platform for further mining and utilization, so as to ensure that all data streams are transmitted quickly with low latency.

[0125] The group analysis and opinion decision module relies on distributed computing and group intelligence technology to analyze user group behavior data, including user group classification, similarity calculation between user groups, screening of current interaction data of user groups, sentiment analysis of group users, keyword extraction, topic identification, and opinion aggregation. Combined with large model processing, deep cognition and decision support based on the wishes of user groups are achieved. This module is responsible for integrating and processing the data of all users and providing a basis for subsequent personalized output.

[0126] The group analysis and opinion decision module is the "cognitive center" of the entire system. It analyzes and infers the data based on the large amount of user behavior and comment data received from the group interaction data collection module, realizes user group processing, classifies users of the same type into a user group, and finally outputs representative opinions and behavior suggestions for different groups of users. The core of this module is to use large models (such as language models, sentiment analysis models, and topic modeling, etc.) to intelligently process group data, automatically determine the main focus of the current group discussion, explore decision-making needs in group intelligence, and make data induction and reasoning in multiple dimensions.

[0127] The group analysis and opinion decision module can use a large model to perform sentiment analysis on the collected group comments and interactive information, extracting important emotional factors in the opinions and possible controversial points. It can also extract high-frequency keywords from the comment text and classify hot topics through topic modeling analysis. For example, combined with user comment information, the user's concerned topics such as performance and price are extracted from the interaction process as the basis for analysis and feedback. Based on the system's intelligent reasoning in group intelligence analysis, the above two results are synthesized using a large model to provide reference decision information and user priority speech lists for the subsequent personalized group dialogue generation module.

[0128] Specifically, the group analysis and opinion decision module can first receive the data pushed by the group interaction data collection module at the beginning. In order to improve the efficiency of data analysis, the original data can be preliminarily sorted and preprocessed, including cleaning noise, deleting meaningless messages, and completing data, and at the same time standardizing and formatting comment data and log data to make them have the basis for inputting intelligent models. For the sorted comment information, a large language model such as a sentiment classification model is started to determine the emotional tendency (positive, negative or neutral) of each comment. According to the results of sentiment analysis, all analyzed user comments are labeled with unified labels. These labels will point to positive emotions (excited, optimistic or supportive), negative emotions (complaints or dissatisfaction) and neutral comments, and provide sentiment classification basis for subsequent derivation. At the same time, the language processing tool can also be used to perform topic analysis and keyword extraction to identify the core theme of the text content and the main opinions of users represented by it. For example, the keywords that users talk about more are "price", "after-sales service" or "power consumption". This step automatically classifies the topics discussed by the group and marks them. This extraction can help find the common hot spots and long-tail issues that all users are concerned about during the interaction process. After completing the topic analysis, the group analysis and opinion decision module identified multiple high-frequency, key issues or topics discussed by major users, such as "performance evaluation", "gifts" or "cashback". Further, deep learning technology is used to synthesize information such as emotional tags and keywords to obtain a comprehensive opinion map. This analysis is not only based on the analysis of a single comment, but also attempts to associate group reactions of different dimensions through the "emotion-keyword matrix" and determine what kind of interactive feedback should be generated. For example, most users not only expressed concerns about the power consumption of a certain mobile phone, but also the emotions were mostly negative. Therefore, it was decided to strengthen the system to give professional explanations and highlight the interactive data of mobile phone energy efficiency management design. It is understandable that group interaction will generate consensus decisions on multiple different topics. The group analysis and opinion decision module can sort out these decisions from multiple analysis results and form widely recognized opinions and suggestions among multiple concurrent issues. The ultimate goal is to provide a sufficiently rich and logically based group wisdom conclusion to the personalized group dialogue generation module. Finally, an interaction plan is generated based on the group consensus judgment made. For example, users who are concerned about price issues are given feedback interaction data based on "discount-related offers", and users who focus on the appearance of the phone are given close-up scenes of the appearance, making the interactive content more targeted. The final derivation decision generated by this module will pass the summarized emotional decisions and topic consensus results to the next layer of personalized group dialogue generation module in real time through the API, so that the system can use these opinions to generate personalized speech and multi-faceted customized speech responses.

