AI live broadcast method and system based on large language model

Abstract

The invention relates to an AI live broadcast method and system based on a large language model. The method comprises the following steps: generating an AI anchor in a network platform live broadcast room; creating a voice chat link with the user according to a connection request initiated by the user; converting the voice data of the user in the voice chat link into a voice question text; inputting the voice question text into a large language model, and linking the large language model to a product database; outputting a reply text by combining a large language model with a product database; and converting the reply text into audio, and replying to the user through the AI anchor. The voice question of the user is converted into the text in real time, and the personalized voice reply is generated by using the large language model, so that richer and more vivid communication experience is provided. And compared with the existing AI live broadcast based on characters, the tedious process that the user needs to typewrite to ask questions is solved, and the real-time interaction effect is improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to an AI live broadcast method and system based on a large language model. Background Art

[0002] AI live streaming refers to the use of artificial intelligence technology to assist or completely drive the generation and management of live content. Live streaming is done through AI-driven virtual characters (such as virtual anchors or virtual assistants). These virtual anchors can interact with viewers in real time, answer questions, and even adjust content based on audience emotions and feedback.

[0003] Existing technology can only realize AI live broadcast in the form of text-based broadcasting. It is unable to conduct voice communication like a real anchor, and is unable to conduct real emotional interaction. The user experience is relatively bland and lacks interactivity and appeal. Summary of the invention

[0004] In order to at least overcome to a certain extent the problem that the related technology can only realize AI live broadcast in the form of text-based broadcasting and cannot meet the users' various live broadcasting needs, the present application provides an AI live broadcast method and system based on a large language model.

[0005] The scheme of this application is as follows:

[0006] According to a first aspect of an embodiment of the present application, there is provided an AI live broadcast method based on a large language model, comprising:

[0007] Generate AI anchors in the live broadcast room of the network platform;

[0008] Create a voice chat link with the user based on the connection request initiated by the user;

[0009] Convert the user's voice data in the voice chat link into voice question text;

[0010] Inputting the voice question text into a large language model, and linking the large language model to a product database;

[0011] Output the reply text through the combination of large language model and product database;

[0012] The reply text is converted into audio, and the AI ​​anchor responds to the user.

[0013] Preferably, the method further comprises:

[0014] Input the bullet comment text sent by the user into the large language model;

[0015] Output the reply text through the combination of large language model and product database;

[0016] The reply text is converted into audio, and the AI ​​anchor responds to the user.

[0017] Preferably, the method further comprises:

[0018] Performing sensitive content recognition on the voice question text;

[0019] If there is sensitive content in the voice question text, the voice question text will be intercepted.

[0020] Preferably, the method further comprises:

[0021] Count all the bullet texts in the current time period;

[0022] Prioritize all bullet texts in the current time period;

[0023] Take the bullet text with a preset number of digits in the priority sorting and input it into the large language model;

[0024] By combining a large language model with the product database, the corresponding reply text for each bullet comment text is output;

[0025] The reply text is converted into audio, and the AI ​​anchor replies in order according to the priority order corresponding to each reply text.

[0026] Preferably, the priority sorting of all bullet-screen texts in the current time period includes:

[0027] Parse the barrage text, divide the barrage text into barrage question text and barrage chat text, and remove the barrage chat text;

[0028] Determine the questions in each bullet screen question text;

[0029] According to the preset question priority, determine the first score value of each barrage question text;

[0030] According to the frequency of occurrence of various types of questions in all the barrage question texts in the current time period, the frequency of various types of questions is sorted;

[0031] Determine a second score value for each barrage question text according to the frequency ranking;

[0032] Determine a third score value of each barrage question text according to the live broadcast room level of the user sending each barrage question text in the current live broadcast room;

[0033] According to the first score value, the second score value and the third score value of each barrage question text, and the weight assigned to each score value, a final score value of each barrage question text is obtained;

[0034] All the barrage question texts in the current time period are prioritized according to the final score value of each barrage question text.

[0035] Preferably, the method further comprises:

[0036] Real-time detection of whether the voice question text, barrage text or reply text mentions product-related content;

[0037] When it is detected that product-related content is mentioned in the voice question text, barrage text or reply text, the product mentioned in the voice question text, barrage text or reply text is displayed.

