Live broadcast interaction method and system based on large model, terminal and storage medium

By using a large model in the live broadcast system to analyze and automatically respond to questions asked by audiences, the phenomenon of the anchor missing audience problems is solved, the live broadcast interactive experience and efficiency is improved, and real-time and personalized interactive replies are achieved.

CN119011887APending Publication Date: 2024-11-22BEIJING UNISOUND INFORMATION TECH CO LTD
View PDF 0 Cites 3 Cited by

Patent Information

Application Number
CN202410953601.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-16
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

During the existing live broadcast interaction, the problem of anchor missing viewers often occurs, resulting in a low user experience.

Method used

A live broadcast interaction method based on a large model is adopted to determine the audience's intentions by obtaining questions from audiences and conducting question-and-answer analysis. If the interaction conditions are met, a reply will be automatically generated and automatically interacted when the anchor did not reply in time to ensure that the audience's questions are answered.

Benefits of technology

It improves the user's live broadcast interactive experience, prevents the anchor from missing audience problems, realizes real-time, personalized and accurate reply, reduces manual intervention, and improves the professionalism and attractiveness of live broadcasts.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119011887A_ABST
    Figure CN119011887A_ABST
Patent Text Reader

Abstract

The invention provides a live broadcast interaction method and system based on a large model, a terminal and a storage medium, and the method comprises the steps: obtaining audience questions in a target live broadcast room, inputting the audience questions into a pre-trained intention large model for question and answer analysis, and obtaining audience intentions; if the audience intention meets a live broadcast interaction condition, determining an interaction reply according to the audience intention, and performing interaction prompt on an anchor in the target live broadcast room according to the interaction reply and the audience question; acquiring live broadcast voice of the anchor in a first preset time length, and performing reply detection on the live broadcast voice; and if the reply detection of the live broadcast voice is unqualified, performing live broadcast interaction in the target live broadcast room according to the interaction reply. According to the embodiment of the invention, the interaction prompt is performed on the anchor in the target live broadcast room based on the interaction reply and the audience questions, so that the anchor is prevented from missing the audience questions needing to be replied, and the live broadcast interaction experience of the user is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of Internet technology, and in particular to a live interactive method, system, terminal, and storage medium based on a large model. Background Technology

[0002] In recent years, the internet live streaming industry has gradually transitioned from extensive competition to intensive competition, with user experience becoming the focus of competition among major live streaming platforms. As the number of live streaming users continues to expand, content competition has entered a fierce stage. In the internet world, live streaming has become one of the most effective media for communication and interaction with users.

[0003] In existing live streaming interactions, there are frequent instances where the streamer misses out on viewers, which reduces the user's live streaming experience. Summary of the Invention

[0004] The purpose of this invention is to provide a live interactive method, system, terminal, and storage medium based on a large model, so as to solve the problem of poor user live interactive experience in the prior art.

[0005] This invention is implemented as follows: a live interactive method based on a large model, the method comprising:

[0006] The system acquires questions from viewers in the target live stream and inputs these questions into a pre-trained intent model for question-and-answer analysis to obtain the viewers' intent.

[0007] If the viewer's intention meets the conditions for live interaction, then an interactive response is determined based on the viewer's intention, and interactive prompts are given to the host of the target live room based on the interactive response and the viewer's question;

[0008] Acquire the live broadcast audio of the anchor within a first preset duration, and perform response detection on the live broadcast audio;

[0009] If the response to the live audio message fails the test, then live interaction will be conducted in the target live room based on the interactive response.

[0010] Preferably, after inputting the audience's question into a pre-trained intent model for question-answering analysis to obtain the audience's intent, the method further includes:

[0011] Obtain product information within a second preset duration in the target live stream room, and match the product names in the product information with the audience's intent;

[0012] If the product name in the product information matches the viewer's intent, then the viewer's intent is determined to meet the conditions for live interaction.

[0013] If the product name in the product information does not match the viewer's intent, then the viewer's intent is determined not to meet the conditions for live streaming interaction.

[0014] Preferably, determining the interactive response based on the audience's intent includes:

[0015] The product name matched in the product information is identified as the target product, and the product database of the target product is queried;

[0016] The audience's intent is matched with the product database using a question-and-answer matching method, and the interactive response is generated based on the matching results.