[0129] The personalized group dialogue generation module can use the large model to generate personalized interactive decisions and words for different user groups or single users in real time based on the output results of the group analysis and opinion decision module, support multiple content outputs in parallel, and meet the customized needs of different users at the same time. For example, it can push high-definition performance descriptions to users who are concerned about video playback effects, and push technical data descriptions to users who are concerned about power consumption.

[0130] The personalized group dialogue generation module generates customized response words for different group needs by deeply analyzing the group analysis data provided by the above modules (including sentiment analysis results of user groups, keyword hot topics and group opinions, etc.). The core of this module relies on a large-scale language generation model to generate real-time multi-version dialogues for different concerns according to the different requirements of group users. This generation can automatically balance various cross-requirements to ensure efficient response. In addition, the module can generate multiple path output solutions customized for different user groups, and these paths are derived from the upstream "emotional decision" and "group analysis results". In the implementation process, a specific large model is introduced to generate the next step of personalized natural language recommendation response through existing input in the existing context. Furthermore, the module also has the ability to dynamically optimize feedback, and continuously adjust the presentation mode and focus of the dialogue based on the interaction, so that the dialogue conforms to the continuously changing interactive scene.

[0131] Specifically, the personalized group dialogue generation module receives data from the group analysis and decision-making module, and can also obtain the user demand list and opinion generation results automatically generated by the system as the basis for the work of this module. This module can classify and divide the user group according to the behavioral characteristics and comment classification of the group users. For example, users can be divided into several categories such as "price-sensitive users", "technology enthusiasts" and "new product rush buyers" according to interest tags. Detailed user grouping is crucial for personalized content generation. The big data model in this module determines the group classification standard through behavior frequency and emotional weight. In order to analyze user behavior more deeply, a group portrait of each user group can also be established. The construction of these user portraits is based on the past viewing behavior, emotional analysis results, comment hotspots and user language habits of each user in the user group. For different user groups, the portrait attributes of demand segmentation can be automatically generated, such as viewing time, interaction frequency and importance, and specific priority objects can be marked and filtered. After collecting and building user portraits, the system calls a large-scale pre-trained language model to complete the generation of natural language, relying on the powerful language generation model capabilities to automatically generate the most personalized words for users. The personalized group dialogue generation module can generate corresponding answers based on the popular keyword combinations mentioned by users, or users can make multiple-choice inferences on the main controversial points of the live broadcast, such as giving detailed and simplified answers to different needs for product introduction content. Through the context understanding and language output capabilities of the big model, personalized recommendation words are generated according to the established user group preferences. This recommendation can be achieved by the big model autonomously adjusting the language tone to ensure that it conforms to the audience's style. For example, a relaxed, lively and humorous dialogue is generated for young people; while for professional technology enthusiasts, professional terms are used for in-depth interpretation. In addition to generating recommendation words, this module also further generates specific solution responses (such as promoting participation and instant discounts) through analysis, thereby actively guiding users to perform corresponding operations, similar to promotional calls or issuing discount suggestions. The personalized group dialogue generation module can also automatically output multiple versions of dialogue options, each of which is matched one by one with the previously assigned user group to ensure that the content conforms to the preferences, language habits and expectations of each group of users. This action is repeated many times, and multiple paths are generated for different groups. For example, for users who are interested in discount information, price-oriented words are generated; for users who are concerned about technical parameters, performance dialogues are generated. The personalized group dialogue generation module can also automatically formulate more response plans based on the actual needs of the user group analysis. The plan may include suggestions for solving after-sales problems, such as introducing freight subsidies, and further specific actions based on user needs for product features or live broadcast mechanisms, such as immediately guiding preferential orders and limited purchases, etc. Accordingly, the multiple versions of responses from "personalized speech" and "specific action plans" are merged and sorted through the background, and sent to multiple interactive channels of the live broadcast according to a certain priority.Different user groups have different language content feedback, and multiple speech paths meet the interaction needs of different user groups. In addition, this module also has self-feedback capabilities. After the actual content is pushed, it can monitor new interaction data and continuously adjust the subsequent dialogue generation strategy of the language generation model. Finally, the processed multiple personalized responses are packaged into structured returns after all merging and screening steps, and sent to the corresponding user groups, so that different users can receive recommendation results or promote them to interact with the next step of the live broadcast, such as purchasing or sharing, based on their own behavioral characteristics and preferences. Users get the greatest convenience and information satisfaction from multiple versions of customized content and interactive participation suggestions.