[0038] Preferably, the product is displayed, including:

[0039] Retrieving product information and product links from the product database;

[0040] The product information and product link are pushed to the live broadcast room for display.

[0041] Preferably, the product is displayed, including:

[0042] Transport products to designated locations via conveyor belts;

[0043] The robot grabs the product and moves it along the preset path for display;

[0044] Alternatively, a robotic arm can grab the product and place it on a display stand for rotational display.

[0045] Preferably, the method further comprises:

[0046] After the human anchor comes online, turn off the AI ​​anchor according to the instructions and switch to the human anchor;

[0047] The reply text output by the large language model is displayed to the artificial anchor.

[0048] According to a second aspect of an embodiment of the present application, there is provided an AI live broadcast system based on a large language model, comprising:

[0049] Processor and memory;

[0050] The processor and the memory are connected via a communication bus:

[0051] Wherein, the processor is used to call and execute the program stored in the memory;

[0052] The memory is used to store a program, and the program is used to execute at least one of the AI ​​live broadcast methods based on a large language model as described above.

[0053] The technical solution provided by this application may have the following beneficial effects:

[0054] This technical solution can convert users' voice questions into text in real time, and use a large language model to generate personalized voice replies, providing a richer and more vivid communication experience. Compared with the existing text-based AI live broadcast, this solves the tedious process of users needing to ask questions by typing, and improves the effect of real-time interaction. Through the combination of a large language model and a product database, the system can generate answers that are highly relevant to the live broadcast content. It can not only answer audience questions, but also display product information in real time and promote products. This technical effect greatly enriches the function of live broadcast, breaks through the limitation of traditional live broadcast that relies only on human hosting, and enhances the dual effects of commercialization and user experience.

[0055] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0057] Figure 1 This is a flowchart of an AI live broadcast method based on a large language model provided by an embodiment of the present application;

[0058] Figure 2 It is a flowchart of an AI live broadcast method based on a large language model provided by another embodiment of the present application;

[0059] Figure 3 It is a structural schematic diagram of a product transport conveyor belt provided in one embodiment of the present application.

[0060] Reference numerals: cargo lifting belt-1; product storage frame-2; system conveyor belt-3; link shaft-4; product-5; photosensitive module-6. DETAILED DESCRIPTION

[0061] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0062] Embodiment 1

[0063] Figure 1This is a flowchart of an AI live broadcast method based on a large language model provided by an embodiment of the present application, with reference to Figure 1 , an AI live broadcast method based on a large language model, comprising:

[0064] S11: Generate an AI anchor in the live broadcast room of the network platform;

[0065] AI live streaming refers to the use of artificial intelligence technology to assist or completely drive the generation and management of live content. Live streaming is carried out through AI-driven virtual characters (such as virtual anchors or virtual assistants).

[0066] S12: creating a voice chat link with the user according to the connection request initiated by the user;

[0067] A voice chat link can also be called a voice chat room, which is a scene where multiple people can have online voice chats.

[0068] S13: converting the user's voice data in the voice chat link into voice question text;

[0069] Speech recognition technology: The core technology is to convert audio signals into text. Common speech recognition methods include acoustic models based on deep learning (such as deep neural networks, convolutional neural networks) and language models. The system compares the speech patterns in the audio signal with the pre-trained model to identify the corresponding text content.

[0070] S14: inputting the voice question text into the large language model, and linking the large language model to the product database;

[0071] A large language model refers to a large-scale neural network model trained based on deep learning technology, which can understand and generate natural language text. Large language models are usually trained using a large amount of text data to learn information such as language structure, grammar, semantics, and context.

[0072] S15: Output the reply text through the large language model combined with the product database;

[0073] S16: Convert the reply text into audio, and reply to the user through the AI ​​anchor.

[0074] In actual practice, a small amount of audio files can be used to train a reply voice that matches the host's voice, and then the reply text can be converted into audio with the host's voice, and the AI ​​host can reply to the user.

