Response text generation method and device, electronic equipment and storage medium

Abstract

The embodiment of the invention provides a reply text generation method and device, electronic equipment and a storage medium, and belongs to the technical field of artificial intelligence. The method comprises the following steps: determining a corresponding target domain agent based on an identified target technical field through a domain scheduling agent, and obtaining the target domain agent through cooperative work of at least two cascaded first draft agents and quality inspection agents in the target domain agent, optimizing the generated reply content in combination with a plurality of preset reply book scoring standards and quality up-to-standard conditions so as to regenerate the reply content; and when each of the quality scores of the plurality of reply book scoring standards meets a predetermined quality standard condition, performing reply text generation on the reply content to obtain a target reply text. According to the embodiment of the invention, the reply text is automatically generated through cooperative work of a plurality of agents, the reply text does not need to be participated in the reply generation process by continuously inputting prompt words manually, and the reply text generation efficiency is remarkably improved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and particularly to a method and device for generating response texts, an electronic device, and a storage medium. Background Art

[0002] Currently, traditional methods for generating response texts usually generate corresponding response texts for examination opinions through artificial intelligence and natural language generation (NLG) technology. For example, in the process of patent response, the main examination content of the notice of examination opinion and the cited comparative documents can be analyzed through an artificial intelligence model. However, continuous input of prompt words is required to schedule the artificial intelligence model to implement the analysis of the notice of examination opinion and the cited comparative documents. That is to say, related technologies need to continuously input prompt words manually to instruct the artificial intelligence model to generate examination content, and further sort out the examination content to generate response texts through NLG technology and predefined rules and templates. However, this method can only focus on single links such as patent analysis, examination opinion analysis, or document generation, and only realizes automation in a single link. However, the process of generating response texts involves multiple links, and the above method still requires manual participation in the process of generating response texts, resulting in low efficiency of generating response texts. Therefore, how to improve the efficiency of generating response texts has become an urgent problem to be solved. Summary of the Invention

[0003] The main objective of the embodiments of this application is to propose a method and device for generating response texts, an electronic device, and a storage medium, aiming to improve the efficiency of generating response texts.

[0004] To achieve the above objective, in the first aspect of the embodiments of this application, a method for generating response texts is proposed. The method includes:

[0005] Obtain a target examination opinion text; wherein, the target examination opinion text includes a notice of examination opinion text, a comparative text, and an application text;

[0006] Perform domain recognition on the target examination opinion text through a pre-trained domain scheduling agent to obtain a target technical domain, and determine a target domain agent from pre-trained candidate domain agents based on the target technical domain; wherein, the target domain agent includes at least two cascaded draft agents, a target quality review agent, and a target response text agent;

[0007] Generate a response content by using the at least two cascaded draft agents for the target technical domain, the notice of examination opinion text, the comparative text, and the application text;

[0008] The target quality review agent scores the content quality of the response content to obtain quality scores for multiple response letter scoring criteria. If at least one of the quality scores for the multiple response letter scoring criteria does not meet the predetermined quality compliance condition, at least one of the at least two cascaded draft agents is optimized based on the response letter scoring criteria that do not meet the predetermined quality compliance condition to regenerate the response content;

[0009] If each of the quality scores for the multiple response letter scoring criteria meets the predetermined quality compliance condition, the target response text agent generates a target response text based on the preset response text format and the response content.

[0010] In some embodiments, the at least two cascaded draft agents include a target field draft generation agent and a target draft polishing agent;

[0011] The generation of the response content by the at least two cascaded draft agents for the target technical field, the examination opinion notice text, the comparison text, and the application text includes:

[0012] The target field draft generation agent generates a response opinion for the target technical field, the examination opinion notice text, the comparison text, and the application text to obtain an initial response content; wherein, the target field draft generation agent is used to respond to the viewpoints in the examination opinion based on the comparison text and the application text;

[0013] The target draft polishing agent optimizes the response opinion of the initial response content to obtain an optimized response content.

[0014] In some embodiments, the at least two cascaded draft agents further include a target draft reflection agent between the target field draft generation agent and the target draft polishing agent;

[0015] The optimization of the response opinion of the initial response content by the target draft polishing agent to obtain an optimized response content includes:

[0016] The target draft reflection agent performs content logic analysis on the initial response content to obtain content optimization ideas;

[0017] The target draft polishing agent optimizes the response opinion of the initial response content based on the content optimization ideas to obtain the optimized response content.

[0018] In some embodiments, optimizing at least one of the at least two cascaded draft agents based on the reply book scoring criteria that do not meet the predetermined quality compliance conditions includes:

[0019] If the reply book scoring criteria that do not meet the predetermined quality compliance conditions is the accurate quality score of the reply content, then optimize the target domain draft generation agent and the target draft reflection agent among the at least two cascaded draft agents;

[0020] If the reply book scoring criteria that do not meet the predetermined quality compliance conditions is the compliance quality score of the reply content, then optimize the target domain draft generation agent and the target draft reflection agent among the at least two cascaded draft agents;

[0021] If the reply book scoring criteria that do not meet the predetermined quality compliance conditions is the logical quality score of the reply content, then optimize the target domain draft generation agent and the target draft reflection agent among the at least two cascaded draft agents.