[0132] The content adjustment and customized presentation module is responsible for dynamically adjusting and optimizing the order, format and push channel of the main live content presentation based on the personalized generated results. The main screen can be used to prioritize content, highlight key product explanations and other different strategies to adjust different focus points among the audience groups and maximize the personalized interactive experience.

[0133] The content adjustment and customized presentation module receives corresponding data from the personalized group dialogue generation module and the group decision module, and combines the historical interaction records of the user group to infer the content type that they are most concerned about at present. Usually, this link is combined with behavioral analysis through machine learning models or weighted based on a certain emotional trajectory. For example: for users with positive emotions, it can be inferred that they are more willing to buy, and the product details page will be pushed first; for negative emotions, such as complaining about slow delivery of goods, remedial measures to solve the problem or reminders of precautions may be displayed. Through the interest analysis results, the system can build a content priority system. Each specific product and message fragment (such as discount promotions, specification display) will be given a certain weight to define certain content that is more attractive to specific users, and determine the playback time, length and display position of such content in the live video. The content adjustment and customized presentation module can refine the push content according to the user group. During the live interaction process, due to different attributes and preferences of user groups, the content adjustment and customized presentation module can target the interaction history of specific audiences, such as age group deviation, behavioral data, and focus direction, to ensure that different product display focuses are presented. The content adjustment and customized presentation module can also adjust the layout of personalized content according to the type of user group currently interacting and the focus recommendation position of the current scene / activity. For example, if a product has a strong interaction with the technical user group, the system will give priority to placing the content description in the center or high exposure position, while for the user group that pays attention to other types of discount details, it may adopt a more stable auxiliary zoning display method, and display it in a small window at the border of the non-important focus position. The content adjustment and customized presentation module can also adjust the main screen content playback priority, especially for the group with concentrated interactive interests. Modify the playback priority and coverage time point in time. For example, when introducing a new clothing product, the timing is suitable for the main screen scene to insert a more specific video or picture introduction clip, and turn the function explanation to the user focus. In the multi-area display mode, the system will use the secondary screen area (including the bullet screen area, the small product recommendation area, etc.) to adjust the user experience. Users can view the secondary screen product recommendation information while enjoying the high-quality main screen push during the live interaction. When it is found that the user group's interaction rate with certain advertisements or comments increases, the secondary screen content area will generate dynamic adjustments or focus on displaying "participatory activities". The content adjustment and customized presentation module can also have the function of intelligently and dynamically switching the main content. The real-time content switching logic determines when and at what time to make intelligent recommendations for which users. For example, when sellers interact intensely, a flash sale or limited-time promotional script can be automatically arranged for insertion. This task can trigger the exchange action after being automatically generated using NLP. In addition, during periods of high negative feedback, quick remedial displays corresponding to the scene can be inserted to prevent emotional fluctuations from reducing conversion rates. According to the inferred scheduling rules, personalized live content presentation can also be displayed for each user side. That is, although the live broadcast interfaces of different user terminals in the same live broadcast room have a unified audio stream, the displayed screen content may be different.Whether it is the priority of the main video screen or the dynamic allocation of recommended information, all final decisions reflect the viewing experience that is most beneficial to the user. For example, the same live broadcast room presents discount instructions for one type of user's main video, while showing product demonstration clips for another type. The display interface of the live broadcast can adjust the layout according to the strategy, such as the product page is also accompanied by the host's discount announcement, and real-time triggering calls push information, etc.

[0134] The personal intelligent assistant module can be a module independent of the group's main interaction process. When users directly query additional product information and communicate their individual needs through text or voice, such as product after-sales information query, payment process issues, etc., this intelligent assistant can generate timely and personalized responses for each user. Therefore, the personal intelligent assistant module can automatically identify, process and respond to users' independent needs, and instantly feedback personalized information query results, so that users' problems in the interaction process can be solved quickly and intelligently.

[0135] The concurrency management and response coordination module can realize the coordinated response and scheduling of multiple user requests, ensure the stable operation of the system under high concurrency, and avoid performance bottlenecks through load balancing and resource allocation. It ensures the real-time connection of data flow and processing flow of various modules, reduces delays and improves response speed.

