Live broadcast script generation method and device, storage medium and program product

Abstract

Embodiments of the present application provide a live script generation method and device, a storage medium and a program product, relating to the technical field of artificial intelligence. The method comprises: in response to triggering of a script generation task, determining target sub-agents that need to be called in sequence in a workflow according to the completeness of first script generation configuration parameter information, wherein if the completeness is insufficient, the target sub-agents include a parameter acquisition agent, an outline generation agent and a script generation agent; inputting second script generation configuration parameter information determined by the parameter acquisition agent into the outline generation agent to obtain a script outline output by the outline generation agent, wherein the second script generation configuration parameter information is a result of parameter completion on the first script generation configuration parameter information; and inputting the script outline and the second script generation configuration parameter information into the script generation agent to obtain a script file output by the script generation agent. The present scheme can generate a high-quality script file.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a method, device, storage medium and program product for generating live streaming scripts. Background Technology

[0002] With the development of technology, the live streaming industry has expanded rapidly, especially e-commerce live streaming, which has become an important way to boost consumption and inventory turnover. A live stream can bring high-frequency orders and category explosion effects in a short period of time, but it requires extremely high quality and efficiency in generating live stream scripts. A live stream script refers to a detailed planning and execution manual designed in advance for a live stream.

[0003] Currently, fixed pipelines are generally used to generate live streaming scripts, meaning the entire processing flow is predefined, like a production line. For example, the script generation process for a live stream might be forcibly defined as "execute steps A, B, C, and D in sequence." Regardless of the user's input, each live streaming script generation task will be executed in this order. While this can automate the generation of live streaming scripts to some extent and improve efficiency, the steps and their execution order in the "fixed pipeline" are "hard-coded." The generated live streaming scripts cannot be dynamically changed according to the actual situation, resulting in poor script quality. Summary of the Invention

[0004] This application provides a method, device, storage medium, and program product for generating live streaming scripts, used to generate high-quality script files.

[0005] In a first aspect, embodiments of this application provide a method for generating live streaming scripts. The method is applied to a main intelligent agent, which is configured with a workflow consisting of multiple sub-intelligent agents. The method includes: In response to the triggering of the script generation task, the target sub-intelligent agents that need to be called sequentially in the workflow are determined based on the completeness of the configuration parameter information generated by the first script. If the completeness is insufficient, the target sub-intelligent agents include a parameter acquisition agent, an outline generation agent, and a script generation agent. The second script generation configuration parameter information determined by the intelligent agent through the parameters is input into the outline generation intelligent agent to obtain the script outline output by the outline generation intelligent agent. The second script generation configuration parameter information is the result of parameter completion of the first script generation configuration parameter information. Input the script outline and the second script generation configuration parameter information into the script generation agent to obtain the script file output by the script generation agent.

[0006] Secondly, embodiments of this application provide a live streaming script generation device. The device is applied to a main intelligent agent, which has a workflow consisting of multiple sub-intelligent agents. The device includes: The determination module is used to respond to the triggering of the script generation task, and determine the target sub-intelligent agents that need to be called sequentially in the workflow based on the completeness of the configuration parameter information generated by the first script. If the completeness is insufficient, the target sub-intelligent agents include the parameter acquisition agent, the outline generation agent, and the script generation agent. The acquisition module is configured to input the second script generation configuration parameter information determined by the acquisition agent through the parameters into the outline generation agent to obtain the script outline output by the outline generation agent, wherein the second script generation configuration parameter information is the result of parameter completion of the first script generation configuration parameter information; and to input the script outline and the second script generation configuration parameter information into the script generation agent to obtain the script file output by the script generation agent.

[0007] Thirdly, embodiments of this application provide an electronic device, including: a memory, a processor, and a communication interface; wherein, the memory stores executable code, and when the executable code is executed by the processor, the processor executes the live script generation method as described in the first aspect.

[0008] Fourthly, embodiments of this application provide a non-transitory machine-readable storage medium storing executable code, which, when executed by a processor of an electronic device, enables the processor to at least implement the live script generation method as described in the first aspect.

[0009] Fifthly, embodiments of this application provide a computer program product, the computer program product including a computer program, which, when executed by a processor, can implement the live script generation method as described in the first aspect.

[0010] In the live script generation scheme provided in this application embodiment, after the main intelligent agent responds to the script generation task triggered by the user, it can detect the completeness of the first script generation configuration parameter information, and then determine the target sub-intelligent agents that need to be called sequentially in the workflow based on the completeness of the first script generation configuration parameter information, so as to complete the script generation task.

[0011] In practical applications, if the configuration parameters for the first script generation are incomplete, the main agent can automatically call the parameter acquisition agent to complete the parameters and obtain the configuration parameters for the second script generation. This avoids the inefficiency caused by forcibly calling multiple sub-agents in the workflow due to incomplete input parameters. After obtaining the configuration parameters for the second script generation, the main agent inputs them into the outline generation agent to obtain the script outline output by the outline generation agent. This is equivalent to obtaining the core skeleton of the live script, laying the foundation for generating a complete script file. Afterward, the main agent only needs to input the script outline and the second script generation configuration parameters into the script generation agent to successfully obtain a complete and accurate script file.

[0012] It should be understood that in the process of generating live streaming scripts using traditional fixed pipelines, if parameters are missing, automatic parameter completion is not possible. Script files generated based on incomplete parameters are bound to be poor in content, inaccurate, and of low quality. However, this application can dynamically call the corresponding sub-agents to automatically complete the parameters based on the completeness of the configuration parameter information generated by the first script. Then, based on the configuration parameter information generated by the second script with completed parameters, a script file is generated, ensuring the accuracy and high quality of the script file. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 A flowchart illustrating a live streaming script generation method provided in this application embodiment; Figure 2 This application provides an schematic diagram illustrating the generation of configuration parameter information using a second script, as shown in an embodiment of this application. Figure 3 This application provides an schematic diagram of the application of a script outline. Figure 4 This application provides an illustration of the application of a script file. Figure 5 Another flowchart of a live script generation method provided in this application embodiment; Figure 6 Another flowchart of a live streaming script generation method provided in this application embodiment; Figure 7 This application provides an schematic diagram of a live streaming script generation system according to an embodiment of the present application. Figure 8 A schematic diagram of a live script generation device provided in an embodiment of this application; Figure 9 This is a schematic diagram of the structure of an electronic device provided in this embodiment. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application. In addition, the timing of the steps in the following method embodiments is only an example and not a strict limitation.

[0016] It should be noted that, in the cases involving user information in the embodiments of this application, the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in the embodiments of this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use, and processing of related data must comply with the relevant laws, regulations, and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse. In addition, the various models involved in this application (including but not limited to large language models or other models) comply with relevant laws and standards.

[0017] Additionally, it should be noted that the various functional models involved in the embodiments of this application can be language models (LM) or multimodal models (MM) based on artificial intelligence. The embodiments of this application do not limit the number of model parameters supported by the model, aiming to meet actual needs. If the model has relatively more parameters, the model will be relatively larger in scale and have relatively better performance; however, it will consume more time and resources during inference or training. If the model has relatively fewer parameters, the model will be relatively smaller in scale, and while meeting performance requirements, it will be more lightweight, consuming less time and resources during inference or training. The aforementioned model can be a deep learning model used to process and generate natural language text or multimodal data, and can be implemented based on a neural network architecture, which can be pre-trained on large amounts of data. In an optional implementation, the aforementioned model can include an encoder, a decoder, a self-attention layer, and a feed-forward neural network, etc. The encoder is mainly used to convert input data (usually in sequence form) into vector representation. This process can capture the semantic features of the input data. The decoder is responsible for converting the intermediate representation generated by the encoder into output data (usually in sequence form). The self-attention layer is a mechanism that allows the model to pay attention to other positions in the sequence to better encode the current position information. The feedforward neural network can perform nonlinear transformations on the output of the self-attention layer to enhance the model's expressive power. All parts work together, enabling the model built on them to perform well in various complex processing tasks, such as natural language processing, computer vision, speech recognition, machine translation, text summarization, and intelligent question answering.

[0018] The following explains the terms or concepts involved in the embodiments of this application: An agent is an intelligent proxy system that uses the language model mentioned above as its core controller. Essentially, it is an agent system that controls the language model to solve user problems. It relies on the language model as its core decision-making and processing unit and can use the logical reasoning ability of the language model to break down the solution steps of the user problem, call tools, and summarize the solution results of multiple steps.