[0022] In some embodiments, optimizing at least one of the at least two cascaded draft agents based on the reply book scoring criteria that do not meet the predetermined quality compliance conditions further includes:

[0023] If the reply book scoring criteria that do not meet the predetermined quality compliance conditions is the language quality score of the reply content, then optimize the target domain draft generation agent and the target draft polishing agent among the at least two cascaded draft agents.

[0024] In some embodiments, the domain recognition of the target examination opinion text by the pre-trained domain scheduling agent to obtain the target technical domain includes:

[0025] Extract the text domain features of the target examination opinion text through the domain scheduling agent to obtain the text domain features;

[0026] Perform domain analysis on the text domain features to obtain the target technical domain.

[0027] In some embodiments, the generation of the target reply text by the target reply text agent for the preset reply text format and the reply content includes:

[0028] Extract the format features of the reply text format through the target reply text agent to obtain the text format features, and extract the content features of the reply content through the target reply text agent to obtain the reply content features;

[0029] Insert the response content into the response text format in sequence according to the text format feature and the response content feature to obtain the target response text.

[0030] To achieve the above object, a second aspect of the embodiments of the present application provides a response text generation device, the device includes:

[0031] An examination opinion text acquisition module, configured to acquire a target examination opinion text; wherein, the target examination opinion text includes an examination opinion notice text, a comparison text, and an application text;

[0032] A field scheduling agent module, configured to perform field recognition on the target examination opinion text through a pre-trained field scheduling agent to obtain a target technical field, and determine a target field agent from pre-trained candidate field agents based on the target technical field; wherein, the target field agent includes at least two cascaded draft agents, a target quality review agent, and a target response text agent;

[0033] A response opinion generation module, configured to generate response opinions on the target technical field, the examination opinion notice text, the comparison text, and the application text through the at least two cascaded draft agents to obtain a response content;

[0034] A response opinion optimization module, configured to perform content quality scoring on the response content through the target quality review agent to obtain quality scores of multiple response book scoring criteria, and if at least one of the quality scores of the multiple response book scoring criteria does not meet a predetermined quality compliance condition, optimize at least one of the at least two cascaded draft agents based on the response book scoring criteria that do not meet the predetermined quality compliance condition to regenerate the response content;

[0035] A response text generation module, configured to, if each of the quality scores of the multiple response book scoring criteria meets the predetermined quality compliance condition, generate a response text on a preset response text format and the response content through the target response text agent to obtain a target response text.

[0036] To achieve the above object, a third aspect of the embodiments of the present application provides an electronic device, the electronic device includes a memory and a processor, the memory stores a computer program, and when the processor executes the computer program, the method described in the first aspect above is implemented.

[0037] To achieve the above object, a fourth aspect of the embodiments of the present application provides a computer-readable storage medium, the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method described in the first aspect above is implemented.

[0038] The reply text generation method and device, electronic device, and storage medium proposed in this application first use a domain scheduling agent to identify the domain of the target examination opinion text (specifically including the notice of examination opinion text, comparison text, and application text), which can identify the target technical domain, and determine the corresponding target domain agent based on the target technical domain, realizing the dynamic role allocation of agents for different technical domains; secondly, through the collaborative work of at least two cascaded draft agents and quality review agents in the target domain agent, patent analysis, examination opinion analysis, and reply text generation can be achieved in one step, and the quality of the reply content is gradually improved by combining multiple preset reply book scoring criteria and quality compliance conditions, realizing the continuous optimization of the reply opinion, which helps to improve the accuracy of the reply content generation; finally, the target reply text is generated through the reply text agent and the preset reply text format. It can be seen that this application can automate all links of patent analysis, examination opinion analysis, and reply opinion generation in the reply text generation process, without relying on manual continuous input of prompt words to participate in the reply generation process, significantly improving the efficiency of reply text generation. Description of the Drawings

[0039] Figure 1 is a flowchart of the reply text generation method provided by an embodiment of this application;

[0040] Figure 2 is Figure 1 a flowchart of step S102 in

[0041] Figure 3 is Figure 1 a flowchart of step S103 in

[0042] Figure 4 is Figure 3 a flowchart of step S302 in

[0043] Figure 5 is Figure 1 a flowchart of step S104 in

[0044] Figure 6 is Figure 1 a flowchart of step S105 in

[0045] Figure 7 is a schematic structural diagram of the reply text generation device provided by an embodiment of this application;

[0046] Figure 8 is a schematic hardware structure diagram of the electronic device provided by an embodiment of this application. Detailed Embodiments

[0047] In order to make the objectives, technical solutions and advantages of the present application clearer and more understandable, the present application will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0048] It should be noted that although functional modules are divided in the device schematic diagram and the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different module division in the device or a different order in the flowchart. Terms such as "first" and "second" in the description, claims and the above-mentioned drawings are used to distinguish similar objects and do not necessarily need to describe a specific order or sequence.

[0049] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0050] First, several nouns involved in the present application are analyzed:

[0051] Artificial intelligence (AI): It is a new technical science that studies, develops theories, methods, technologies and application systems for simulating, extending and expanding human intelligence; artificial intelligence is a branch of computer science. Artificial intelligence attempts to understand the essence of intelligence and produce a new intelligent machine that can react in a way similar to human intelligence. The research in this field includes robots, speech recognition, image recognition, natural language processing and expert systems, etc. Artificial intelligence can simulate the information process of human consciousness and thinking. Artificial intelligence also uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results of theories, methods, technologies and application systems.