[0136] In a specific example, during an e-commerce live broadcast, more than 1,000 users were watching a live commentary of a newly released smartphone. Some users were concerned about the screen brightness and sound experience of the phone, while others had questions about battery performance and power consumption. In traditional live broadcast systems, anchors usually need to receive and view comments, and then select a question from the comments to respond one by one. This can neither accurately and comprehensively capture the interaction focus of all users, nor can it cope with frequent and diverse problem scenarios, which can easily cause dissatisfaction among some users. In an e-commerce live broadcast system based on group intelligent interaction, the behavioral comments of all viewers are collected in real time through the group interaction data collection module. Under the action of the group analysis and opinion decision module, the comments, likes, screen typing and other information in the live broadcast room are aggregated and processed, and at the same time, through sentiment analysis and behavior preference recognition, opinions and judgments on high-frequency user questions are obtained. Then, with the model support of the personalized group dialogue generation module, based on key issues (such as screen experience and power consumption), the system generates two main different speech techniques to make customized content recommendations for two types of user groups. On the main screen, based on the dynamic regulation of content adjustment and customized presentation modules, the system shows the optical technology advantages behind the mobile phone's high-definition screen technology to users who focus on screen experience at 30-second intervals. At the same time, a short explanatory video for battery-conscious users is automatically played in the side window, explaining the low power consumption and AI scheduling mechanism of the mobile phone. The main screen can also explain user groups' questions about screen brightness and sound experience through digital humans.

[0137] In a specific example, suppose that while Ms. Zhang is watching a fashion live broadcast, she needs to quickly check the shipping policy, but the shipping policy has not been mentioned in the host's introduction. She sent a voice query about the delivery time to the system's intelligent assistant module, and the request was intercepted and processed by the personal intelligent assistant module. The personal intelligent assistant module uses the internal FAQ (Frequently Asked Questions) database and cooperates with the query matching function to directly feedback the approximate time when her current city can be shipped. The personal intelligent assistant works independently, allowing Ms. Zhang to get a quick answer. At this time, other viewers in the main live broadcast scene are still continuing to discuss the color and material of the clothes in a group.

[0138] In a specific example, in a certain scenario, the sales of live products soared to 10,000 people watching online. A large number of real-time barrage questions, user likes, and constant page refreshes brought high concurrency. For this concurrent scenario, in the background of the system, the concurrency management and response coordination module performs real-time concurrent load monitoring. On the one hand, this module ensures that the group interaction data collection module can normally process a large amount of behavioral data. At the same time, it also greatly compresses the response delay of each message in calling "personalized dialogue generation" and "intelligent assistant service". Specific tuning includes user requests being broken up by load balancing, sharding through distributed queues and then sent to each computing node, and calling fast cache to adjust the query flow, avoiding the main server processor or memory bottleneck under the distributed cloud architecture, and ensuring the efficient completion of individual responses. In the end, the questions raised by the user are processed and fed back with almost zero perceived delay.

[0139] In the above technical solution, user input is recorded in the group interaction data collection module, and the group analysis and opinion decision-making module analyzes and classifies the massive information of multiple users in real time, and feedback is given based on different key information. A personalized group dialogue generation module is used to create diversified solutions, and the content adjustment and customized presentation module is used to output them. At the same time, additional questions required by individual users can be answered individually through the personal intelligent assistant module. During the entire response process, the concurrency management and coordination module always adjusts and load manages each interaction link to ensure smooth data flow between different individuals and the whole, which can greatly improve the quality of interactive experience of concurrent live users.

[0140] In the technical solution disclosed herein, the collection, storage, use, processing, transmission, provision and disclosure of user personal information (such as real-time interaction data, voice information, user attribute data, etc.) are in compliance with the relevant laws and regulations and do not violate public order and good morals.

[0141] It should be noted that any arrangement and combination of the technical features in the above embodiments also falls within the protection scope of the present disclosure.

[0142] In one example, Figure 5 This is a structural diagram of a real-time interactive device based on a digital human provided by an embodiment of the present disclosure. The embodiment of the present disclosure is applicable to the situation where a digital human interacts with a batch of similar users in a multi-person real-time interactive scenario. The device is implemented by software and / or hardware and is specifically configured in an electronic device. The electronic device can be a terminal device or a server device, as long as it can execute the real-time interactive method based on a digital human. The embodiment of the present disclosure does not limit the specific device type of the electronic device.