[0017] Preferably, providing interactive prompts to the streamer in the target live stream room based on the interactive responses and audience questions includes:

[0018] Obtain the information display address of the broadcaster, and obtain the username corresponding to the viewer's question;

[0019] Based on the user name and the viewer's intent, viewer questions are displayed in the first area of ​​the information display address;

[0020] The question and answer are displayed in the second area of ​​the information display address based on the interactive response.

[0021] Preferably, the response detection for the live audio includes:

[0022] The live broadcast audio is subjected to speech recognition to obtain speech text, and the speech text is matched with the interactive response to obtain text similarity;

[0023] If the text similarity is less than the similarity threshold, the response detection of the live audio is deemed unqualified.

[0024] If the text similarity is greater than or equal to the similarity threshold, the response detection of the live audio is deemed qualified.

[0025] Preferably, after conducting live interaction in the target live room based on the interactive response, the method further includes:

[0026] If it is detected that the anchor has not conducted a voice live broadcast within a third preset time period, the product information of the current live broadcast is obtained, and a co-broadcast interactive voice is generated based on the product information of the current live broadcast.

[0027] Live interaction is conducted in the target live room based on the secondary broadcaster's interactive voice until the main broadcaster's live voice is captured.

[0028] Preferably, before inputting the audience's question into a pre-trained intent model for question-answering analysis to obtain the audience's intent, the method further includes:

[0029] Obtain historical Q&A data from live streams, and generate Q&A text corpus based on the historical Q&A data from live streams;

[0030] The question-and-answer text corpus is cleaned and labeled to obtain model training corpus, and the model training corpus is input into the intent big model for question-and-answer prediction to obtain predicted responses;

[0031] The model loss is calculated based on the predicted response, and the parameters of the intent model are updated based on the model loss until the intent model converges, thus obtaining the pre-trained intent model.

[0032] Another objective of this invention is to provide a live interactive system based on a large model, the system comprising:

[0033] The question-and-answer analysis module is used to acquire audience questions in the target live broadcast room and input the audience questions into a pre-trained intent model for question-and-answer analysis to obtain the audience intent.

[0034] An interactive prompt module is used to determine an interactive response based on the viewer's intent if the viewer's intent meets the live interaction conditions, and to provide interactive prompts to the host of the target live room based on the interactive response and the viewer's question.

[0035] The response detection module is used to acquire the live broadcast audio of the anchor within a first preset duration and to perform response detection on the live broadcast audio.

[0036] The live interaction module is used to conduct live interaction in the target live room based on the interaction response if the response to the live voice fails to pass the detection.

[0037] In this embodiment of the invention, by inputting audience questions into a pre-trained intent model for question-and-answer analysis, audience intent can be effectively obtained. By detecting whether the audience intent meets the conditions for live interaction, it can be effectively determined whether a response is needed for the audience question. If the audience intent meets the conditions for live interaction, an interactive response is automatically determined based on the audience intent. Based on the interactive response and the audience question, interactive prompts are given to the host in the target live room to prevent the host from missing audience questions that need to be answered, thus improving the user's live interaction experience. By detecting responses to live audio, it can be determined whether the host has responded to audience questions that need to be answered. If no response has been given, live interaction is automatically initiated in the target live room based on the interactive response, further preventing the omission of audience questions that need to be answered. Attached Figure Description

[0038] Figure 1 This is a flowchart of the live interactive method based on a large model provided in the first embodiment of the present invention;

[0039] Figure 2 This is a schematic diagram of the structure of the live interactive system based on a large model provided in the second embodiment of the present invention;

[0040] Figure 3 This is a schematic diagram of the structure of the terminal device provided in the third embodiment of the present invention. Detailed Implementation

[0041] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0042] To illustrate the technical solution described in this invention, specific embodiments are described below.

[0043] Example 1

[0044] Please see Figure 1 This is a flowchart of a live streaming interaction method based on a large model provided in the first embodiment of the present invention. This live streaming interaction method based on a large model can be applied to any device or system, and includes the following steps:

[0045] Step S10: Obtain audience questions from the target live stream and input the audience questions into a pre-trained intent big data model for question-and-answer analysis to obtain the audience intent;

[0046] Specifically, when any viewer asks a question in the target live stream, the text of the viewer's question is obtained in real time. The pre-trained intent model performs question-and-answer analysis on the viewer's question to obtain the viewer's intent.