[0019] Currently, fixed pipelines are generally used to generate live streaming scripts. However, the steps in a fixed pipeline and the execution order of each step are "hard-coded," and the live streaming scripts generated cannot be dynamically changed according to the actual situation, resulting in poor quality.

[0020] In view of this, this application provides a live script generation method that solves the above problems through the following approach: A workflow consisting of multiple sub-intelligent agents is set up in the main intelligent agent. After the main intelligent agent responds to a script generation task triggered by the user, it first evaluates the completeness of the first script generation configuration parameter information corresponding to the script generation task to determine whether the initial parameters input by the user are sufficient. Then, based on the evaluation results, it dynamically determines the sub-intelligent agents that need to be called sequentially. This avoids the problem that traditional fixed pipelines cannot dynamically adjust the execution path according to the input. In practical applications, if the initial parameters output by the user (i.e., the first script generation configuration parameter information) are complete, the outline generation intelligent agent and the script generation intelligent agent that need to be called sequentially in the workflow can be directly determined. Conversely, if the first script generation configuration parameter information is incomplete, the parameter acquisition intelligent agent is automatically called to complete the parameters and obtain the second script generation configuration parameter information. This not only avoids the inefficiency caused by forcibly calling multiple sub-intelligent agents in the workflow due to incomplete input parameters, but also ensures the quality of the final script file generated based on the script outline and the second script generation configuration parameter information. It should be understood that traditional fixed pipelines cannot automatically complete missing parameters, and script files generated based on incomplete parameters are bound to be poor in content, inaccurate, and of poor quality.

[0021] Figure 1 This application provides a flowchart of a live streaming script generation method. The method is applied to a main agent, which is configured with a workflow consisting of multiple sub-agents, such as... Figure 1 As shown, the method includes the following steps: 101. In response to the triggering of the script generation task, the main agent determines the target sub-agents that need to be called sequentially in the workflow based on the completeness of the configuration parameter information generated by the first script. If the completeness is insufficient, the target sub-agents include the parameter acquisition agent, the outline generation agent, and the script generation agent.

[0022] 102. The main agent will input the second script generation configuration parameter information determined by the parameter acquisition agent into the outline generation agent to obtain the script outline output by the outline generation agent. The second script generation configuration parameter information is the result of parameter completion of the first script generation configuration parameter information.

[0023] 103. The main agent inputs the script outline and the second script generation configuration parameter information into the script generation agent to obtain the script file output by the script generation agent.

[0024] In practical applications, a user triggers a script generation task, and the main agent acquires first script generation configuration parameter information corresponding to the script generation task (such as product identifier, specific product information, live broadcast guest, live broadcast duration, live broadcast theme, etc.), and determines target sub-agents that need to be called sequentially in the workflow according to the completeness of the first script generation configuration parameter information, specifically: As an implementation, the main agent may input the script generation requirement description information input by the user into a requirement analysis agent, so that the requirement analysis agent parses the script generation requirement description information according to a set parameter set to obtain structured first script generation configuration parameter information, so as to determine the target sub-agents that need to be called sequentially in the workflow according to the completeness of the first script generation configuration parameters; and acquire the first script generation configuration parameter information fed back by the requirement analysis agent and the target sub-agents that need to be called sequentially in the workflow.

[0025] It should be noted that the requirement analysis agent includes a requirement analysis model. In practical applications, a prompt word containing the script generation requirement description information input by the user is input into the requirement analysis model, so that the requirement analysis model can parse the script generation requirement description information according to the set parameter set to obtain structured first script generation configuration parameter information, so as to determine the target sub-agents that need to be called sequentially in the workflow according to the completeness of the first script generation configuration parameters.

[0026] In specific implementation, for example, assume that the script generation requirement description information input by the user is "I want to generate a live broadcast script for Live Room A. The ID of the product sold in Live Room A is a, the invited live broadcast guest is Jia, the broadcast is scheduled to start at 10 a.m., the live broadcast duration is 1 hour, and words such as super and medicinal shall not be used".

[0027] The prompt word may be: "You are an intelligent task parsing and scheduling expert. Please extract key information according to the script generation requirement description information input by the user and generate structured first script generation configuration parameter information. The output shall include two parts: (1) Task list: list the sub-agents involved in the first script generation configuration parameter information, in the format of "task name->[sub-agent name]" connected by arrows to represent the execution flow; if multiple sub-agents are called, sort them by priority.

[0028] (2) Forbidden word update: extract forbidden words from the user's input and list them separated by commas. Requirements: the language shall be concise, only the required content shall be output, no explanations or additional descriptions shall be added. If no forbidden words are mentioned, this part does not need to be displayed. The output format shall be strictly followed, and Chinese punctuation marks shall be used." Following the above example, the structured first script generation configuration parameter information may be: { "Product ID": a, "Live Stream Guest": A, "Live broadcast time": 10:00 AM "Livestream Duration": 1 hour Banned words: "super", "medicinal". } The output of the requirements analysis model includes: The task list is as follows: "Product Data Query -> [Parameter Acquisition Agent] -> Script Outline Generation -> [Outline Generation Agent] -> Script File Generation -> [Script Generation Agent]". This task list reflects the order in which the target sub-agents are called.

[0029] Banned words updated: "super", "medicinal".

[0030] By inputting user-generated script-based requirement descriptions into the requirement analysis agent, the agent parses these descriptions according to a set of parameters. This yields structured first-stage script generation configuration parameters, automatically transforming unstructured, ambiguous user input into clear, executable structured instructions and a dynamic workflow. This lays the foundation for automated execution of subsequent processes. Then, based on the completeness of the first-stage script generation configuration parameters, the target sub-agents to be called sequentially in the workflow are determined. Essentially, the target sub-agents and their calling order are dynamically generated based on the actual situation, offering high flexibility.

[0031] As another implementation method, corresponding rules can be pre-defined, and the target sub-agents that need to be called sequentially in the workflow can be determined based on these rules.

[0032] To make it easier to understand, for example, the rule could be: 1. If the product ID exists, but the specific product information does not exist, then it is necessary to call the parameters to obtain the intelligent agent.

[0033] 2. If the live stream guest exists, but the guest's specific information does not exist, then it is necessary to call the parameters to obtain the intelligent agent.

[0034] 3. If the live stream topic does not exist, the topic setting agent needs to be invoked.

[0035] 4. If the product ID exists, it is necessary to call the outline to generate the intelligent agent and the script to generate the intelligent agent.

[0036] Assuming the configuration parameter information generated by the first script includes: "Product ID: a" and "Live Stream Guest: A", then, based on the specific content of the above rules, it can be determined that: triggering the above rules 1, 2, and 4, the final order in which the target sub-intelligent agents that need to be called in the workflow is "Parameter Acquisition Agent" -> "Outline Generation Agent" -> "Script Generation Agent".

[0037] The target sub-agents that need to be called sequentially in the workflow are determined by pre-defining corresponding rules, which is simple to implement and the rules are transparent.

[0038] During the execution of the above two implementation methods: If the configuration parameter information generated by the first script is complete enough, there is no need to call the parameters of multiple sub-agents to obtain the agent. Instead, the outline generation agent and the script generation agent can be called in sequence to generate the subsequent script files.

[0039] If the completeness of the configuration parameter information generated by the first script is insufficient, the parameter acquisition agent is invoked to determine the configuration parameter information generated by the second script. This information is then used to complete the configuration parameter information generated by the first script. Based on the configuration parameter information generated by the second script, the outline generation agent and the script generation agent are used to generate the subsequent script file. To facilitate understanding, an example is given below. Suppose the script generation task triggered by the user is "I want to generate a live broadcast script for live broadcast room A. The product ID sold in live broadcast room A is 'a', the invited live broadcast guest is 'A', the broadcast is scheduled to start at 10:00 AM, and the broadcast duration is 1 hour." After the main agent determines the configuration parameter information generated by the user through the script generation task (product ID is 'a', live broadcast guest is 'A', the broadcast time is 10:00 AM, and the broadcast duration is 1 hour), it detects that the configuration parameter information generated by the first script is missing "specific product information (such as product details, price, etc.)" and "live broadcast theme," indicating insufficient completeness.