[0052] Agent: It refers to a computer program or system that can simulate human intelligent behavior, perceive the environment, make autonomous decisions and execute tasks. The agent perceives environmental information through sensors or data input, conducts reasoning and decision-making based on predefined rules or learning algorithms, and can complete tasks in an autonomous manner. The goal of the agent is to simulate, extend and expand human intelligent behavior and provide efficient and intelligent solutions for the automation of complex tasks.

[0053] The embodiments of the present application provide a method and device for generating reply texts, an electronic device and a storage medium, aiming to improve the efficiency of generating reply texts.

[0054] The response text generation method, device, electronic device, and storage medium provided by the embodiments of the present application will be specifically described through the following embodiments. First, the response text generation method in the embodiments of the present application will be described.

[0055] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence (AI) is a theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use knowledge to obtain the best results.

[0056] Artificial intelligence basic technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, and mechatronics. Artificial intelligence software technologies mainly include several major directions such as computer vision technology, robotics, biometric technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0057] The response text generation method provided by the embodiments of the present application relates to the field of artificial intelligence technology. The response text generation method provided by the embodiments of the present application can be applied to a terminal, can also be applied to a server side, or can also be software running on a terminal or a server side. In some embodiments, the terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, etc.; the server side can be configured as an independent physical server, can also be configured as a server cluster or a distributed system composed of multiple physical servers, or can also be configured as a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application that implements the response text generation method, etc., but is not limited to the above forms.

[0058] This application can be used in numerous general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multi-processor systems, microprocessor-based systems, set-top boxes, programmable consumer electronic devices, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, and so on. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. This application can also be practiced in a distributed computing environment where tasks are executed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0059] Figure 1 is an optional flowchart of the response text generation method provided by an embodiment of this application. Figure 1 The method in may include but is not limited to steps S101 to S105.

[0060] Step S101, obtain the target examination opinion text; wherein, the target examination opinion text includes the examination opinion notice text, the comparison text, and the application text.

[0061] Step S102, perform domain recognition on the target examination opinion text through a pre-trained domain scheduling agent to obtain the target technical domain, and determine the target domain agent from the pre-trained candidate domain agents based on the target technical domain; wherein, the target domain agent includes at least two cascaded draft agents, a target quality review agent, and a target response text agent.

[0062] Step S103, generate response opinions for the target technical domain, the examination opinion notice text, the comparison text, and the application text through at least two cascaded draft agents to obtain the response content.

[0063] Step S104, perform content quality scoring on the response content through the target quality review agent to obtain quality scores for multiple response book scoring criteria, and if at least one of the quality scores for multiple response book scoring criteria does not meet the predetermined quality compliance condition, then optimize at least one of the at least two cascaded draft agents based on the response book scoring criteria that do not meet the predetermined quality compliance condition to regenerate the response content.

[0064] In step S105, if each of the quality scores of multiple response letter scoring criteria meets the predetermined quality compliance condition, the target response text agent generates a response text based on the preset response text format and response content to obtain the target response text.

[0065] Steps S101 to S105 illustrated in the embodiments of the present application first use the domain scheduling agent to perform domain identification on the target examination opinion text (specifically including the examination opinion notice text, the comparison text, and the application text), can identify the target technical field, and determine the corresponding target domain agent based on the target technical field, realizing the dynamic role assignment of agents for different technical fields. Secondly, through the collaborative work of at least two cascaded draft agents and quality review agents in the target domain agent, patent analysis, examination opinion analysis, and response text generation can be achieved in one step. Moreover, by combining multiple preset response letter scoring criteria and quality compliance conditions, the quality of the response content is gradually improved, realizing the continuous optimization of the response opinion, which helps to improve the accuracy of the response content generation. Finally, the target response text is generated through the response text agent and the preset response text format. It can be seen that the present application can automate all links of patent analysis, examination opinion analysis, and response opinion generation in the process of response text generation, without relying on manual continuous input of prompt words to participate in the response generation process, significantly improving the efficiency of response text generation.

[0066] In step S101 of some embodiments, specifically, the target examination opinion text refers to the set of official examination opinion notices and related documents issued by the examiner for the application text during the patent examination process, mainly including the examination opinion notice text, the comparison text, and the application text.

[0067] Specifically, the examination opinion notice text refers to the official examination opinion put forward by the examiner for the patent application, including but not limited to the examination conclusion, the basis for citing the comparison document, and the examination opinion on the defects of the application text, etc. The comparison text is the prior art document cited by the examiner and used as the basis for the examination opinion. The application text is the patent document submitted by the applicant, including the abstract of the specification, the claims, the specification, and the drawings of the specification, etc.

[0068] Before step S102 in some embodiments, in an optional embodiment of the present application, multiple agents need to be pre-trained. Specifically, the functions and behaviors of each agent are defined by the prompt design of different agents. In the specific implementation process, the function of each agent is defined by prompt strategies such as preset task descriptions, input-output formats, and context constraints, so that the agent can show targeted capabilities in different task scenarios.