[0143] like Figure 5The real-time interaction device 500 based on digital human shown in the figure comprises: a real-time interaction data acquisition module 510, a classified user group acquisition module 520, a current classified user group screening module 530, a target interaction data determination module 540 and a group real-time interaction module 550.

[0144] A real-time interaction data acquisition module 510 is used to acquire the real-time interaction data of each user in the current multi-user real-time interaction scene;

[0145] A classified user group acquisition module 520 is used to classify the users according to the real-time interaction data to obtain multiple classified user groups;

[0146] The current classified user group screening module 530 is used to screen the current classified user group from the classified user groups;

[0147] A target interaction data determination module 540 is used to determine the target interaction data of the currently classified user group;

[0148] The group real-time interaction module 550 is used to generate digital human interaction data according to the target interaction data, so that the digital human can interact with the current classified user group in real time according to the digital human interaction data.

[0149] The disclosed embodiment obtains the real-time interaction data of each user in the current multi-user real-time interaction scenario, classifies the users according to the real-time interaction data, obtains multiple classified user groups, and screens the current classified user group from the classified user groups to determine the target interaction data of the current classified user group, thereby generating digital human interaction data according to the target interaction data, so that the digital human can interact with the current classified user group in real time according to the digital human interaction data, solves the problem of unsatisfactory real-time interaction effect in the existing multi-user concurrent real-time interaction scenario, can realize real-time interaction of batch users, thereby improving the interaction effect in the multi-user concurrent real-time interaction scenario, and further improving the interaction experience.

[0150] Optionally, the current classified user group screening module 530 is further used to: determine the number of users in each of the classified user groups; sort the classified user groups in descending order according to the number of users; and use the classified user group with the largest number of users as the current classified user group.

[0151] Optionally, the current classified user group screening module 530 is also used to: determine a group label for each of the classified user groups based on the user association data of the user; calculate the weight of the group label of each of the classified user groups; sort the classified user groups in descending order of the weight of the group label; and take the classified user group with the largest weight of the group label as the current classified user group.

[0152] Optionally, the above-mentioned device also includes a current classified user group update module, which is used to: when it is determined that the target interaction data of the current classified user group has completed the interaction, select the classified user group with the largest number of users from the remaining classified user groups in turn according to the sorting results of the classified user groups as the target replacement classified user group; update the current classified user group according to the target replacement classified user group.

[0153] Optionally, the above-mentioned device also includes a current classified user group update module, which is used to: when it is determined that the target interaction data of the current classified user group has completed the interaction, calculate the similarity between the remaining classified user groups and the current classified user group; according to the similarity between the remaining classified user groups and the current classified user group, the remaining classified user groups with the greatest similarity are selected in turn as target replacement classified user groups; and update the current classified user group according to the target replacement classified user group.

[0154] Optionally, the current classified user group updating module is further used to: obtain real-time group interaction data between the remaining classified user groups and the current classified user group; and calculate the group similarity between the remaining classified user groups and the current classified user group based on the real-time group interaction data.

[0155] Optionally, the current classified user group update module is also used to: obtain real-time group interaction data between the remaining classified user groups and the current classified user group; determine the current group interaction problems between the remaining classified user groups and the current classified user group based on the real-time group interaction data; and calculate the similarity of interaction problems between the remaining classified user groups and the current group interaction problems of the current classified user group.

[0156] Optionally, the target interaction data includes user question data, and the number of the user question data is multiple; the group real-time interaction module 550 is also used to: classify the multiple user question data of the current classified user group to obtain classified question data; filter the current classified question data from each of the classified question data; generate answer data for the current classified question data as the digital human interaction data.

[0157] Optionally, the group real-time interaction module 550 is also used to: sort each of the classified question data according to priority sorting reference factors; wherein the priority sorting reference factors include at least one of the number of user questions, the frequency of question occurrences, and the user's urgency; and filter the current classified question data according to the sorting results of the classified question data.

[0158] Optionally, the above-mentioned device also includes an interface attribute dynamic adjustment module, which is used to: determine the interface attributes of the current real-time interaction interface of the current multi-user real-time interaction scene; wherein the interface attributes include the current layout format and / or the current digital human image; and dynamically adjust the interface attributes of the current real-time interaction interface in real time according to the current classified user group.