[0040] In this process, for example, regarding the method by which the main agent detects the completeness of the configuration parameter information generated by the first script: One implementation approach is to maintain a standardized parameter template within the main intelligent agent. This template defines the ideal parameters required to generate a high-quality live streaming script. In practice, after obtaining the first script generation configuration parameter information, the parameters in this information are compared with the ideal parameters in the template. If all parameters in the template are present in the first script generation configuration parameter information, then the first script generation configuration parameter information is considered sufficiently complete. Conversely, if some parameters in the template are not mentioned in the first script generation configuration parameter information, then the completeness of the first script generation configuration parameter information is considered insufficient.

[0041] As another implementation approach, since different script types have different parameter requirements, the main agent can first determine the required script type based on the script generation task triggered by the user, and then check whether the key parameters on which the script type depends are complete. For example, suppose the script generation task triggered by the user is "generate a live stream script, with relevant information including: product ID is 'a', expected to start at 10:00 AM, and live stream duration of 1 hour". Based on this script generation task, the user's script type can be determined (let's say "product sales type"). Then, by checking the key parameters related to the product sales type, it is found that the first script generation configuration parameter information (product ID is 'a', live stream time is 10:00 AM, live stream duration is 1 hour) lacks "specific product information". At this point, it can be determined that the first script generation configuration parameter information is incomplete. Conversely, if the user's input includes key parameters related to the product sales type, then the first script generation configuration parameter information is considered sufficiently complete.

[0042] When the completeness of the first script generation configuration parameter information is determined to be insufficient through the above completeness detection method, the main agent will input the second script generation configuration parameter information determined by the parameter acquisition agent into the outline generation agent to obtain the script outline output by the outline generation agent, specifically: As one implementation method, the main agent determines the script type based on the configuration parameter information generated by the second script; the script type and the configuration parameter information generated by the second script are input into the outline generation agent to obtain the script outline corresponding to the script type output by the outline generation agent.

[0043] In practical applications, the main agent can determine the script type by generating configuration parameter information based on a second script according to predefined rules. For example, the preset rule could be: "1. If there is relevant information about the live stream guests, and the information about the live stream guests is rich in content, then the script type is determined to be guest interaction type."

[0044] 2. If there are multiple product IDs and no information about the live stream guests, then the script type is determined to be product introduction type.

[0045] 3. If information related to "promotion" exists and the live stream topic contains holiday keywords, then the script type is determined to be a holiday promotion type, etc.

[0046] Assume the second script generation configuration parameter information obtained so far includes: { The "Product List" contains the following product IDs: a, b, c, d, e, etc. The products corresponding to these product IDs are all infant and toddler products, such as milk powder, calcium tablets, biscuits, toys, etc., with prices ranging from 50 yuan to 500 yuan.

[0047] "Live Stream Guest": A, "Live broadcast time": 10:00 AM "Livestream Duration": 1 hour "Live Stream Theme": Featured Product Recommendations Banned words: "super", "medicinal". } Following the example above, the configuration parameter information generated by the second script, compared to that generated by the first script, completes the "specific information of the product" and the "live broadcast theme". When determining the script type, substituting the configuration parameter information generated by the second script into the aforementioned preset rules, it can be determined that: the configuration parameter information generated by the second script conforms to rule 2, that is, the configuration parameter information generated by the second script contains multiple product IDs and no information related to the live broadcast guests, thus confirming that the script type is a product introduction type.

[0048] It should be noted that, to facilitate users' viewing and modification of configuration parameters, the configuration parameters generated by the second script can be displayed to users in a structured format. For details, please refer to [link / reference needed]. Figure 2 .exist Figure 2 The following parameter information is illustrated (for illustrative purposes only and is not intended to be limiting): Livestream Event Title: Win Great Prizes for Mid-Autumn Festival; Live broadcast time: October 11, 2025; Live broadcast time: 12:00-13:00; Script type: Guest interaction type; Host: A; Live Stream Guest: B; Live Stream Theme: Celebrating the Mid-Autumn Festival; Product List: Product ID: a {Product Name: Milk Powder}, {Live Stream Price: 339}, {Market Retail Price: 376}; Product ID: b {Product Name: Calcium Tablets}, {Live Stream Price: 338}, {Market Retail Price: 376}; Product ID: c {Product Name: Biscuits}, {Live Stream Price: 70}, {Market Retail Price: 100}.

[0049] After determining the script type, the script type and the second script generation configuration parameters can be input into the outline generation agent to obtain the script outline corresponding to the script type, output by the outline generation agent. It should be noted that the outline generation agent contains an outline generation model. In practice, prompts containing the script type and the second script generation configuration parameters can be input into the outline generation model to cause it to output the corresponding script outline.

[0050] To facilitate understanding, the following example illustrates this: the prompt containing information about the script type and the second script generation configuration parameters can be: "You are a professional live stream script planner. Please generate a script outline for a 'product introduction' live stream:" [Basic Live Streaming Information (i.e., configuration parameter information generated by the second script)]: The "Product List" contains the following product IDs: a, b, c, d, e, etc. The products corresponding to these product IDs are all infant and toddler products, such as milk powder, calcium tablets, biscuits, toys, etc., with prices ranging from 50 yuan to 500 yuan.

[0051] Live broadcast time: 12:00-13:00 "Livestream Duration": 1 hour "Live Stream Theme": Featured Product Recommendations Banned words: "super", "medicinal". [Task Requirements]: Please generate a concise and clear script outline. Note: 1. Formatting requirements (must be strictly followed): - Format for each stage: Stage Name (Start Time - End Time) - The time must be accurate to the minute, such as (0-5 minutes), (5-15 minutes). - Total duration must equal live stream duration 2. Content Requirements (must be strictly followed): - Keep it simple - Avoid excessive detail - Highlight key points - Highlight key points 3. Output requirements (must be strictly followed): - Output a concise outline of the live stream script, without any additional opening remarks, closing remarks, or explanatory text.

[0052] Please output the script outline.

[0053] Figure 3 This is a schematic diagram illustrating the application of a script outline provided in an embodiment of this application. Figure 3 The document contains four time periods: Time Segment 1 (Opening): 12:00-12:05. During this time, host A warmly welcomes the viewers, introduces the theme of the live stream, highlights the featured products, and emphasizes their quality and special offers.

[0054] Time Period 2 (Product Introduction): 12:05-12:35. During this time period, host A will provide an in-depth explanation of the product's functions.

[0055] Time Segment 3 (Interactive): 12:35-12:55. During this time period, host A interacts with the audience and answers questions from customers.

[0056] Time Period 4 (Summary): 12:55-13:00. During this time, host A thanks the viewers again for their support, emphasizes today's special offers, encourages final purchases, and bids farewell to the viewers.

[0057] Based on the above, it's important to note that different types of live streams have completely different rhythms and focuses (for example, holiday promotional live streams require dense promotional moments and a sense of urgency, while guest interaction live streams require ample space for dialogue, etc.). Determining the script type through the second script generation configuration parameters, and then inputting the script type and the second script generation configuration parameters into the outline generation agent to obtain the script outline, can improve the accuracy of the generated script outline. Furthermore, the main agent can automatically adjust the script outline according to different script types, laying the foundation for generating complete and accurate script files subsequently.

[0058] As an alternative implementation, the main agent can input the second script generation configuration parameter information into the outline generation agent to obtain the script outline output by the outline generation agent. This method can directly obtain the script outline based on the second script generation configuration parameter information without considering the "script type". It should be noted that the outline generation agent contains an outline generation model. In specific implementation, a prompt word containing the second script generation configuration parameter information can be input into the outline generation model to make the outline generation model output the corresponding script outline. The appearance of this prompt word can refer to the prompt word shown in the previous implementation method (simply delete the "script type" part), which will not be elaborated here.

[0059] It should be noted that, during the process of inputting the second script generation configuration parameter information into the outline generation agent based on the above two implementation methods, if the first script generation configuration parameter information lacks target product attribute information and / or live streamer introduction information, then the target product identifier and / or live streamer identifier corresponding to the script generation task are input into the data collection agent to obtain the target product attribute information and / or live streamer introduction information output by the data collection agent after searching in the knowledge base; if the first script generation configuration parameter information lacks a live stream topic, then the first script generation configuration parameter information is input into the topic setting agent so that the topic setting agent can perform parameter analysis on the first script generation configuration parameter information and generate a live stream topic; the second script generation configuration parameter information containing target product attribute information and live stream topic is input into the outline generation agent.