[0069] In one example, the initial domain scheduling agent (also known as the Bole Agent) is guided by a preset domain Prompt to identify the patent domain, and in combination with the input training examination opinion text, dynamically defines the roles of subsequent agents (such as at least two cascaded draft agents, quality review agents, and response text agents) to ensure that the response tasks are executed by suitable role agents; further, the draft agent relies on the draft Prompt to set the technical discussion structure to ensure that the generated response content conforms to the logic of patent documents; the reflection agent is guided by the reflection Prompt to check the logical integrity and technical accuracy of the response content and put forward optimization suggestions; the polishing agent is guided by the polishing Prompt to perform grammar adjustment and expression optimization based on the optimization suggestions; the quality review agent combines the idea of reinforcement learning and uses the quality review Prompt to set the scoring criteria to guide the large model to automatically evaluate the response content based on preset rules such as compliance, logic, accuracy, and language quality, and can train different agents by matching different Prompts for different agents, so as to obtain each pre-trained agent.

[0070] For example, the Prompt design of the Bole Agent can include domain task descriptions, input formats, output formats, and context constraints. Among them, the task description can be: You are an expert in patent domain identification. Your task is to identify the technical domain to which the input patent document belongs and select the most suitable agent from the candidate agent library to process the patent; the input format can be {Notice of Examination Opinion Attachment}, {Application Text Attachment}, and {Comparative Document Attachment}; the output format can be Technical Domain: {Identified Technical Domain} and Adapted Domain Agent: {List of Selected Domain Agents}.

[0071] For example, the prompt design for the draft agent can include a domain task description, input format, output format, and technical argument structure constraints. The task description might be: "You are a patent response expert. Your task is to generate a response that conforms to the patent response logic based on the examination conclusions, comparative documents, and application text in the office action notice." The input format might be {Office Action Notice Attachment}, {Application Text Attachment}, and {Comparative Document Attachment}. The technical argument structure constraints might be: 1. Analyze the examiner's examination conclusions and objections to extract the examination opinions. 2. Generate technical argument rebuttals for each examination opinion based on the application text. 3. Ensure that the structure of the reply conforms to the logic of the opinion statement, including the following parts: examination opinion, modification explanation (the specific modification content of the claims and the cited application text paragraphs need to be explained), point-by-point rebuttal to the examination opinion (technical discussion of each examiner's point, which needs to be analyzed in combination with the specific technical details in the application text and comparative documents) and conclusion (summarize the reply content and explicitly request the examiner to reconsider the patent application); the output format can be examination opinion: {examination conclusion}, modification explanation: {the content of the modified claims and the cited application text paragraphs}, point-by-point rebuttal to the examination opinion: {description of the examination opinion, the rebuttal content to the examination opinion}, conclusion: {summary of the reply content}.

[0072] For example, the prompt design for the reflective agent can include a task description, input format, output format, and check rule constraints. The check rule constraints can include: 1. Check the logical integrity of the response, ensuring the sufficiency of the rebuttal reasons for each examination conclusion and that the rebuttal reasons are logically clear and complete. 2. Check the technical accuracy of the response, ensuring that the technical discussion of the rebuttal reasons is consistent with the technical details in the patent text and that the cited legal provisions are accurate. 3. Identify logical loopholes or technical inaccuracies in the response and provide specific optimization suggestions.

[0073] In one embodiment of the present application, the prompt strategies of the polishing agent and the quality review agent are similar to those of the above-mentioned Bole Agent and the first draft agent, and are not described in detail here.

[0074] This application adopts different Prompt strategies for different intelligent agents, can define the functions of each intelligent agent, and obtain pre-trained intelligent agents, so that users can schedule intelligent agents only by indicating the field, and can realize the automatic generation of reply content, reply content optimization and reply text through the joint operation of various intelligent agents. There is no need to rely on manual input of prompt words to participate in the reply generation process, which significantly improves the efficiency of reply text generation.

[0075] See also Figure 2, in some embodiments, step S102 includes but is not limited to steps S201 to S202:

[0076] Step S201, the domain scheduling agent extracts the text domain features from the target examination opinion text to obtain the text domain features.

[0077] Step S202, perform domain analysis on the text domain features to obtain the target technical domain.

[0078] In step S201 of some embodiments, specifically, the domain scheduling agent is a trained Bole Agent, which is used to automatically identify the technical domain to which the text belongs according to the text content, and determine the target domain agent corresponding to the technical domain from multiple candidate domain agents based on this technical domain.

[0079] Specifically, the text domain features refer to the semantic features reflecting the technical domain of the target examination opinion text.

[0080] Specifically, the domain scheduling agent analyzes keywords, technical terms, domain information, patent classification numbers, etc. in the target examination opinion text, and extracts information that can reflect the technical domain features.

[0081] For example, in patents in the field of artificial intelligence, the domain scheduling agent will extract technical terms such as deep learning, convolutional neural network, image recognition, technical domain information of image processing, and classification numbers such as G06T.

[0082] In step S202 of some embodiments, specifically, the target technical domain refers to the technical domain to which the target examination opinion text belongs. This technical domain includes but is not limited to: artificial intelligence, biomedicine, mechanical engineering, electronic communication, chemical materials, etc.