[0047] Optionally, after inputting the audience's question into a pre-trained intent model for question-answering analysis and obtaining the audience's intent, the method further includes:

[0048] Obtain product information within a second preset duration in the target live stream room, and match the product names in the product information with the audience's intent; wherein, the second preset duration can be set according to needs, for example, the second preset duration can be set to the duration range 5 minutes before the current time point, that is, match the product names in the product information within the duration range 5 minutes before the current time point with the audience's intent;

[0049] If the product name in the product information matches the viewer's intent, then the viewer's intent is determined to meet the conditions for live interaction; wherein, if the product name in the product information matches the viewer's intent, then the viewer's question is determined to require an interactive response.

[0050] If the product name in the product information does not match the viewer's intent, the viewer's intent is determined not to meet the conditions for live interaction; wherein, if the product name in the product information does not match the viewer's intent, the viewer's question is determined not to require an interactive response.

[0051] Furthermore, before obtaining the audience's intent by inputting the audience's question into a pre-trained intent model for question-answering analysis, the process also includes:

[0052] Obtain historical Q&A data from live streams, and generate Q&A text corpus based on the historical Q&A data from live streams;

[0053] The question-and-answer text corpus is cleaned and labeled to obtain model training corpus, and the model training corpus is input into the intent big model for question-and-answer prediction to obtain predicted responses;

[0054] The model loss is calculated based on the predicted response, and the parameters of the intent model are updated based on the model loss until the intent model converges, thus obtaining the pre-trained intent model.

[0055] Step S20: If the viewer's intention meets the live interaction conditions, then determine the interaction response based on the viewer's intention, and provide interaction prompts to the anchor of the target live room based on the interaction response and the viewer's question;

[0056] If a viewer's intention meets the conditions for live interaction, the system automatically determines the interactive response based on the viewer's intention. Based on the interactive response and the viewer's question, the system provides interactive prompts to the host in the target live room to prevent the host from missing any viewer questions that need to be answered, thereby improving the user's live interaction experience.

[0057] Optionally, determining the interactive response based on the audience's intent includes:

[0058] The product name matched in the product information is identified as the target product, and the product database of the target product is queried;

[0059] The audience's intent is matched with the product database using a question-and-answer matching method, and the interactive response is generated based on the matching results.

[0060] This involves obtaining the database address corresponding to the target product and determining the product database based on the database address. The product database stores the current data information of the target product, including information such as the current promotion, current price, remaining quantity, and delivery time.

[0061] Furthermore, based on the interactive responses and audience questions, interactive prompts are provided to the streamer in the target live stream room, including:

[0062] Obtain the information display address of the broadcaster and obtain the username corresponding to the audience's question; wherein, the information display address corresponds to the broadcaster's live streaming viewing device, which can be a mobile phone, tablet or computer or other devices.

[0063] Based on the user name and the viewer's intent, the viewer's question is displayed in the first area of ​​the information display address, and based on the interactive response, the question reply is displayed in the second area of ​​the information display address.

[0064] The locations of the first and second zones can be set according to needs. For example, the central control screen of the live streaming device can display a text prompt to the host, such as "User XX asked, how many days will it take for product number 5 to ship?" The interactive reply can be "Ships within three days, baby."

[0065] Step S30: Obtain the live broadcast audio of the anchor within a first preset duration, and perform response detection on the live broadcast audio;

[0066] The first preset duration can be set according to needs. For example, the time for the host to interact with the host based on interactive replies and audience questions can be used as the starting point to obtain 40 seconds of live audio. The host can then perform response detection on the live audio to determine whether the host has responded to the audience questions that need to be answered.

[0067] Optionally, response detection is performed on the live audio, including:

[0068] The live broadcast audio is subjected to speech recognition to obtain speech text, and the speech text is matched with the interactive response to obtain text similarity;

[0069] If the text similarity is less than the similarity threshold, the response detection of the live audio is deemed unqualified.

[0070] If the text similarity is greater than or equal to the similarity threshold, the response detection of the live audio is deemed qualified.

[0071] The similarity threshold can be set according to needs. If the text similarity is less than the similarity threshold, it is determined that the host has not interacted or responded. If the text similarity is greater than or equal to the similarity threshold, it is determined that the host has interacted or responded.