[0060] In practical applications, if the first script generates configuration parameters that lack target product attribute information (such as product ID, product name, etc.) and / or live streamer introduction information (host's name and related introduction information, live streamer's name and related introduction letter), the main intelligent agent can input the target product identifier and / or live streamer identifier corresponding to the script generation task into the data collection intelligent agent. It should be understood that the data collection intelligent agent includes a data collection model. In specific implementation, inputting prompts containing the target product identifier and / or live streamer identifier into the data collection model will retrieve the target product attribute information and / or live streamer introduction information from the knowledge base.

[0061] To facilitate understanding, the following example illustrates this: the prompt containing the target product identifier and / or the live streamer identifier can be: "You are an intelligent information retrieval assistant. You need to determine which retrieval tool to use to obtain information based on the user's needs."

[0062] Available search tools: 1. Tool A: xxxx; 2. Tool B: xxxx; 3. Tool C: xxxx; Please determine what you need to search for based on the following information: Target product identifier: xxx; Live streamer identifier: xxx; Please analyze the above information and then determine which tools to use for the search: 1. If only the target product identifier is available, use tool A and tool B to search for target product attribute information related to that target product identifier.

[0063] 2. If only the streamer's identifier is available, use tool C to search for information about the streamer related to that identifier.

[0064] 3. If there are both target product tags and live streamer tags, then tools A, B, and C can be used simultaneously.

[0065] Please begin your search and obtain relevant information.

[0066] It should be noted that the data collection model can perform multi-source searches in the knowledge base based on different retrieval tools. For example, it can retrieve specific information about a product, such as its name and price, from the product knowledge graph; retrieve historical scripts for the product from the historical script database; retrieve relevant reviews of the product from external databases; and retrieve the background, achievements, and livestreaming style of the livestreamers from the personnel database. When the target product attribute information and / or the livestreamer's introduction information are missing from the first script generation configuration parameters, the data collection agent can quickly and accurately obtain these information by searching the knowledge base, laying the foundation for subsequent script outline generation.

[0067] If the first script generation configuration parameter information is missing a live broadcast topic, then the first script generation configuration parameter information is input into the topic setting agent so that the topic setting agent can perform parameter analysis on the first script generation configuration parameter information and generate a live broadcast topic; the second script generation configuration parameter information, which includes the target product attribute information and the live broadcast topic, is input into the outline generation agent.

[0068] Similarly, if the configuration parameter information generated by the first script lacks a live broadcast topic, the main agent can input the configuration parameter information generated by the first script into the topic setting agent. It should be understood that the topic setting agent includes a topic setting model. In practice, inputting prompts containing the configuration parameter information generated by the first script into the topic setting model will yield the live broadcast topic output by the topic setting model after searching the knowledge base.

[0069] To facilitate understanding, the following example illustrates this: the prompt containing the configuration parameter information generated by the first script can be: "You are a live streaming theme setting expert. You need to determine the live streaming theme based on the parameter information provided by the user."

[0070] The parameter information provided by the user (i.e., the configuration parameter information generated by the first script) is as follows: { "Product ID": a, "Live broadcast time": 10:00 AM "Livestream Duration": 1 hour Banned words: "super", "medicinal". } Please begin setting the live stream topic.

[0071] By using a topic setting agent to determine the live stream topic when the configuration parameter information generated in the first script is missing, the "live stream topic" information is completed, laying the foundation for generating a more complete script outline in the future.

[0072] After obtaining the target product attribute information and the live broadcast theme, input the second script generation configuration parameter information containing the target product attribute information and the live broadcast theme into the outline generation agent to obtain the script outline.

[0073] Next, the main agent inputs the script outline and the second script generation configuration parameters into the script generation agent to obtain the script file output by the script generation agent, specifically: As one implementation method, if the second script generation configuration parameter information indicates script file format reference acquisition information, then the script file format reference information is obtained according to the script file format reference acquisition information; the script outline, the second script generation configuration parameter information, and the script file format reference information are input into the script generation agent to obtain the script file output by the script generation agent.

[0074] In practical applications, after obtaining the configuration parameter information for the second script generation, it can be determined whether there is script file format reference information in the configuration parameter information for the second script generation (i.e., whether the user has a description similar to "refer to historical materials"). If so (for example, the user indicates: "refer to the script file from the last Mid-Autumn Festival", "refer to the script file from the same type of product", etc.), then the relevant historical live broadcast materials can be located in the knowledge base. These historical live broadcast materials can be in document format, table format, image format, etc., which are not limited here.

[0075] In practice, taking the historical live broadcast materials in table format as an example, a multimodal model can be used to parse the historical live broadcast materials and extract the column structure from them, such as "serial number", "product", "stage", "time", "specific content", "remarks", etc. The column result is then standardized into "script file format reference information", which is a structured format description.

[0076] Next, the script outline (providing the skeleton and rhythm of the live stream content), the second script generation configuration parameter information (providing details such as the specific product content, live stream time, and live stream duration), and the script file format reference information (i.e., the column structure mentioned above, providing the final script file presentation structure) are input into the script generation agent to obtain the script file output by the script generation agent. It should be noted that the script generation agent contains a script generation model. In specific implementation, prompts containing the script outline, second script generation configuration parameter information, and script file format reference information can be input into the outline generation model to cause the outline generation model to output the corresponding script file.

[0077] To facilitate understanding, the following example illustrates this: the prompt containing the script outline, second script generation configuration parameters, and script file format reference information can be: "You are a professional live streaming script writing expert. Please generate an executable script file based on the following information:" [Basic Live Streaming Information (i.e., configuration parameter information generated by the second script)]: The "Product List" contains the following product IDs: a, b, c, d, e, etc. The products corresponding to these product IDs are all infant and toddler products, such as milk powder, calcium tablets, biscuits, toys, etc., with prices ranging from 50 yuan to 500 yuan.

[0078] Live broadcast time: 12:00-13:00 "Livestream Duration": 1 hour "Live Stream Theme": Featured Product Recommendations Banned words: "super", "medicinal". [Script Outline]: xxx (i.e., the content of the script outline obtained in the above embodiment).

[0079] [Task Requirements]: Please strictly follow the following table structure when generating the script, which must include the following: {"Serial Number", "Product", "Stage", "Time", "Specific Details", "Remarks"} Please output the script file.

[0080] Figure 4 This is a schematic diagram illustrating the application of a script file provided in an embodiment of this application. Figure 4 The document includes details such as "Live Stream Theme: Celebrating Mid-Autumn Festival", "Live Stream Time: 12:00-13:00", "1 hour", "Host: A", as well as specific information such as "Serial Number", "Products", "Segment", "Time", "Specific Content", and "Remarks".

[0081] Based on the above, if the second script generation configuration parameter information indicates script file format reference information, then the script file format reference information is obtained according to the script file format reference information, realizing true "out-of-the-box use". The generated script file no longer requires staff to manually adjust the format or reformat it. By using the script file format reference information (the format of historically successful script files), not only can the consistency and professionalism of brand output be ensured, but work efficiency can also be improved.

[0082] As another implementation method, a series of script templates and filling rules can be predefined for each stage of the script outline (such as "opening interaction", "product explanation", and "sales promotion and closing"). The script generation agent selects the corresponding template according to each stage of the script outline and fills in the specific content (such as product information, live broadcast time, etc.) in the second script generation configuration parameter information to obtain the final script file.

[0083] Based on the above, the live script generation method provided in this application embodiment can detect the completeness of the first script generation configuration parameter information after the main intelligent agent responds to the script generation task triggered by the user. If the completeness of the first script generation configuration parameter information is insufficient, the parameter acquisition intelligent agent is called to complete the parameters and obtain the second script generation configuration parameter information. Then, a script file is generated based on the second script generation configuration parameter information after parameter completion, which ensures the accuracy and high quality of the script file.

[0084] Figure 5 Another flowchart of a live script generation method provided in this application embodiment is shown below. Figure 5 As shown, the method includes the following steps: 501. In response to the triggering of the script generation task, the main agent determines the target sub-agents that need to be called sequentially in the workflow based on the completeness of the configuration parameter information generated by the first script. If the completeness is insufficient, the target sub-agents include the parameter acquisition agent, the outline generation agent, and the script generation agent.

[0085] 502. The main agent will input the second script generation configuration parameter information determined by the parameter acquisition agent into the outline generation agent to obtain the script outline output by the outline generation agent. The second script generation configuration parameter information is the result of parameter completion of the first script generation configuration parameter information.