[0075] This technical solution can convert users' voice questions into text in real time, and use a large language model to generate personalized voice replies, providing a richer and more vivid communication experience. Compared with the existing text-based AI live broadcast, this solves the tedious process of users needing to ask questions by typing, and improves the effect of real-time interaction. Through the combination of a large language model and a product database, the system can generate answers that are highly relevant to the live broadcast content. It can not only answer audience questions, but also display product information in real time and promote products. This technical effect greatly enriches the function of live broadcast, breaks through the limitation of traditional live broadcast that relies only on human hosting, and enhances the dual effects of commercialization and user experience.

[0076] Embodiment 2

[0077] Reference Figure 2 , the method further comprises:

[0078] Input the bullet comment text sent by the user into the large language model;

[0079] Output the reply text through the combination of large language model and product database;

[0080] Convert the reply text into audio and reply to the user through the AI ​​anchor.

[0081] Users usually interact with each other in live broadcast platforms through voice and bullet comments. As a major form of interaction, bullet comments can display audience questions, comments or suggestions in real time. If you only rely on voice or a single text input method, the interactivity of the live broadcast will be limited. By inputting the bullet comment text into the large language model, the AI ​​anchor can respond to a large number of audience questions or comments in a short period of time, improving the user experience of participation.

[0082] The processing of barrage text can automatically analyze the audience's questions and generate appropriate responses in a very short time. AI anchors can not only interact with each other in real time, but also quickly generate contextual responses based on the audience's barrage content, avoiding the lag problem when human anchors cannot cope with a large number of barrages. In addition, since the large language model has powerful language understanding and generation capabilities, the generated responses are usually more natural and accurate, effectively improving the audience experience.

[0083] By inputting the barrage text into a large language model and combining it with a product database, AI anchors can provide more personalized responses to user questions. This customized feedback can not only answer specific questions, but also automatically guide users to related products, services, or other content based on the content of the question, enhancing the commercial value of the live broadcast platform. For example, if a user asks about the usage or purchase information of a product, the AI ​​anchor can obtain relevant information from the product database in real time and show the user detailed information and purchase links of the product.

[0084] In traditional live broadcasts, human hosts usually need to manually select and respond to a large amount of barrage content, especially when the number of viewers is large, the burden on human hosts will increase. By automatically inputting barrage content into a large language model for processing and generating replies, AI hosts can automatically respond to a large amount of user interactive content, greatly improving the response efficiency of live broadcasts. Human hosts can focus on more important tasks, such as answering complex questions and managing special scenarios.

[0085] Embodiment 3

[0086] Reference Figure 2 , the method further comprises:

[0087] Identify sensitive content in voice question text;

[0088] If there is sensitive content in the voice question text, the voice question text will be intercepted.

[0089] In actual live broadcast scenarios, the voice questions of connected users may contain inappropriate content, such as malicious comments, advertisements, etc. Therefore, combined with sensitive content filtering technology, voice question text can be reviewed and filtered before entering the large language model to ensure the safety and health of live broadcast content. This provides an effective content monitoring and management mechanism for live broadcast platforms, reducing the negative impact of inappropriate speech on the live broadcast atmosphere.

[0090] Embodiment 4

[0091] It should be noted that the method also includes:

[0092] Count all the bullet texts in the current time period;

[0093] Prioritize all bullet texts in the current time period;

[0094] Take the bullet text with a preset number of digits in the priority sorting and input it into the large language model;

[0095] By combining a large language model with the product database, the corresponding reply text for each bullet comment text is output;

[0096] The reply text is converted into audio, and the AI ​​anchor replies in order according to the priority of each reply text.

[0097] Among them, all bullet texts in the current time period are prioritized, including:

[0098] Parse the barrage text, divide the barrage text into barrage question text and barrage chat text, and remove the barrage chat text;

[0099] Determine the questions in each bullet screen question text;

[0100] According to the preset question priority, determine the first score value of each barrage question text;

[0101] According to the frequency of occurrence of various types of questions in all the barrage question texts in the current time period, the frequency of various types of questions is sorted;

[0102] According to the frequency sorting, determine the second score value of each barrage question text;

[0103] Determine a third score value of each barrage question text according to the live broadcast room level of the user sending each barrage question text in the current live broadcast room;

[0104] According to the first score value, the second score value and the third score value of each barrage question text, and the weight assigned to each score value, a final score value of each barrage question text is obtained;

[0105] All the barrage question texts in the current time period are prioritized according to the final score value of each barrage question text.