[0159] Optionally, the interface attribute dynamic adjustment module is also used to: dynamically adjust the interface attributes of the current real-time interaction interface in real time according to the target interaction data of the current classified user group; or dynamically adjust the interface attributes of the current real-time interaction interface in real time according to the group label of the current classified user group.

[0160] Optionally, the above-mentioned device also includes an independent interaction module, which is used to: filter the target real-time interaction data of the target user from the real-time interaction data of each of the users; generate independent interaction data according to the data content of the target real-time interaction data; and send the independent interaction data to the target user through an independent channel for real-time interaction.

[0161] The above-mentioned real-time interaction device based on digital human can execute the real-time interaction method based on digital human provided by any embodiment of the present disclosure, and has the corresponding functional modules and beneficial effects of the execution method. For technical details not described in detail in this embodiment, please refer to the real-time interaction method based on digital human provided by any embodiment of the present disclosure.

[0162] Since the above-mentioned real-time interactive device based on digital human is a device that can execute the real-time interactive method based on digital human in the embodiment of the present disclosure, based on the real-time interactive method based on digital human in the embodiment of the present disclosure, the technical personnel in the field can understand the specific implementation of the real-time interactive device based on digital human in the embodiment of the present disclosure and its various variations, so how the real-time interactive device based on digital human implements the real-time interactive method based on digital human in the embodiment of the present disclosure is not introduced in detail here. As long as the technical personnel in the field implement the device adopted by the real-time interactive method based on digital human in the embodiment of the present disclosure, it belongs to the scope of protection to be provided by the present disclosure.

[0163] In one example, the present disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0164] Figure 6 A schematic block diagram of an example electronic device 600 that can be used to implement an embodiment of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or required herein.

[0165] like Figure 6 As shown, the device 600 includes a computing unit 601, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 602 or a computer program loaded from a storage unit 608 into a random access memory (RAM) 603. In the RAM 603, various programs and data required for the operation of the device 600 can also be stored. The computing unit 601, the ROM 602, and the RAM 603 are connected to each other via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0166] A number of components in the device 600 are connected to the I / O interface 605, including: an input unit 606, such as a keyboard, a mouse, etc.; an output unit 607, such as various types of displays, speakers, etc.; a storage unit 608, such as a disk, an optical disk, etc.; and a communication unit 609, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 609 allows the device 600 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0167] The computing unit 601 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 601 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The computing unit 601 performs the various methods and processes described above, such as a real-time interaction method based on a digital human.

[0168] Optional real-time interaction methods based on digital humans may include:

[0169] Obtain the real-time interaction data of each user in the current multi-user real-time interaction scenario;

[0170] Classifying the users according to the real-time interaction data to obtain multiple classified user groups;

[0171] Filter the current classified user group from each of the classified user groups;

[0172] Determining target interaction data of the currently classified user group;

[0173] Digital human interaction data is generated according to the target interaction data, so that the digital human can interact with the current classified user group in real time according to the digital human interaction data.

[0174] For example, in some embodiments, the real-time interaction method based on digital human can be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as storage unit 608. In some embodiments, part or all of the computer program can be loaded and / or installed on the device 600 via ROM 602 and / or communication unit 609. When the computer program is loaded into RAM 603 and executed by computing unit 601, one or more steps of the real-time interaction method based on digital human described above can be performed. Alternatively, in other embodiments, computing unit 601 can be configured to perform the real-time interaction method based on digital human by any other appropriate means (for example, by means of firmware).

[0175] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0176] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0177] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0178] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0179] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0180] A computer system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The relationship between the client and the server is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services. The server can also be a server of a distributed system, or a server combined with a blockchain.

[0181] The disclosed embodiment obtains the real-time interaction data of each user in the current multi-user real-time interaction scenario, classifies the users according to the real-time interaction data, obtains multiple classified user groups, and screens the current classified user group from the classified user groups to determine the target interaction data of the current classified user group, thereby generating digital human interaction data according to the target interaction data, so that the digital human can interact with the current classified user group in real time according to the digital human interaction data, solves the problem of unsatisfactory real-time interaction effect in the existing multi-user concurrent real-time interaction scenario, can realize real-time interaction of batch users, thereby improving the interaction effect in the multi-user concurrent real-time interaction scenario, and further improving the interaction experience.