[0083] Specifically, the domain scheduling agent will calculate the matching degree between the target examination opinion text and each technical domain according to the technical term weights and semantic association degrees in the text domain features, and select the domain with the highest matching degree as the target technical domain.

[0084] For example, if the weights of technical terms such as "convolutional neural network" and "image recognition" in the text domain features are relatively high, the domain scheduling agent will analyze the target technical domain as the artificial intelligence technical domain; if the weights of technical terms such as "gene sequence" and "drug compound" in the text domain features are relatively high, it will be analyzed as the biomedicine technical domain.

[0085] In the embodiments of the present application, through domain analysis of the text domain features, the domain scheduling agent can accurately identify the technical domain to which the target examination opinion text belongs, providing a domain information basis for the subsequent dynamic allocation of agents.

[0086] In step S102 of some embodiments, specifically, if the target technical field is identified as the artificial intelligence technical field, an agent in the artificial intelligence field is determined from the pre-trained candidate field agents as the target field agent.

[0087] Specifically, the target field agent includes at least two cascaded draft agents, a target quality review agent, and a target reply text agent. Among them, the draft agent includes a draft Agent, a reflection Agent, and a polishing Agent, etc. The draft Agent is used to reply to the viewpoints in the examination opinion based on the comparison text and the application text. The reflection Agent is used to preliminarily optimize the preliminary reply content to ensure the integrity, logical rationality, and technical accuracy of the reply content. The polishing Agent is used to optimize the preliminary reply content based on the improvement ideas of the reflection Agent. The target quality review agent is used to evaluate the quality of the optimized reply content. The target reply text agent (such as a document generation Agent) is used to generate the final reply text.

[0088] In this embodiment, determining the target field agent from the pre-trained candidate field agents based on the target technical field can intelligently adjust the roles and tasks of the subsequent agents according to the technical field of the patent, can flexibly allocate the tasks of different role agents according to the characteristics of the patent field, and helps to improve the efficiency of reply text generation.

[0089] Please refer to Figure 3 , in some embodiments, step S103 includes but is not limited to steps S301 to S302:

[0090] Step S301, a target field draft generation agent generates a reply opinion on the target technical field, the examination opinion notice text, the comparison text, and the application text to obtain an initial reply content; among them, the target field draft generation agent is used to reply to the viewpoints in the examination opinion based on the comparison text and the application text.

[0091] Step S302, a target draft polishing agent optimizes the reply opinion of the initial reply content to obtain an optimized reply content.

[0092] In step S301 of some embodiments, specifically, at least two cascaded draft agents include a target field draft generation agent and a target draft polishing agent. Among them, the target field draft generation agent is used to reply to the viewpoints in the examination opinion based on the comparison text and the application text; the target draft polishing agent is used to optimize the initial reply content based on the improvement ideas of the reply content.

[0093] Specifically, the target domain initial draft generation agent does not simply fill in the text. Instead, based on the specific requirements of the examination opinions and in combination with the patent application content, it ensures the logical consistency and clarity of the response through the method of step-by-step reasoning (Chain-of-Thought, CoT).

[0094] For example, when the examination opinion points out that the application text lacks creativity based on the prior art, the target domain initial draft generation agent analyzes the reasons for rejection and provides targeted refutation logic, such as identifying the distinguishing technical features between the application text and the comparative document, and discussing the creativity of the distinguishing technical features based on functional differences, structural differences, or experimental data, etc.

[0095] Furthermore, the target domain initial draft generation agent also generates the response content structurally according to a preset format (such as the text content format of the statement of opinion), ensuring that it is well-structured and meets the specification requirements of the patent examination process.

[0096] In this embodiment, by generating response opinions for the target technical field, examination opinion notice text, comparative text, and application text through the target domain initial draft generation agent, not only can the target domain initial draft generation agent generate the initial draft content of the response book that conforms to the characteristics of the target technical field, but also ensure the logic and accuracy of the initial draft content.

[0097] Refer to Figure 4 , in some embodiments, step S302 includes but is not limited to steps S401 to S402:

[0098] Step S401, the target initial draft reflection agent performs content logic analysis on the initial response content to obtain content optimization ideas.

[0099] Step S402, the target initial draft polishing agent optimizes the response opinions of the initial response content based on the content optimization ideas to obtain optimized response content.

[0100] In step S401 of some embodiments, specifically, at least two cascaded initial draft agents further include a target initial draft reflection agent between the target domain initial draft generation agent and the target initial draft polishing agent; wherein, the target initial draft reflection agent is used to perform preliminary optimization on the initial response content to obtain content optimization ideas for the initial response content.

[0101] Specifically, the target initial draft reflection agent analyzes the logical structure of the initial response content, the accuracy of technical discussions, the accuracy of legal clause citations, and the content integrity to identify possible logical loopholes or imperfections in the initial response content, and generates corresponding content optimization ideas for the logically flawed or imperfect content.

[0102] For example, by analyzing whether the refuting content in the initial reply content exactly corresponds to the viewpoints in the examination opinion, whether there are omissions or unclear expressions when refuting the viewpoints in the examination opinion, and checking whether the technical terms in the refuting content meet the patent requirement standards. If the refutation of Comparative Document 1 lacks specific technical details and the description of the refuting content does not cite specific paragraphs in the specification, then an optimization idea for supplementing the specific technical feature details of the refuting Comparative Document 1 by citing relevant paragraphs in the specification is generated.