[0072] Step S40: If the response detection of the live voice is unqualified, then live interaction is carried out in the target live room according to the interactive response;

[0073] If the host does not respond, a live interaction will be automatically initiated in the target live room based on the response. This live interaction can be conducted via voice, text, or images.

[0074] Optionally, after conducting live interaction in the target live room based on the interactive response, the method further includes:

[0075] If it is detected that the host has not conducted a voice live broadcast within a third preset time period, the product information of the current live broadcast is obtained, and a secondary broadcast interactive voice is generated based on the product information of the current live broadcast; wherein, the third preset time period can be set according to needs. For example, if it is detected that the host has not conducted a voice live broadcast within 40 seconds, the secondary broadcast interactive voice is automatically generated based on the product information of the current live broadcast.

[0076] Live interaction is conducted in the target live room based on the secondary broadcast interactive voice until the main broadcaster's live voice is captured; wherein, conducting live interaction in the target live room through the secondary broadcast interactive voice can effectively create an atmosphere and prevent the target live room from becoming quiet.

[0077] In this step, when the host is unsure how to continue the live stream (the host's audio stream is silent for more than 40 seconds), a co-host automatically takes over, creating an atmosphere and boosting sales by providing examples and emphasizing the remaining quantity of products. The system also acquires the host's audio stream, transcribes it into text, and uses a large model to determine if supplementary product information, answers user questions, or encourages further attention is needed, providing supplementary content as needed.

[0078] In this embodiment, by inputting audience questions into a pre-trained intent model for question-and-answer analysis, audience intent can be effectively obtained. By detecting whether the audience intent meets the conditions for live interaction, it can be effectively determined whether a response is needed for the audience question. If the audience intent meets the conditions for live interaction, an interactive response is automatically determined based on the audience intent. Based on the interactive response and the audience question, interactive prompts are given to the host in the target live room to prevent the host from missing audience questions that need to be answered, thus improving the user's live interaction experience. By detecting responses to live audio, it is determined whether the host has responded to audience questions that need to be answered. If no response has been made, live interaction is automatically initiated in the target live room based on the interactive response, further preventing the omission of audience questions that need to be answered. This embodiment can quickly obtain real-time product information, making product promotion more accurate and effective. Through close cooperation with the host, it can handle various live streaming scenarios, whether the host's speech is unclear or the host is unsure how to continue, effectively addressing these issues to achieve good product promotion and audience experience. By obtaining audience questions in real time and responding quickly, the problem of interaction delay in traditional live streaming is solved. By utilizing the deep learning and natural language processing capabilities of the large model, personalized and accurate responses are provided. It enables automated control of the live streaming room, reducing manual intervention and improving efficiency. The large-scale model can not only answer audience questions, but also assist the host in introducing products and creating a lively atmosphere, enhancing the professionalism and appeal of the live stream.

[0079] Example 2

[0080] Please see Figure 2 This is a schematic diagram of the structure of a live interactive system 100 based on a large model provided in the second embodiment of the present invention, including:

[0081] The question-and-answer analysis module 10 is used to obtain audience questions in the target live broadcast room and input the audience questions into a pre-trained intent big data model for question-and-answer analysis to obtain the audience intent.

[0082] Optionally, the question-and-answer analysis module 10 is further configured to: obtain product information within a second preset duration in the target live streaming room, and match the product names in the product information with the audience's intent;

[0083] If the product name in the product information matches the viewer's intent, then the viewer's intent is determined to meet the conditions for live interaction.

[0084] If the product name in the product information does not match the viewer's intent, then the viewer's intent is determined not to meet the conditions for live streaming interaction.

[0085] Furthermore, the question-and-answer analysis module 10 is also used to: acquire historical question-and-answer data from the live stream, and generate question-and-answer text corpus based on the historical question-and-answer data from the live stream;

[0086] The question-and-answer text corpus is cleaned and labeled to obtain model training corpus, and the model training corpus is input into the intent big model for question-and-answer prediction to obtain predicted responses;

[0087] The model loss is calculated based on the predicted response, and the parameters of the intent model are updated based on the model loss until the intent model converges, thus obtaining the pre-trained intent model.