[0106] During the live broadcast, viewers will continue to send bullet comments, which may include a large number of questions, comments, suggestions, etc. Due to the requirements of real-time and interactivity, it is impossible to respond to every bullet comment immediately. Therefore, the system prioritizes all bullet comments in the current time period and determines which bullet comments should be responded to first.

[0107] The priority sorting of bullet comments can take into account multiple factors, such as:

[0108] The urgency or importance of the question: For example, some questions may involve product features or common problems, and the system will give priority to these comments.

[0109] Frequency of questions asked: If the same or similar questions appear frequently in the barrage, this indicates that this is a question that the audience generally cares about, and the system should give priority to responding.

[0110] User interaction level or contribution: Some live streaming platforms may give higher priority to users’ comments based on their contribution on the platform (such as level, points, etc.) to encourage active users to participate in the interaction.

[0111] Priority sorting allows AI hosts to respond to audiences’ concerns first, thereby improving the efficiency and quality of interaction and ensuring that important questions are answered in a timely manner. When the number of viewers is large, AI hosts can allocate time reasonably and first deal with questions or concerns that are valuable to more viewers. This approach allows viewers to feel that their questions are being responded to in a timely manner, improving user participation and the interactivity of the live broadcast.

[0112] Once the barrage texts are prioritized, the system will input the high-priority barrage texts in the preset positions into the large language model to generate the corresponding reply texts. These texts will then be converted into audio and voice replies will be made by the AI ​​anchor. The AI ​​anchor does not rely on human anchors to process the barrages one by one, but can automatically generate responses based on the barrage content, significantly improving the real-time interactive capabilities of live broadcasts.

[0113] Specifically, first, the system will classify the barrage text and distinguish between question text and chat text. Among them, chat text is mainly users' idle chats, comments, etc., which usually have no direct impact on the content of the live broadcast, so no reply is required. The system will remove these chat texts and focus on the content of the questions. By classifying the barrage text, the system can focus on processing the question text, avoid wasting computing resources to process irrelevant chat text, and improve processing efficiency. Users' questions are responded to first, avoiding the audience's questions being drowned in a large amount of irrelevant chat content during the live broadcast, and ensuring the response rate of questions.

[0114] Then, for the question text, the system will score and sort it according to the following steps:

[0115] First score value: score according to the preset question priority. For example, some common questions (such as product usage, payment process, etc.) may be given a higher priority.

[0116] Second score: Sort questions by frequency. Questions with higher frequency will get higher scores. This way, we can focus resources on answering more questions that the audience cares about.

[0117] The third score value: give the bullet comments a higher weight based on the user level of the questioner (such as VIP level, activity, etc.). This practice can encourage active users to ask questions and increase their sense of participation.

[0118] By comprehensively considering the priority, frequency and activity of the questioner, the system can intelligently assign weights to each question text, so that the most urgent and common questions during the live broadcast are responded to first. Through this precise sorting mechanism, the system can respond to more core questions of the audience within a limited time, improving the interactive effect of the live broadcast and the audience's satisfaction.

[0119] Embodiment 5

[0120] It should be noted that the method also includes:

[0121] Real-time detection of whether product-related content is mentioned in voice question text, barrage text or reply text;

[0122] When it is detected that product-related content is mentioned in the voice question text, barrage text or reply text, the product mentioned in the voice question text, barrage text or reply text is displayed.