[0103] In this embodiment, through the target initial draft reflection agent's content logic analysis of the initial reply content, an content optimization idea is obtained, which can identify possible logical loopholes, unclear arguments or insufficient evidence in the initial reply content, and accordingly form an content optimization idea to provide clear guidance for subsequent content optimization, helping to improve the accuracy of the generated reply content.

[0104] In step S402 of some embodiments, specifically, the target initial draft polishing agent can optimize the language expression, logical structure and technical rigor of the initial reply content based on the content optimization idea.

[0105] Furthermore, the target initial draft polishing agent will adjust the sentence structure of the initial reply content according to the content optimization idea to make it more in line with the specification requirements of the patent reply document, further improve the logical rigor of the technical discussion, ensure that the refuting opinions correspond one by one to the viewpoints in the examination opinion, and verify the cited legal provisions and technical terms to ensure the accuracy and standardization of the technical terms.

[0106] For example, if the content optimization idea is to supplement the specific technical feature details of the refuting comparative document by citing relevant paragraphs in the specification, then the polishing agent optimizes the initial reply content by supplementing the specific technical feature details of the refuting comparative document by citing relevant paragraphs in the specification, and for the reply content involving numerical ranges, experimental data or formula derivations in the application text, further optimizes the expression method to ensure that the quantitative data is reproducible, thereby generating a reply content with fluent language, clear logic, complete content and compliance with regulations.

[0107] In this embodiment, through the target initial draft polishing agent's optimization of the reply opinions based on the content optimization idea, the generation quality and readability of the optimized reply content are significantly improved, and the accuracy of the generated reply content is further improved.

[0108] The embodiments of the present application shown in steps S301 to S302 can utilize the collaborative work of the target field draft generation agent, the target draft reflection agent, and the target draft polishing agent to achieve the full - process automation from the initial response content to the optimized response content, solve the problem that artificial intelligence technology can only focus on single links such as patent retrieval, examination opinion analysis, or document generation, and without relying on manual participation in the generation of response texts, significantly improve the efficiency and quality of response text generation.

[0109] In step S104 of some embodiments, specifically, through the target quality review agent, the response content can be scored for accurate quality of the response content, compliance quality of the response content, logical quality of the response content, and language quality of the response content, etc., to obtain the content accurate quality score, content compliance quality score, content logical quality score, and content language quality score corresponding to the response book scoring criteria.

[0110] Furthermore, the content accurate quality score is used to determine whether the technical description of the response content is correct, unambiguous, and whether there is content inconsistent with the examination opinion; the content compliance quality score is used to ensure that the response content complies with the laws, regulations, and format requirements of patent examination, avoiding the risk of rejection due to format errors or violations of laws and regulations; the content logical quality score is used to evaluate the rigor of the argumentation in the response content, whether it can effectively support the innovative discussion of the application text, and whether it clearly responds to the key rejection points of the examination opinion, etc.; the content language quality score is used to check whether the language expression of the response content conforms to the specifications of patent response documents, avoiding long - winded, repetitive, or ambiguous expressions.

[0111] Furthermore, the predetermined quality compliance condition is determined based on the response book scoring criteria of the total score of the response content and the content accurate quality score, content compliance quality score, content logical quality score, and content language quality score.

[0112] For example, the predetermined quality compliance condition can be that the total score of the response content is greater than 85 points, and any one of the content accurate quality score, content compliance quality score, content logical quality score, and content language quality score cannot be lower than 20 points.

[0113] Furthermore, if at least one of the content accurate quality score, content compliance quality score, content logical quality score, and content language quality score does not meet 20 points, or the total score of the response content is equal to or less than 85 points, it indicates that the response content is unqualified.

[0114] In this embodiment, the target quality review agent performs content quality scoring on the response content to obtain quality scores for multiple response letter scoring criteria, enabling the quantification of the quality assessment criteria for the response content. Based on the quality scores and multiple response letter scoring criteria, problems such as possible logical loopholes, inaccurate technical features, inaccurate language expressions, or incorrect legal citations in the response content are effectively solved, significantly improving the accuracy of the response content.

[0115] Please refer to Figure 5 , in some embodiments, step S104 includes but is not limited to steps S501 to S503:

[0116] Step S501, if the response letter scoring criterion for not meeting the predetermined quality compliance condition is the accuracy quality score of the response content, optimize the target domain draft generation agent and the target draft reflection agent in at least two cascaded draft agents.

[0117] Step S502, if the response letter scoring criterion for not meeting the predetermined quality compliance condition is the compliance quality score of the response content, optimize the target domain draft generation agent and the target draft reflection agent in at least two cascaded draft agents.

[0118] Step S503, if the response letter scoring criterion for not meeting the predetermined quality compliance condition is the logic quality score of the response content, optimize the target domain draft generation agent and the target draft reflection agent in at least two cascaded draft agents.

[0119] In step S501 of some embodiments, specifically, if the response letter scoring criterion for not meeting the predetermined quality compliance condition is the accuracy quality score of the response letter, since the accuracy quality score of the response letter mainly evaluates the accuracy of the technical discussion in the response content, it is necessary to optimize the target domain draft generation agent and the target draft reflection agent.