[0088] The interactive prompt module 11 is used to determine an interactive response based on the viewer's intention if the viewer's intention meets the live interaction conditions, and to provide interactive prompts to the host of the target live room based on the interactive response and the viewer's question.

[0089] Optionally, the interactive prompt module 11 is also used to: determine the product name matched in the product information as the target product, and query the product database of the target product;

[0090] The audience's intent is matched with the product database using a question-and-answer matching method, and the interactive response is generated based on the matching results.

[0091] Furthermore, the interactive prompt module 11 is also used to: obtain the information display address of the anchor and obtain the username corresponding to the audience's question;

[0092] Based on the user name and the viewer's intent, viewer questions are displayed in the first area of ​​the information display address;

[0093] The question and answer are displayed in the second area of ​​the information display address based on the interactive response.

[0094] The response detection module 12 is used to acquire the live broadcast audio of the anchor within a first preset duration and to perform response detection on the live broadcast audio.

[0095] Optionally, the response detection module 12 is further configured to: perform speech recognition on the live broadcast audio to obtain speech text, and perform similarity matching between the speech text and the interactive response to obtain text similarity;

[0096] If the text similarity is less than the similarity threshold, the response detection of the live audio is deemed unqualified.

[0097] If the text similarity is greater than or equal to the similarity threshold, the response detection of the live audio is deemed qualified.

[0098] The live interaction module 13 is used to conduct live interaction in the target live room based on the interaction response if the response detection of the live voice is unqualified.

[0099] Optionally, the live streaming interaction module 13 is also used to: if it is detected that the anchor has not conducted a voice live stream within a third preset time period, obtain the product information of the current live stream, and generate a co-broadcast interactive voice based on the product information of the current live stream;

[0100] Live interaction is conducted in the target live room based on the secondary broadcaster's interactive voice until the main broadcaster's live voice is captured.

[0101] In this embodiment, by inputting audience questions into a pre-trained intent model for question-and-answer analysis, audience intent can be effectively obtained. By detecting whether the audience intent meets the conditions for live interaction, it can be effectively determined whether a response is needed for the audience question. If the audience intent meets the conditions for live interaction, an interactive response is automatically determined based on the audience intent. Based on the interactive response and the audience question, interactive prompts are given to the host in the target live room to prevent the host from missing audience questions that need to be answered, thus improving the user's live interaction experience. By detecting responses to live audio, it can be determined whether the host has responded to audience questions that need to be answered. If no response has been given, live interaction is automatically initiated in the target live room based on the interactive response, further preventing the omission of audience questions that need to be answered.

[0102] Example 3

[0103] Figure 3 This is a structural block diagram of a terminal device 2 provided in the third embodiment of this application. For example... Figure 3 As shown, the terminal device 2 in this embodiment includes a processor 20, a memory 21, and a computer program 22 stored in the memory 21 and executable on the processor 20, such as a program for a live interactive method based on a large model. When the processor 20 executes the computer program 22, it implements the steps in the various embodiments of the live interactive method based on the large model described above.

[0104] For example, the computer program 22 may be divided into one or more modules, which are stored in the memory 21 and executed by the processor 20 to complete this application. The one or more modules may be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program 22 in the terminal device 2. The terminal device may include, but is not limited to, the processor 20 and the memory 21.

[0105] The processor 20 may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0106] The memory 21 can be an internal storage unit of the terminal device 2, such as a hard drive or memory of the terminal device 2. The memory 21 can also be an external storage device of the terminal device 2, such as a plug-in hard drive, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the terminal device 2. Furthermore, the memory 21 can include both internal and external storage units of the terminal device 2. The memory 21 is used to store the computer program and other programs and data required by the terminal device. The memory 21 can also be used to temporarily store data that has been output or will be output.

[0107] Furthermore, the functional modules in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0108] If an integrated module is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. This computer-readable storage medium can be non-volatile or volatile. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable storage medium can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the contents of a computer-readable storage medium may be appropriately added to or subtracted from the contents as required by the legislation and patent practice in a jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, a computer-readable storage medium may not include electrical carrier signals and telecommunication signals.