[0123] This technical solution will not only process voice question texts and barrage texts, but also analyze whether these texts involve product-related content. For example, users may ask about the characteristics, price, usage methods, etc. of a certain product, or mention a certain product in the barrage. The system needs to analyze these texts in real time to detect whether there is information related to the product. By detecting in real time whether the text content contains product-related information, the system can accurately identify the user's interest and needs for the product, ensuring that product-related information can be processed and displayed in a timely manner. When a text (voice, barrage or reply) is detected to mention a product, the system will display relevant information about the product. Product display can be carried out in a variety of ways (such as text information display, link push, video display, etc.) to meet the needs of different users. By combining products with interactive content, the interactivity of live broadcasts and the audience's sense of participation are enhanced, allowing users to directly obtain product information during the interaction, making live broadcasts not only entertainment or information sharing, but also product promotion.

[0124] Furthermore, the product is displayed in two forms:

[0125] 1) Retrieve product information and product links from the product database;

[0126] Push product information and product links to the live broadcast room for display.

[0127] When a product is mentioned in the text, the system will then search for information in the product database. This information may include product name, price, description, picture, video, etc., and the system will also retrieve information such as product purchase links and promotions. By retrieving detailed information from the product database, the system can provide accurate product descriptions and purchase information to ensure that viewers can fully understand the product features when they see the product display. By displaying product information and purchase links, users can quickly click to enter the purchase page, which improves product conversion rates and increases the commercial benefits of the live broadcast platform.

[0128] 2) Transport the product to the designated location via a conveyor belt;

[0129] The robot grabs the product and moves it along the preset path for display;

[0130] Alternatively, a robotic arm can grab the product and place it on a display stand for rotational display.

[0131] Reference Figure 3The cargo lifting belt 1 is a conveyor belt that reciprocates up and down to store products. The cargo lifting belt 1 has multiple product storage frames 2, each of which stores products. The cargo lifting belt 1 is driven by a stepper motor, which can accurately control the position of each product storage frame 2 and calculate the position of the product storage frame 2 according to the number of rotations and the length of the lifting belt. The link shaft 4 is used to fix the product storage frame 2 on the cargo lifting belt 1. The cargo lifting belt 1 can transfer the specified product to the system conveyor belt 3 through instructions, and the photosensitive module 6 is used to determine whether the product has been transferred to the specified position.

[0132] The conveyor belt moves the product from the backstage or designated location to the display stand or in front of the camera to increase the visual sense of dynamics. The robotic arm can grab the product and display it to the designated location of the live broadcast room according to the preset path, or place the product on a rotating display stand, so that the product can be displayed in all directions and provide the audience with a more intuitive product view. Through the dynamic display of the conveyor belt or robotic arm, the product display is no longer monotonous and static, which can attract the audience's attention, increase the visual impact of the display, and enhance the attractiveness of the product. The physical display method not only makes the product display more vivid, but also increases the audience's sense of immersion, making the atmosphere of the live broadcast room more active, and enhancing the audience's interest and participation in the product. Through dynamic display, the product can be displayed more comprehensively, and the audience can see the details of the product from different angles, which helps to increase the exposure and purchase desire of the product.

[0133] Embodiment 6

[0134] It should be noted that the method also includes:

[0135] After the human anchor comes online, turn off the AI ​​anchor according to the instructions and switch to the human anchor;

[0136] The reply text output by the large language model is displayed to the human anchor.

[0137] In some cases, the AI ​​anchor may need to be replaced by a human anchor, for example, when a human anchor is needed to provide a more personalized or emotional response during a live broadcast, or when a human anchor can provide a better experience during complex interactions. At this time, the system will transfer control from the AI ​​anchor to the human anchor through the triggering of the command.

[0138] Even after the human anchor takes over, the system can still provide support to the human anchor. For example, the system can generate reply texts based on user questions through a large language model, and these reply texts can be displayed to the human anchor after the human anchor takes over. The human anchor can provide further answers or dialogue based on these texts.

[0139] In this way, the human anchor can use the efficient reply text generated by the system to respond to the audience's questions, ensuring the timeliness and accuracy of the answer. At the same time, the human anchor can also modify the text according to his own style to make the interaction more emotional and humane. With the help of the content generated by the system, the human anchor can save time and reduce errors or omissions in the conversation while efficiently handling the audience's questions.