[0182] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.

[0183] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A real-time interaction method based on digital human, comprising: Obtain the real-time interaction data of each user in the current multi-user real-time interaction scenario; Classifying the users according to the real-time interaction data to obtain multiple classified user groups; Filter the current classified user group from each of the classified user groups; Determining target interaction data of the currently classified user group; Digital human interaction data is generated according to the target interaction data, so that the digital human can interact with the current classified user group in real time according to the digital human interaction data.

2. The method according to claim 1, wherein: The step of screening the current classified user group from each of the classified user groups includes: Determining the number of users in each of the classified user groups; Sorting each of the classified user groups in descending order according to the number of users; The classified user group with the largest number of users is used as the current classified user group.

3. The method according to claim 1, wherein: The step of screening the current classified user group from each of the classified user groups includes: Determining a group label for each of the classified user groups according to the user association data of the user; Calculating the weight of the group label of each of the classified user groups; Sorting the classified user groups in descending order according to the weight of the group labels; The classified user group with the largest weight of the group label is used as the current classified user group.

4. The method according to claim 2 or 3, further comprising: When it is determined that the target interaction data of the current classified user group has completed the interaction, the classified user group with the largest number of users is selected from the remaining classified user groups in sequence according to the sorting result of the classified user groups as the target replacement classified user group; The current classified user group is updated according to the target replacement classified user group.

5. The method according to claim 2 or 3, further comprising: When it is determined that the target interaction data of the current classified user group has completed the interaction, calculating the similarity between the remaining classified user groups and the current classified user group; According to the similarity between the remaining classified user groups and the current classified user group, the remaining classified user groups with the greatest similarity are selected in turn as target replacement classified user groups; The current classified user group is updated according to the target replacement classified user group.

6. The method according to claim 5, wherein: The calculating the similarity between the remaining classified user groups and the current classified user group includes: Acquire real-time group interaction data between the remaining classified user groups and the current classified user group; The group similarity between the remaining classified user groups and the current classified user group is calculated according to the group real-time interaction data.

7. The method according to claim 5, wherein: The calculating the similarity between the remaining classified user groups and the current classified user group includes: Acquire real-time group interaction data between the remaining classified user groups and the current classified user group; Determine the current group interaction problem between the remaining classified user groups and the current classified user group according to the real-time group interaction data; The similarity of the interaction problems between the remaining classified user groups and the current group interaction problems of the current classified user group is calculated.

8. The method according to claim 7, wherein: The target interaction data includes user question data, and the number of the user question data is multiple; generating digital human interaction data according to the target interaction data includes: Classify and process the multiple user question data of the current classified user group to obtain classified question data; Filtering current classified question data from each of the classified question data; Generate answer data for the current classified question data as the digital human interaction data.

9. The method according to claim 8, wherein: Filtering the current classified question data from each of the classified question data includes: Sorting each of the classified question data according to a priority sorting reference factor; wherein the priority sorting reference factor includes at least one of the number of user questions, the frequency of question occurrence, and the user's urgency; The current classified question data is filtered according to the sorting result of the classified question data.

10. The method according to claim 1, further comprising: Determining interface attributes of the current real-time interaction interface of the current multi-user real-time interaction scene; wherein the interface attributes include a layout format and / or a digital human image; The interface attributes of the current real-time interactive interface are dynamically adjusted in real time according to the current classified user group.

11. The method according to claim 10, wherein: The real-time dynamic adjustment of the interface attributes of the current real-time interactive interface according to the current classified user group includes: Dynamically adjusting the interface attributes of the current real-time interaction interface in real time according to the target interaction data of the current classified user group; or The interface attributes of the current real-time interactive interface are dynamically adjusted in real time according to the group label of the current classified user group.

12. The method according to claim 1, further comprising: Filtering target real-time interaction data of a target user from the real-time interaction data of each of the users; Generate independent interaction data according to the data content of the target real-time interaction data; The independent interaction data is sent to the target user through an independent channel for real-time interaction.