[0109] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A live interactive method based on a large model, It is characterized in that The method comprises: Obtain audience questions in the target live broadcast room, and input the audience questions into the pre-trained intention model for question-answer analysis to obtain audience intentions; If the audience intention meets the live broadcast interaction conditions, an interactive response is determined according to the audience intention, and an interactive prompt is given to the host of the target live broadcast room according to the interactive response and the audience question; Acquire the live broadcast voice of the anchor within a first preset time period, and perform reply detection on the live broadcast voice; If the response detection of the live broadcast voice fails, live broadcast interaction is performed in the target live broadcast room according to the interactive response.

2. The live interactive method based on a large model as claimed in claim 1, It is characterized in that Inputting the audience's question into the pre-trained intention model for question-answer analysis, and obtaining the audience's intention, further comprising: Acquire product information within a second preset time period in the target live broadcast room, and match the product name in the product information with the viewer's intention; If the product name in the product information successfully matches the viewer's intention, it is determined that the viewer's intention meets the live broadcast interaction condition; If the product name in the product information fails to match the viewer's intention, it is determined that the viewer's intention does not meet the live broadcast interaction condition.

3. The live interactive method based on a large model as claimed in claim 2, It is characterized in that Determine interactive responses based on stated audience intent, including: Determine the product name matched in the product information as the target product, and query the product database of the target product; The audience intention is matched with the product database through question and answer, and the interactive response is generated according to the question and answer matching result.

4. The live interactive method based on a large model as claimed in claim 1, It is characterized in that Providing interactive prompts to the host of the target live broadcast room according to the interactive response and the audience question, including: Obtain the information display address of the anchor, and obtain the user name corresponding to the audience's question; Displaying audience questions in the first area of ​​the information display address according to the user name and the audience intention; The question reply is displayed in the second area of ​​the information display address according to the interactive reply.

5. The live interactive method based on a large model as claimed in claim 1, It is characterized in that Performing reply detection on the live voice, including: Performing speech recognition on the live broadcast speech to obtain speech text, and performing similarity matching between the speech text and the interactive reply to obtain text similarity; If the text similarity is less than the similarity threshold, it is determined that the reply detection of the live voice is unqualified; If the text similarity is greater than or equal to the similarity threshold, it is determined that the reply detection of the live voice is qualified.

6. The live interactive method based on a large model as claimed in claim 1, It is characterized in that After performing live broadcast interaction in the target live broadcast room according to the interactive reply, the method further includes: If it is detected that the anchor has not performed voice live broadcast within the third preset time period, obtaining the product information of the current live broadcast, and generating the secondary anchor interactive voice according to the product information of the current live broadcast; Live interaction is performed in the target live broadcast room according to the secondary broadcaster's interactive voice until the live broadcast voice of the host is collected.

7. The live interactive method based on a large model as claimed in claim 1, It is characterized in that Inputting the audience's question into the pre-trained intention model for question-answer analysis, before obtaining the audience's intention, further includes: Acquire historical live broadcast question and answer data, and generate question and answer text corpus based on the historical live broadcast question and answer data; Performing data cleaning and data annotation on the question-answer text corpus to obtain model training corpus, and inputting the model training corpus into the intention big model to perform question-answer prediction to obtain a predicted reply; The model loss is calculated according to the predicted response, and the parameters of the intention big model are updated according to the model loss until the intention big model converges to obtain the pre-trained intention big model.

8. A live interactive system based on a large model, It is characterized in that The system comprises: A question-and-answer analysis module is used to obtain audience questions in the target live broadcast room, and input the audience questions into the pre-trained intention model for question-and-answer analysis to obtain audience intentions; An interactive prompt module, for determining an interactive response according to the audience intention if the audience intention meets the live broadcast interactive condition, and providing an interactive prompt to the host of the target live broadcast room according to the interactive response and the audience question; A reply detection module, used to obtain the live broadcast voice of the anchor within a first preset time period, and perform reply detection on the live broadcast voice; The live broadcast interaction module is used to conduct live broadcast interaction in the target live broadcast room according to the interactive reply if the reply detection of the live broadcast voice fails.

9. A terminal device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, It is characterized in that When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program. It is characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Citation Information

Cited By

  • Display method and device of virtual live broadcast room, equipment and storage medium

    CN120075483A

  • Information processing method, system and equipment based on live broadcast platform

    CN120111263A

  • Live streaming interaction method and apparatus, electronic device, and computer-readable storage medium

    WO2026178991A1