[0086] 503. The main agent sends the script outline to the user for review.

[0087] 504. The main agent receives the review results from the user. If the review result is "not approved", it updates the second script generation configuration parameter information based on the review result and re-enters the updated second script generation configuration parameter information into the outline generation agent to obtain a new script outline.

[0088] 505. The main agent inputs the new script outline and the second script generation configuration parameter information into the script generation agent to obtain the script file output by the script generation agent.

[0089] For the specific execution process of steps 501, 502 and 505, please refer to the above embodiments, which will not be repeated here.

[0090] For steps 503 and 504, it should be understood that the script outline is the skeleton of the script file. Modifying the script outline is far less costly than modifying the complete script file. Therefore, in this embodiment, after receiving the script outline output by the outline generation agent, the main agent can send the script outline to the user for review and receive the user's review results. If the review result is positive, the subsequent script file generation operation can continue based on the original script outline. If the review result is negative, as one implementation method, the second script generation configuration parameter information can be updated based on the review result. For example, assuming the original second script generation configuration parameter information includes a live stream theme of "Featured Product Recommendations," and the user's review comments indicate that the live stream theme is not approved and needs to be changed to "Celebrating Mid-Autumn Festival," the main agent will use "Celebrating Mid-Autumn Festival" to update the content of the live stream theme in the original second script generation configuration parameter information, obtaining the updated second script generation configuration parameter information. Afterwards, it reverts to the outline generation agent, re-inputs the updated second script generation configuration parameter information into the outline generation agent, obtains a new script outline, and then uses the new script outline to generate the script file. Alternatively, the audit result can include a script outline that the user has directly modified. The original script outline can be updated using the user-modified script outline to obtain a new script outline, which can then be used to generate script files.

[0091] By manually reviewing the generated script outline and updating the second script generation configuration parameters if the user feedback indicates a failure, and then re-inputting the updated second script generation configuration parameters into the outline generation agent to obtain a new script outline, we can, on the one hand, allow users to intervene manually before the official script file is generated, reducing the cost of modifying the complete script file later. On the other hand, it should be understood that users may fail the review and regenerate the script outline more than once. Each "failure" and subsequent regeneration of the script outline is actually a high-quality learning sample. Recording all this data for subsequent training of the outline generation model contained in the outline generation agent allows it to better understand the user's potential preferences, obtain a script outline that better meets the user's needs, reduce the number of times the user modifies the script file later, and improve the efficiency of script file generation.

[0092] Figure 6 Another flowchart of a live script generation method provided in this application embodiment is shown below. Figure 6 As shown, the method includes the following steps: 601. In response to the triggering of the script generation task, the main intelligent agent determines the target sub-intelligent agents that need to be called sequentially in the workflow based on the completeness of the configuration parameter information generated by the first script. If the completeness is insufficient, the target sub-intelligent agents include the parameter acquisition intelligent agent, the outline generation intelligent agent, and the script generation intelligent agent.

[0093] The target sub-agent also includes a compliance detection agent.

[0094] 602. The main agent will input the second script generation configuration parameter information determined by the parameter acquisition agent into the outline generation agent to obtain the script outline output by the outline generation agent. The second script generation configuration parameter information is the result of parameter completion of the first script generation configuration parameter information.

[0095] 603. The main agent inputs the script outline and the second script generation configuration parameter information into the script generation agent to obtain the script file output by the script generation agent.

[0096] 604. The main intelligent agent performs multi-dimensional detection on the script file through the compliance detection intelligent agent to obtain multi-dimensional detection results. The multi-dimensional detection results include: user-defined prohibited words, prohibited words defined by the live streaming platform, and preset general detection items.

[0097] 605. If the test result indicates that the preset general test item failed the test, the script generation task will be re-executed.

[0098] 606. If the detection result indicates that the script file contains prohibited words, then perform semantic replacement on the prohibited words in the script file.

[0099] For the specific execution process of steps 601-603, please refer to the above embodiments, which will not be repeated here.

[0100] For steps 604-606, the target sub-agent also includes a compliance detection agent, which contains a compliance detection model. In practical applications, the main agent inputs prompts, including script files, into the compliance detection model. The model then performs multi-dimensional detection on the script files to obtain the results. These multiple dimensions include: user-defined prohibited words, prohibited words defined by the live streaming platform, and preset general detection items. User-defined prohibited words can be any user-defined prohibited terms such as "super" or "medicinal," and are not limited here. Prohibited words defined by the live streaming platform can include false promises, which are not listed here. The preset general detection items include serious violations, including some sensitive words.

[0101] In practice, for ease of understanding, the following examples illustrate the prompts that include script files: "You are a compliance testing expert in the field of live e-commerce scripts, possessing a complete database of prohibited words related to live streaming. Your core task is to receive an original live streaming script file, replace all non-compliant content in place, and finally output an optimized script file with the same format and completely compliant content to the client."

[0102] Maintaining the core formatting principle (highest priority): You must strictly adhere to the following formatting guidelines: 1. Maintain the original structure.

[0103] 2. Targeted replacement.

[0104] 3. Minimize changes.

[0105] 4. Maintain the special format.

[0106] Core Revision Principles (Complete Knowledge Base) While adhering to the above core formatting guidelines, you must rigorously review and replace all prohibited words in the following categories. This section lists the prohibited words that need to be replaced in each live stream as defined by the service domain. Due to the involvement of service domain data, only a simple example is provided, such as: medical terminology and efficacy claims (for general food / non-special cosmetics). The use of any terms related to the treatment, diagnosis, prevention of diseases, or specific physiological effects is strictly prohibited.

[0107] Output format requirements Strictly adhere to the following output format; do not add any additional descriptions or explanations: Output the content of the compliant and corrected script file directly, keeping the original format and without adding any opening remarks, closing remarks, or explanatory text.

[0108] Task Instructions Now, please strictly adhere to the above format, maintain the core guidelines and complete the core correction principles, and perform compliance correction on the following provided script files: Script file: xxx (the specific content of the script file) It should be noted that the compliance detection model can either perform compliance checks on the script file alone, or it can correct the script file simultaneously with the compliance check; this is not limited here. In practical applications, if the detection result indicates that a preset general detection item has failed the test, the script generation task will be re-executed. For example, suppose the script file contains the following description: "This herbal foot bath pack uses a warming and dispersing formula. Soaking in it before bed can cure cold legs." However, the compliance detection model finds that the foot bath pack is a general health care product and should strictly prohibit mentioning any disease treatment or prevention effects in the wording. "Cure cold legs" violates the requirements of the preset general detection item. Therefore, the script generation task needs to be re-executed to regenerate this part of the script file. For example, the above description can be modified to: "This herbal foot bath pack uses a warming and dispersing formula. Soaking in it before bed can help relieve cold legs, promote blood circulation, and make your body feel more relaxed and comfortable."

[0109] Additionally, if the detection results indicate that the script file contains prohibited words, then the prohibited words in the script file will be semantically replaced. For example, suppose the script file contains the following description: "This herbal foot bath pack uses a formula that warms the meridians and dispels cold. Soaking in it before bed can help relieve cold legs and promote blood circulation. It's super powerful." However, after the compliance detection model detects that "super powerful" is a user-defined prohibited word and / or a prohibited word defined by the live streaming platform, then the prohibited word in the script file will be semantically replaced. For example, "super powerful" will be replaced with "good experience" or "good comfort."

[0110] By using a compliance detection agent to perform multi-dimensional checks on script files and making corresponding modifications based on the detection results, the automatic processing of illegal or prohibited words is achieved. This ensures the compliance of script files while preserving their fluency and attractiveness to the greatest extent possible.

[0111] In addition, the target sub-agent also includes an optimization agent. In this embodiment, after obtaining the second script generation configuration parameter information, the historical user interaction data of the historical script file matching the script generation task can be determined based on the second script generation configuration parameter information; the historical user interaction data is input into the optimization agent so that the optimization agent can estimate the interaction effect index and user question data corresponding to the script file, and optimize the script file based on the interaction effect index and user question data.

[0112] In practical applications, after obtaining the configuration parameter information for the second script generation, the main agent can filter according to set rules in a knowledge base that records historical script files. For example, it can perform feature similarity matching to determine the historical live broadcast with the highest feature similarity and extract their historical user interaction data.