13. A real-time interactive device based on digital human, comprising: A real-time interaction data acquisition module is used to acquire the real-time interaction data of each user in the current multi-user real-time interaction scenario; A classified user group acquisition module is used to classify the users according to the real-time interaction data to obtain multiple classified user groups; A current classified user group screening module, used for screening the current classified user group from each of the classified user groups; A target interaction data determination module, used to determine the target interaction data of the currently classified user group; The group real-time interaction module is used to generate digital human interaction data according to the target interaction data, so that the digital human can interact with the current classified user group in real time according to the digital human interaction data.

14. The device according to claim 13, wherein: The current classification user group screening module is also used for: Determining the number of users in each of the classified user groups; Sorting each of the classified user groups in descending order according to the number of users; The classified user group with the largest number of users is used as the current classified user group.

15. The device according to claim 13, wherein: The current classification user group screening module is also used for: Determining a group label for each of the classified user groups according to the user association data of the user; Calculating the weight of the group label of each of the classified user groups; Sorting the classified user groups in descending order according to the weight of the group labels; The classified user group with the largest weight of the group label is used as the current classified user group.

16. The apparatus according to claim 14 or 15, further comprising a current classified user group updating module, configured to: When it is determined that the target interaction data of the current classified user group has completed the interaction, the classified user group with the largest number of users is selected from the remaining classified user groups in sequence according to the sorting result of the classified user groups as the target replacement classified user group; The current classified user group is updated according to the target replacement classified user group.

17. The apparatus according to claim 14 or 15, further comprising a current classified user group updating module, configured to: When it is determined that the target interaction data of the current classified user group has completed the interaction, calculating the similarity between the remaining classified user groups and the current classified user group; According to the similarity between the remaining classified user groups and the current classified user group, the remaining classified user groups with the greatest similarity are selected in turn as target replacement classified user groups; The current classified user group is updated according to the target replacement classified user group.

18. The device according to claim 17, wherein: The current classification user group updating module is also used for: Acquire real-time group interaction data between the remaining classified user groups and the current classified user group; The group similarity between the remaining classified user groups and the current classified user group is calculated according to the group real-time interaction data.

19. The device according to claim 17, wherein: The current classification user group updating module is also used for: Acquire real-time group interaction data between the remaining classified user groups and the current classified user group; Determine the current group interaction problem between the remaining classified user groups and the current classified user group according to the real-time group interaction data; The similarity of the interaction problems between the remaining classified user groups and the current group interaction problems of the current classified user group is calculated.

20. The device according to claim 19, wherein The target interaction data includes user question data, and the number of the user question data is multiple; the group real-time interaction module is also used for: Classify and process the multiple user question data of the current classified user group to obtain classified question data; Filtering current classified question data from each of the classified question data; Generate answer data for the current classified question data as the digital human interaction data.

21. The device according to claim 20, wherein: The group real-time interaction module is also used for: Sorting each of the classified question data according to a priority sorting reference factor; wherein the priority sorting reference factor includes at least one of the number of user questions, the frequency of question occurrence, and the user's urgency; The current classified question data is filtered according to the sorting result of the classified question data.

22. The device according to claim 13, further comprising an interface attribute dynamic adjustment module, configured to: Determine the interface attributes of the current real-time interaction interface of the current multi-user real-time interaction scene; wherein, The interface attributes include the current layout format and / or the current digital human image; The interface attributes of the current real-time interactive interface are dynamically adjusted in real time according to the current classified user group.

23. The device according to claim 22, wherein: The interface attribute dynamic adjustment module is also used for: Dynamically adjusting the interface attributes of the current real-time interaction interface in real time according to the target interaction data of the current classified user group; or The interface attributes of the current real-time interactive interface are dynamically adjusted in real time according to the group label of the current classified user group.

24. The apparatus according to claim 13, further comprising an independent interaction module, configured to: Filtering target real-time interaction data of a target user from the real-time interaction data of each of the users; Generate independent interaction data according to the data content of the target real-time interaction data; The independent interaction data is sent to the target user through an independent channel for real-time interaction.

25. An electronic device comprising: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the real-time interaction method based on digital humans according to any one of claims 1 to 12.

26. A non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable a computer to execute the real-time interaction method based on digital human according to any one of claims 1 to 12.

27. A computer program product comprising a computer program / instructions, wherein: When the computer program / instruction is executed by a processor, the real-time interaction method based on digital human described in any one of claims 1 to 12 is implemented.

Images