[0120] For example, if the technical features cited in the response content do not match the comparative document, the generation strategy of the target domain draft generation agent will be optimized to make it pay more attention to the association between the technical features and the comparative document, and the target draft reflection agent will further strengthen the logical analysis of the technical discussion to ensure that it fully corresponds to the content refuting the views in the examination opinion.

[0121] In this embodiment, if the response letter scoring criterion for not meeting the predetermined quality compliance condition is the compliance quality score of the response content, the target domain draft generation agent and the target draft polishing agent in at least two cascaded draft agents are optimized, significantly improving the accuracy of the generated response content.

[0122] In step S502 of some embodiments, specifically, if the scoring criterion for the response letter that does not meet the predetermined quality compliance condition is the compliance quality score of the response content, since the compliance quality score of the response content mainly evaluates whether the response content complies with the legal norms and process requirements of patent examination, the intelligent agent for generating the first draft in the target field and the intelligent agent for reflecting on the target first draft are optimized.

[0123] For example, if the relevant legal provisions are not correctly cited in the response content, or the response is not written in accordance with the specification format of patent examination, the generation strategy of the intelligent agent for generating the first draft in the target field is optimized to make it pay more attention to the citation of legal provisions and the standardization of the format. And the intelligent agent for reflecting on the target first draft further strengthens the inspection and correction of compliance issues to ensure that the response content fully complies with the legal requirements and format standards of patent examination.

[0124] In this embodiment, if the scoring criterion for the response letter that does not meet the predetermined quality compliance condition is the compliance quality score of the response content, optimizing the intelligent agent for generating the first draft in the target field and the intelligent agent for reflecting on the target first draft among at least two cascaded first draft intelligent agents can significantly improve the compliance and standardization of the generated response content, making it more in line with the legal regulations requirements of patent examination, and further improving the accuracy of the generated response text.

[0125] In step S503 of some embodiments, specifically, if the scoring criterion for the response letter that does not meet the predetermined quality compliance condition is the logical quality score of the response content, since the logical quality score of the response letter mainly evaluates the logical rigor and structural clarity of the arguments in the response content, the intelligent agent for generating the first draft in the target field and the intelligent agent for reflecting on the target first draft are optimized.

[0126] For example, if the logical chain of technical refutation in the response content is incomplete, or the argument structure for the examination opinion is chaotic, the generation strategy of the intelligent agent for generating the first draft in the target field is optimized to make it pay more attention to the integrity and clarity of the logical structure. And the intelligent agent for reflecting on the target first draft further strengthens the analysis of the logical chain to ensure that each part of the argument closely corresponds to the view of the examination opinion and the overall structure is well - structured.

[0127] In this embodiment, if the scoring criterion for the response letter that does not meet the predetermined quality compliance condition is the logical quality score of the response content, optimizing the intelligent agent for generating the first draft in the target field and the intelligent agent for reflecting on the target first draft among at least two cascaded first draft intelligent agents can significantly improve the logical rigor and structural clarity of the generated response content, making it more in line with the specification requirements of patent response documents, and further improving the accuracy rate of response content generation.

[0128] In step S104 of some embodiments, optimizing at least one of at least two cascaded draft agents based on the reply book scoring criteria that do not meet the predetermined quality compliance conditions further includes: If the reply book scoring criteria that do not meet the predetermined quality compliance conditions is the language quality score of the reply content, then optimizing the target domain draft generation agent and the target draft polishing agent among the at least two cascaded draft agents.

[0129] Specifically, if the reply book scoring criteria that do not meet the predetermined quality compliance conditions is the language quality score of the reply content, since the language quality score of the reply content mainly evaluates whether the language expression of the reply content is clear and fluent, then optimize the target domain draft generation agent and the target draft polishing agent.

[0130] For example, if there are problems such as vague language expression, chaotic sentence structure, or inaccurate use of terms in the reply content, then optimize the generation strategy of the target domain draft generation agent to make it pay more attention to the clarity and professionalism of language expression, and the target draft polishing agent further strengthens the optimization of language expression (such as adjusting the sentence structure, correcting the use of terms, and improving the fluency of the text).

[0131] In this embodiment, if the reply book scoring criteria that do not meet the predetermined quality compliance conditions is the language quality score of the reply content, then optimizing the target domain draft generation agent and the target draft polishing agent among the at least two cascaded draft agents can significantly improve the language quality and readability of the generated reply content, make it easier to understand, and further improve the accuracy of the reply content generation.

[0132] In step S105 of some embodiments, specifically, if each of the quality scores of multiple reply book scoring criteria meets the predetermined quality compliance conditions, indicating that the generated reply content is qualified, then generate a reply text through the target reply text agent for the preset reply text format and the reply content to obtain the target reply text.

[0133] Please refer to Figure 6 , in some embodiments, step S105 includes but is not limited to steps S601 to S602:

[0134] Step S601, extracting format features of the reply text format through the target reply text agent to obtain text format features, and extracting content features of the reply content through the target reply text agent to obtain reply content features.

[0135] Step S602, inserting the reply content into the reply text format in sequence according to the text format features and the reply content features to obtain the target reply text.

[0136] In step S601 of some embodiments, specifically, the reply text format is a predefined reply standard template (such as the text format of an opinion statement), including but not limited to content such as a title, paragraph structure, citation format, and discussion on examination viewpoints, which is used to ensure that the generated reply text meets the specification requirements of patent examination.