[0113] To make it easier to understand, let's take an example and assume that the configuration parameter information generated by the second script includes the following: { Product Name: XX Model Mobile Phone; Product category: Mobile phones; Price: 3999 yuan; Target audience: Young people aged 18-30, photography enthusiasts; } When performing the above feature matching, feature information can be extracted from the configuration parameter information generated by the second script, specifically including: "Product Category: Mobile Phone", "Target Audience: Young people aged 18-30, photography enthusiasts", and "Price: 3999 yuan". In practice, the knowledge base can be searched for one or more historical live streams that are also in the "Mobile Phone" category, target "young people", are priced between "3000 yuan and 5000 yuan", and have "photography" as their core selling point. Their historical user interaction data can be extracted, such as "Is the photo really good?", "Is it a big upgrade compared to the previous generation?", "How long after placing an order will it be shipped?", etc.

[0114] Next, the obtained historical user interaction data is input into the optimization agent. It should be noted that the optimization agent contains an optimization model. In specific implementation, the prompt words containing historical user interaction data are input into the optimization model, which enables the optimization model to predict the interaction effect indicators and user question data corresponding to the script file, and optimize the script file based on the interaction effect indicators and user question data.

[0115] To make it easier to understand, the following are examples of prompts that contain historical user interaction data: "You are a script optimization expert in the live streaming field. Please complete the following task based on the following input information:" Script file: xxx (i.e., the specific content of the script file).

[0116] Historical user interaction data: Frequently asked questions by users: "Does it really take good photos?", "Is it a big upgrade compared to the previous generation?", "How long after placing an order will it be shipped?"

[0117] Optimize tasks: 1. Predict the key metrics for this live stream (watching time, interaction rate, etc.).

[0118] 2. Predict 5 questions that users may ask, with the following requirements: close to the language style of real users; covering different products or scenarios; and including potential doubts or points of hesitation.

[0119] 3. Based on the above analysis, specific optimization suggestions for the script files are proposed, and the script files are optimized based on these suggestions.

[0120] Output optimized script file " After inputting the aforementioned prompts into the optimization model, the optimization script file will be obtained. In practice, the optimization agent can either directly output the optimized script file, or simply output optimization suggestions and package them into an optimization suggestion package, which will then be executed by the main agent or the script generation agent; no limitation is imposed here.

[0121] To facilitate understanding, the optimization suggestions are illustrated below with examples, such as: 1. A "Night Scene Shooting Comparison Demonstration" segment has been added at the 15-20 minute mark.

[0122] 2. The sales pitch incorporates an affirmative explanation of "natural skin tone in the portrait" in advance.

[0123] 3. In the Q&A session, standard answers to the above 5 predicted questions were prepared in advance.

[0124] 4. The promotional information clearly included the offer of "12-month interest-free installments for a limited time".

[0125] By identifying historical user interaction data and inputting it into an optimization agent, the agent can predict the interaction performance metrics and user questions associated with the script file. Based on these metrics, the agent optimizes the script file, fundamentally changing the traditional model of script optimization relying on personal experience. This automates script optimization and significantly improves efficiency. Furthermore, by pre-generating data on potential user questions and preparing answers, the host can handle these questions with ease and professionalism during the live stream, greatly enhancing user trust and engagement, and ultimately increasing order conversion rates.

[0126] Furthermore, the target sub-agent also includes an output integration agent. In practical applications, the script file and the configuration parameters generated by the second script can be input into the output integration agent. The output integration agent then integrates the input content to output the final complete target live streaming script. This process is equivalent to merging the script file with the configuration parameters generated by the second script, ensuring that users can view relevant information from the second script's configuration parameters within the target live streaming script, thus improving the user experience.

[0127] In addition, it should be noted that the main intelligent agent can acquire user interaction data during the execution of the script file, and add the script file to the knowledge base when it is determined that the script file meets the preset excellent script selection criteria based on the user interaction data.

[0128] In practical applications, during live streaming using the script file generated in this application embodiment, a real-time monitoring phase is entered. The main intelligent agent can acquire user interaction data in real time based on a set data monitoring tool, such as bullet screen content, number of likes, number of shares, comments, dwell time, conversion rate, etc. After the live stream ends, the script can be judged as a worthy excellent script based on preset, quantifiable indicators (i.e., preset excellent script selection criteria). For example, these preset excellent script selection criteria could be "average viewing time greater than 20 minutes," "order conversion rate greater than 60%," "negative bullet screen ratio less than 5%," etc., which will not be listed here. When it is determined that the script file meets the above preset excellent script selection criteria, the script file can be added to the knowledge base.

[0129] By continuously adding script files that meet the preset criteria for excellent scripts to the knowledge base, the knowledge base can be continuously updated and optimized to keep up with trends and hot topics, laying the foundation for generating high-quality script files in the future.

[0130] Furthermore, in this embodiment of the application, the live script generation method further includes: in response to a checkpoint operation set for at least some of the sub-intelligent agents, obtaining the state information of at least some of the sub-intelligent agents and storing the state information in a target storage space; if it is determined based on the state information that there is an abnormal intelligent agent among at least some of the sub-intelligent agents that is running abnormally, then re-triggering the operation of the abnormal intelligent agent.

[0131] In practical applications, within a workflow involving multiple sub-agents, the master agent sets checkpoints at key nodes (e.g., after a sub-agent completes its work). Responding to checkpoint operations set for at least some of the sub-agents, the master agent can acquire the state information of at least some of the sub-agents, such as inputs, outputs, intermediate results, and context variables, and store this state information in the target storage space. If the master agent determines, based on the state information, that at least some of the sub-agents are experiencing operational anomalies (i.e., the agent has failed to run, possibly due to network problems, service unavailability, internal errors, etc.), it can resume from the last successfully run checkpoint, i.e., re-trigger the execution of the abnormal agent, without needing to start the entire process from scratch.

[0132] To facilitate understanding, let's take an example. Suppose the current script generation task flow includes: Requirements Analysis Agent (A) → Data Collection Agent (B) → Outline Generation Agent (C) → Script Generation Agent (D). Assume the current scenario is that after the Requirements Analysis Agent (A), Data Collection Agent (B), and Outline Generation Agent (C) have completed their runs, checkpoints have been set, but the Script Generation Agent (D) suddenly fails.

[0133] In practice, after the requirement analysis agent (A), data collection agent (B), and outline generation agent (C) have successfully run, their state information has been saved to the target storage space. At this point, if an abnormality is detected in the script generation agent (D) (e.g., a timeout is detected, or an error code is returned), it is not necessary to re-execute the above process starting from the requirement analysis agent (A). Instead, it is only necessary to resume from the previous checkpoint of the script generation agent (D) (i.e., the position after the outline generation agent (C) has finished running), re-trigger the script generation agent (D), and re-pass the output of the outline generation agent (C) and the necessary context information back to the script generation agent (D).

[0134] Understandably, in complex distributed systems and long-running tasks, partial failures are unavoidable. By using the aforementioned "checkpoint" mechanism, it is ensured that no single point of failure will cause the entire task to be lost or need to be completely restarted, which greatly improves the robustness and reliability of the system. Furthermore, by setting checkpoints, the time and computational costs incurred in starting the entire process from scratch can also be saved.

[0135] To facilitate understanding of this solution, the following will be combined with... Figure 7 Here are some specific examples: Figure 7 This is a schematic diagram illustrating the application of a live streaming script generation system provided in an embodiment of this application. Figure 7 The system includes: a data input module, an intelligent agent collaboration module, a knowledge hub module, a model service module, a manual approval module, a monitoring module, and a persistent storage module.

[0136] The data input module includes: 1. Operational configuration parameters: The parameters of the live broadcast session filled in in advance by the operators or merchants, such as the live broadcast theme, the list of main products, the style of the host (such as humor, calmness, etc.), the duration of the live broadcast, the order conversion indicators, and the restrictions on the wording (i.e. some words that cannot be said in the live broadcast).

[0137] 2. Script material data: Specific information about the product (such as inventory, price, specifications, etc.), product images or videos, historical live broadcast scripts / recorded clips, brand wording guidelines, third-party reviews and compliance evidence materials, etc.

[0138] 3. Behavioral profiling data: User profile data, historical transaction / play / bullet screen / like / order time sequence data, etc.