[0140] Embodiment 7

[0141] An AI live broadcast system based on a large language model, comprising:

[0142] Processor and memory;

[0143] The processor and memory are connected via a communication bus:

[0144] The processor is used to call and execute the program stored in the memory;

[0145] A memory is used to store a program, and the program is used to execute at least one of the AI ​​live broadcast methods based on a large language model in the above embodiments.

[0146] It can be understood that the same or similar parts of the above embodiments can be referenced to each other, and the contents not described in detail in some embodiments can refer to the same or similar contents in other embodiments.

[0147] It should be noted that, in the description of this application, the terms "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance. In addition, in the description of this application, unless otherwise specified, the meaning of "plurality" refers to at least two.

[0148] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code that includes one or more executable instructions for implementing the steps of a specific logical function or process, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments of the present application belong.

[0149] It should be understood that the various parts of the present application can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0150] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.

[0151] In addition, each functional unit in each embodiment of the present application may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0152] The storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk, etc.

[0153] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0154] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. An AI live broadcast method based on a large language model, characterized in that: include: Generate AI anchors in the live broadcast room of the network platform; Create a voice chat link with the user based on the connection request initiated by the user; Convert the user's voice data in the voice chat link into voice question text; Inputting the voice question text into a large language model, and linking the large language model to a product database; Output the reply text through the combination of large language model and product database; The reply text is converted into audio, and the AI ​​anchor responds to the user.

2. The method according to claim 1, characterized in that The method further comprises: Input the bullet comment text sent by the user into the large language model; Output the reply text through the combination of large language model and product database; The reply text is converted into audio, and the user is replied to through the AI ​​anchor.

3. The method according to claim 2, characterized in that The method further comprises: Performing sensitive content recognition on the voice question text; If there is sensitive content in the voice question text, the voice question text will be intercepted.

4. The method according to claim 2, characterized in that: The method further comprises: Count all the bullet comments in the current time period; Prioritize all bullet texts in the current time period; Take the bullet text with a preset number of digits in the priority sorting and input it into the large language model; By combining a large language model with the product database, the corresponding reply text for each bullet comment text is output; The reply text is converted into audio, and the AI ​​anchor replies in order according to the priority order corresponding to each reply text.

5. The method according to claim 4, characterized in that The priority sorting of all bullet comment texts in the current time period includes: Parse the barrage text, divide the barrage text into barrage question text and barrage chat text, and remove the barrage chat text; Determine the questions in each bullet screen question text; According to the preset question priority, determine the first score value of each barrage question text; According to the frequency of occurrence of various types of questions in all the barrage question texts in the current time period, the frequency of various types of questions is sorted; Determine a second score value for each barrage question text according to the frequency ranking; Determine a third score value of each barrage question text according to the live broadcast room level of the user sending each barrage question text in the current live broadcast room; According to the first score value, the second score value and the third score value of each barrage question text, and the weight assigned to each score value, a final score value of each barrage question text is obtained; All the barrage question texts in the current time period are prioritized according to the final score value of each barrage question text.

6. The method according to claim 1, characterized in that The method further comprises: Real-time detection of whether the voice question text, barrage text or reply text mentions product-related content; When it is detected that product-related content is mentioned in the voice question text, barrage text or reply text, the product mentioned in the voice question text, barrage text or reply text is displayed.

7. The method according to claim 6, characterized in that Demonstrate products, including: Retrieving product information and product links from the product database; The product information and product link are pushed to the live broadcast room for display.

8. The method according to claim 6, characterized in that Demonstrate products, including: Transport products to designated locations via conveyor belts; The robot grabs the product and moves it along the preset path for display; Alternatively, a robotic arm can grab the product and place it on a display stand for rotational display.

9. The method according to claim 1, characterized in that: The method further comprises: After the human anchor comes online, turn off the AI ​​anchor according to the instructions and switch to the human anchor; The reply text output by the large language model is displayed to the artificial anchor.

10. An AI live broadcast system based on a large language model, characterized in that: include: Processor and memory; The processor and the memory are connected via a communication bus: Wherein, the processor is used to call and execute the program stored in the memory; The memory is used to store a program, and the program is at least used to execute an AI live broadcast method based on a large language model as described in any one of claims 1-9.

Images