[0139] The intelligent agent collaboration module includes a main intelligent agent and multiple sub-intelligent agents. The multiple sub-intelligent agents include: a requirements analysis intelligent agent, a data collection intelligent agent, a topic setting intelligent agent, an outline generation intelligent agent, a script generation intelligent agent, a compliance detection intelligent agent, an optimization intelligent agent, and an output integration intelligent agent.

[0140] Among them, the main intelligent agent can follow Figure 7 The system sequentially mobilizes multiple sub-agents and is responsible for persistently storing checkpoint, retry, and audit log records. These checkpoints support interruption recovery, audit backtracking, and manual approval callbacks.

[0141] The demand analysis agent is used to parse all the data input from the data input module and generate a structured second script to generate configuration parameter information.

[0142] The data collection agent is used to retrieve product knowledge graph fragments, historical best-selling sales phrases, relevant image / video frames, user behavior samples, and external rule documents as contextual evidence for subsequent generation.

[0143] The topic setting agent is used to analyze the input parameters and generate live broadcast topics.

[0144] The purposes and specific execution processes of the outline generation agent, script generation agent, compliance detection agent, optimization agent, output integration agent, and monitoring module can be found in the above embodiments and will not be repeated here. It should be noted that during the process of the outline generation agent generating the script outline, if it is impossible to accurately generate a unique script outline based on the analysis of the requirements analysis agent, multiple candidate versions can be generated. For example, assuming the main agent cannot determine whether the live stream topic is "product introduction" or "guest interaction" based on the information from the requirements analysis agent, it can generate one script outline containing the live stream topic of "product introduction" and another containing the live stream topic of "guest interaction." Furthermore, after the compliance detection agent outputs the detection results, manual review can be performed, and subsequent operations can continue if the manual review is passed.

[0145] The knowledge hub module includes: a product knowledge graph, a script template library, a custom specification library, a sales script template library, a brand sales script specification library, a compliance rule library, and a historical script and conversion performance database.

[0146] The model service module includes: cloud / private model clusters, industry-fine-tuned models, visual / audio encoders, and a multimodal vector storage and retrieval module. This module provides conditional generation capabilities for content creation, such as generating content plans and complete scripts; it also supports content detection and replacement, such as compliance checks and banned word replacement; it provides multimodal embedding (images / video frames / bullet screen text), parsing multimodal data such as images, beverages, and videos as data for content generation; and it supports cloud-based or private inference to meet latency and privacy requirements.

[0147] Manual approval module: Supports quick approval (quick pass / rejection and replacement), style fine-tuning (modification of wording details), and provides a visual script editor, risk annotation panel, comparison of differences between different versions of the script outline, and one-click approval / rollback buttons.

[0148] The persistent storage module includes: saving intermediate states of the checkpoint mechanism, handling compliance failures of conditional branches, and complete recording of audit logs, etc.

[0149] Based on the above, this solution can break down complex script generation tasks into multiple sub-agents, achieving parallel processing, conditional branching, breakpoint recovery, and flexible human intervention for modification and approval through a checkpoint mechanism. Secondly, by deeply integrating product knowledge graphs, script template libraries, and compliance rules, knowledge augmentation is injected during generation to ensure the correctness and professionalism of each language model. Furthermore, by combining user interaction data, automatic script fine-tuning and strategy correction based on real-time feedback can be achieved.

[0150] The following will describe in detail one or more embodiments of the live script generation apparatus of this application. Those skilled in the art will understand that these apparatuses can all be configured using commercially available hardware components through the steps taught in this solution.

[0151] Figure 8 This is a schematic diagram of the structure of a live script generation device provided in an embodiment of this application, as shown below. Figure 8 As shown, the device includes a determining module 11 and an acquiring module 12.

[0152] The determination module 11 is used to respond to the triggering of the script generation task, and determine the target sub-intelligent agents that need to be called sequentially in the workflow based on the completeness of the configuration parameter information of the first script generation. If the completeness is insufficient, the target sub-intelligent agents include the parameter acquisition intelligent agent, the outline generation intelligent agent and the script generation intelligent agent. The acquisition module 12 is used to input the second script generation configuration parameter information determined by the parameter acquisition agent into the outline generation agent to obtain the script outline output by the outline generation agent, wherein the second script generation configuration parameter information is the result of parameter completion of the first script generation configuration parameter information; and to input the script outline and the second script generation configuration parameter information into the script generation agent to obtain the script file output by the script generation agent.

[0153] Optionally, the determining module 11 is specifically used to: input the script generation requirement description information input by the user into the requirement analysis agent, so that the requirement analysis agent parses the script generation requirement description information according to the set parameter set to obtain the structured first script generation configuration parameter information, so as to determine the target sub-agents that need to be called sequentially in the workflow according to the completeness of the first script generation configuration parameters; and obtain the first script generation configuration parameter information and the target sub-agents that need to be called sequentially in the workflow fed back by the requirement analysis agent.

[0154] Optionally, the acquisition module 12 is specifically used for: determining the script type based on the second script generation configuration parameter information; inputting the script type and the second script generation configuration parameter information into the outline generation agent to obtain a script outline output by the outline generation agent corresponding to the script type. Furthermore, if the second script generation configuration parameter information indicates script file format reference acquisition information, then acquiring the script file format reference information based on the script file format reference acquisition information; inputting the script outline, the second script generation configuration parameter information, and the script file format reference information into the script generation agent to obtain a script file output by the script generation agent.

[0155] Optionally, the target sub-agent further includes a compliance detection agent; the acquisition module 12 is further configured to: perform multi-dimensional detection on the script file through the compliance detection agent to obtain multi-dimensional detection results, the multi-dimensional including: user-defined prohibited words, prohibited words defined by the live streaming platform, and preset general detection items; if the detection result indicates that the preset general detection item fails the detection, the script generation task is re-executed; if the detection result indicates that the script file includes the prohibited words, the prohibited words in the script file are semantically replaced.

[0156] Optionally, the target sub-agent further includes an optimization agent; the device further includes: an optimization module, configured to determine historical user interaction data of a historical script file matching the script generation task based on the configuration parameter information generated by the second script; input the historical user interaction data into the optimization agent, so that the optimization agent can estimate the interaction effect index and user question data corresponding to the script file, and optimize the script file based on the interaction effect index and the user question data.

[0157] Optionally, the device further includes: an adding module, used to acquire user interaction data during the execution of the script file; and to add the script file to the knowledge base when it is determined from the user interaction data that the script file meets preset excellent script screening conditions.

[0158] Optionally, the acquisition module 12 is further configured to: in response to a checkpoint operation set for at least some of the sub-intelligent agents, acquire the state information of the at least some sub-intelligent agents and store the state information in a target storage space; if it is determined based on the state information that there is an abnormal intelligent agent among the at least some sub-intelligent agents that is running abnormally, then re-trigger the operation of the abnormal intelligent agent.

[0159] Optionally, the parameter acquisition agent includes a data collection agent and a topic setting agent; the determining module 11 is further configured to: if the first script generation configuration parameter information lacks target product attribute information and / or live streamer introduction information, input the target product identifier and / or live streamer identifier corresponding to the script generation task into the data collection agent to obtain the target product attribute information and / or live streamer introduction information output by the data collection agent after retrieval in the knowledge base; if the first script generation configuration parameter information lacks a live stream topic, input the first script generation configuration parameter information into the topic setting agent to enable the topic setting agent to perform parameter analysis on the first script generation configuration parameter information and generate the live stream topic; input the second script generation configuration parameter information containing the target product attribute information and the live stream topic into the outline generation agent.

[0160] Optionally, the acquisition module 12 is further configured to: send the script outline to the user for review; receive the review result from the user; if the review result is unsuccessful, update the second script generation configuration parameter information based on the review result, and re-input the updated second script generation configuration parameter information into the outline generation agent to obtain a new script outline.

[0161] Figure 8The device shown can perform the steps in the live script generation method in the foregoing embodiments. For detailed execution process and technical effects, please refer to the description in the foregoing embodiments, which will not be repeated here.

[0162] Figure 9 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 9 As shown, in practice, this electronic device includes a memory 21 and a processor 22.

[0163] Memory 21 is used to store computer programs and can be configured to store various other data to support operation on the electronic device. Examples of this data include instructions for any application or method used to operate on the electronic device, data structures, contact data, phone book data, messages, pictures, videos, etc.

[0164] The processor 22, coupled to the memory 21, is used to execute the computer program in the memory 21 to implement the live script generation method provided in the foregoing embodiments.

[0165] Furthermore, such as Figure 9 As shown, the electronic device also includes other components such as a communication component 23, a display 24, a power supply component 25, and an audio component 26. Figure 9 The diagram only shows some components and does not mean that the electronic device includes only these components. Figure 9 The components shown are as follows. The electronic device in this embodiment can be a terminal device such as a desktop computer, laptop computer, smartphone, or IoT device, or a server device such as a conventional server, cloud server, or server array.

[0166] The aforementioned memory can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random-Access Memory (SRAM), Electrically Erasable Programmable Read Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0167] The aforementioned communication component is configured to facilitate wired or wireless communication between the device containing the communication component and other devices. The device containing the communication component can access wireless networks based on communication standards, such as 2G, 3G, 4G / LTE, 5G, or combinations thereof. In one exemplary embodiment, the communication component receives broadcast signals or broadcast-related information from an external broadcast management system via a broadcast channel.

[0168] The aforementioned display includes a screen, which may include a Liquid Crystal Display (LCD) and a Touch Panel (TP). If the screen includes a Touch Panel, the screen can be implemented as a touchscreen to receive input signals from the user. The Touch Panel includes one or more touch sensors to sense touches, swipes, and gestures on the Touch Panel. The touch sensors can sense not only the boundaries of touch or swipe actions but also the duration and pressure associated with the touch or swipe operation.

[0169] The aforementioned power supply components provide power to various components within the device in which they reside. These power supply components may include a power management system, one or more power sources, and other components associated with generating, managing, and distributing power to the device in which they reside.

[0170] The aforementioned audio component can be configured to output and / or input audio signals. For example, the audio component includes a microphone (MIC) configured to receive external audio signals when the device containing the audio component is in an operating mode, such as call mode, recording mode, or voice recognition mode. The received audio signals can be further stored in memory or transmitted via a communication component. In some embodiments, the audio component also includes a speaker for outputting audio signals.

[0171] Accordingly, embodiments of this application also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, enables the processor to implement the steps in the above-described method embodiments. The computer-readable storage medium includes volatile or non-volatile components, or a combination thereof, and can be removable or non-removable. Examples of computer-readable storage media include, but are not limited to, phase-change random access memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random-access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), flash memory or other memory technologies, CD-ROM, Digital Video Disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium. Accordingly, this application also provides a computer program product, which includes a computer program or instructions that, when executed by a processor, cause the processor to implement the steps in the above method embodiments. It should be understood that each step or combination of steps in the above method flow can be implemented by the computer program or instructions. Furthermore, these computer programs or instructions can be applied to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device, enabling the processor of the general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing device to function as an apparatus for implementing the corresponding functions in the above method embodiments.

[0172] Finally, it should be noted that the above 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.

Claims

1. A method for generating live streaming scripts, characterized in that, Applied to a main intelligent agent, which sets up a workflow consisting of multiple sub-intelligent agents, the method includes: In response to the triggering of the script generation task, the target sub-intelligent agents that need to be called sequentially in the workflow are determined based on the completeness of the configuration parameter information generated by the first script. If the completeness is insufficient, the target sub-intelligent agents include a parameter acquisition agent, an outline generation agent, and a script generation agent. The second script generation configuration parameter information determined by the intelligent agent through the parameters is input into the outline generation intelligent agent to obtain the script outline output by the outline generation intelligent agent. The second script generation configuration parameter information is the result of parameter completion of the first script generation configuration parameter information. Input the script outline and the second script generation configuration parameter information into the script generation agent to obtain the script file output by the script generation agent.

2. The method according to claim 1, characterized in that, The step of determining the target sub-agents that need to be called sequentially in the workflow based on the completeness of the configuration parameter information generated according to the first script includes: The user-input script generates requirement description information, which is then input into the requirement analysis agent. The requirement analysis agent parses the script generates requirement description information according to a set of parameters to obtain structured first script generation configuration parameter information. Based on the completeness of the first script generation configuration parameters, the target sub-agents that need to be called sequentially in the workflow are determined. Obtain the first script generation configuration parameter information fed back by the demand analysis agent, as well as the target sub-agents that need to be called sequentially in the workflow.

3. The method according to claim 1, characterized in that, The step of inputting the second script generation configuration parameter information determined by the intelligent agent through the parameters into the outline generation intelligent agent to obtain the script outline output by the outline generation intelligent agent includes: Based on the configuration parameter information generated by the second script, the script type is determined; The script type and the second script generation configuration parameter information are input into the outline generation agent to obtain the script outline corresponding to the script type output by the outline generation agent.

4. The method according to claim 1, characterized in that, The step of inputting the script outline and the second script generation configuration parameter information into the script generation agent to obtain the script file output by the script generation agent includes: If the second script generation configuration parameter information indicates script file format reference acquisition information, then obtain the script file format reference information according to the script file format reference acquisition information; The script outline, the second script generation configuration parameter information, and the script file format reference information are input into the script generation agent to obtain the script file output by the script generation agent.

5. The method according to any one of claims 1-4, characterized in that, The target sub-agent also includes a compliance detection agent; the method further includes: The compliance detection agent performs multi-dimensional detection on the script file to obtain multi-dimensional detection results. The multi-dimensional detection results include: user-defined prohibited words, prohibited words defined by the live streaming platform, and preset general detection items. If the detection result indicates that the preset general detection item has failed the detection, the script generation task will be re-executed. If the detection result indicates that the script file includes the banned word, then the banned word in the script file is semantically replaced.

6. The method according to any one of claims 1-4, characterized in that, The target sub-agent also includes an optimizing agent; the method further includes: Based on the configuration parameter information generated by the second script, determine the historical user interaction data of the historical script file that matches the script generation task; The historical user interaction data is input into the optimization agent so that the optimization agent can predict the interaction effect index and user question data corresponding to the script file, and optimize the script file based on the interaction effect index and the user question data.

7. The method according to any one of claims 1-4, characterized in that, The method further includes: During the execution of the script file, user interaction data is acquired; When the script file is determined to meet the preset excellent script selection criteria based on the user interaction data, the script file is added to the knowledge base.

8. The method according to any one of claims 1-4, characterized in that, The method further includes: In response to a checkpoint operation set for at least some of the multiple sub-agents, the state information of the at least some sub-agents is obtained and stored in a target storage space; If, based on the state information, it is determined that there is an abnormal intelligent agent among at least some of the sub-intelligent agents that is malfunctioning, then the operation of the abnormal intelligent agent is retried.

9. The method according to any one of claims 1-4, characterized in that, The parameter acquisition agent includes a data collection agent and a topic setting agent; the step of inputting the second script generation configuration parameter information determined by the parameter acquisition agent into the outline generation agent includes: If the target product attribute information and / or the live streamer introduction information are missing in the configuration parameter information generated by the first script, then the target product identifier and / or the live streamer identifier corresponding to the script generation task are input into the data collection agent to obtain the target product attribute information and / or the live streamer introduction information output by the data collection agent after searching in the knowledge base; If the configuration parameter information generated by the first script is missing a live broadcast topic, then the configuration parameter information generated by the first script is input into the topic setting agent, so that the topic setting agent can perform parameter analysis on the configuration parameter information generated by the first script and generate the live broadcast topic. The second script generation configuration parameters, which include the target product attribute information and the live stream theme, are input into the outline generation agent.

10. The method according to any one of claims 1-4, characterized in that, After inputting the second script generation configuration parameter information determined by the intelligent agent through the parameters into the outline generation intelligent agent to obtain the script outline output by the outline generation intelligent agent, the method further includes: Send the script outline to the user for review; The system receives the review result from the user feedback. If the review result is negative, it updates the second script generation configuration parameter information based on the review result and re-inputs the updated second script generation configuration parameter information into the outline generation agent to obtain a new script outline.

11. An electronic device, characterized in that, include: The system includes a memory, a processor, and a communication interface; wherein the memory stores executable code, and when the executable code is executed by the processor, the processor performs the live script generation method as described in any one of claims 1 to 10.

12. A non-transitory machine-readable storage medium, characterized in that, The non-transitory machine-readable storage medium stores executable code, which, when executed by a processor of an electronic device, causes the processor to perform the live script generation method as described in any one of claims 1 to 10.

13. A computer program product, characterized in that, include: A computer program, when executed by a processor of an electronic device, causes the processor to perform the live script generation method as described in any one of claims 1 to 10.

Images