[0137] Specifically, the target reply text agent analyzes each part in the reply text format (such as the title, paragraph structure, citation format, and discussion on examination viewpoints), extracts format structure features and typesetting rules, and forms text format features.

[0138] For example, the target reply text agent will extract the font, font size, and position information of the title, the indentation and line spacing rules of paragraphs, and the standard format of legal clause citations, etc.

[0139] Specifically, the reply content features include title content, modification description content for examination opinions, discussion content for examination opinions, content of cited laws and regulations clauses, etc.

[0140] Furthermore, the target reply text agent analyzes the title content, modification description content for examination opinions, discussion content for examination opinions, and content of cited laws and regulations clauses in the reply content, extracts the semantic features and language styles of each part of the content, and forms content features.

[0141] In this embodiment, by extracting text format features through the target reply text agent, the framework structure of the reply text can be clarified, providing an accurate positioning and layout basis for the subsequent insertion of the reply content. By extracting reply content features through the target reply text agent, the semantics and logic of the reply content can be better understood, ensuring that in the subsequent reply text generation process, the reply content can be accurately embedded into the reply text format and maintain the integrity and coherence of the reply content.

[0142] In step S602 of some embodiments, specifically, through the target reply text agent, each part in the reply content is placed in the corresponding text format structure position one by one according to the text format structure of the text format features, in a logical order and format requirements.

[0143] In this embodiment, by inserting the reply content into the reply text format in sequence according to the text format features and reply content features to obtain the target reply text, a target reply text with a complete structure, standardized format, and coherent content can be formed, which not only meets the requirements of the reply in form but also can clearly and accurately respond to the viewpoints in the examination opinions in content.

[0144] In an embodiment of the present application, a target examination opinion text is obtained; wherein, the target examination opinion text includes an examination opinion notice text, a comparison text, and an application text; the target examination opinion text is subjected to domain recognition by a pre-trained domain scheduling agent to obtain a target technical domain, and a target domain agent is determined from pre-trained candidate domain agents based on the target technical domain; wherein, the target domain agent includes at least two cascaded draft agents, a target quality review agent, and a target response text agent; a response content is generated by at least two cascaded draft agents for the target technical domain, the examination opinion notice text, the comparison text, and the application text; the response content is subjected to content quality scoring by the target quality review agent to obtain quality scores for multiple response book scoring criteria, and if at least one of the quality scores for the multiple response book scoring criteria does not meet a predetermined quality compliance condition, at least one of the at least two cascaded draft agents is optimized based on the response book scoring criteria that do not meet the predetermined quality compliance condition to regenerate the response content; if each of the quality scores for the multiple response book scoring criteria meets the predetermined quality compliance condition, a target response text is generated by the target response text agent for a preset response text format and the response content. In the embodiment of the present application, first, the domain scheduling agent is used to perform domain recognition on the target examination opinion text (specifically including the examination opinion notice text, the comparison text, and the application text), the target technical domain can be recognized, and the corresponding target domain agent is determined based on the target technical domain, realizing the dynamic role allocation of agents for different technical domains, without the need to manually input prompt words to instruct different artificial intelligence models to perform examination opinion analysis; second, through the collaborative work of at least two cascaded draft agents and the quality review agent in the target domain agent, patent analysis, examination opinion analysis, and response text generation can be achieved in one step, and the quality of the response content is gradually improved by combining a preset multiple response book scoring criteria and quality compliance conditions, realizing the continuous optimization of the response content, helping to improve the accuracy of response content generation, and also realizing the full process automation of response content generation; finally, the target response text is generated by the response text agent and the preset response text format, and the automation of all links in the process of response text generation, including patent analysis, examination opinion analysis, and response opinion generation, can be achieved, without relying on manual continuous input of prompt words to participate in the response generation process, significantly improving the efficiency of response text generation.

[0145] Please refer to Figure 7 , an embodiment of the present application further provides a response text generation device, which can implement the above response text generation method. The device includes:

[0146] An examination opinion text acquisition module for acquiring a target examination opinion text; wherein the target examination opinion text includes an examination opinion notice text, a comparison text, and an application text;

[0147] A domain scheduling agent module for identifying the domain of the target examination opinion text through a pre-trained domain scheduling agent to obtain the target technical domain, and determining a target domain agent from the pre-trained candidate domain agents based on the target technical domain; wherein the target domain agent includes at least two cascaded draft agents, a target quality review agent, and a target response text agent;

[0148] A response opinion generation module for generating response opinions on the target technical domain, the examination opinion notice text, the comparison text, and the application text through at least two cascaded draft agents to obtain a response content;

[0149] A response opinion optimization module for scoring the content quality of the response content through the target quality review agent to obtain quality scores for multiple response letter scoring criteria, and if at least one of the quality scores for the multiple response letter scoring criteria does not meet the predetermined quality compliance condition, optimizing at least one of the at least two cascaded draft agents based on the response letter scoring criteria that do not meet the predetermined quality compliance condition to regenerate the response content;

[0150] A response text generation module for generating a target response text from a preset response text format and the response content through the target response text agent if each of the quality scores for the multiple response letter scoring criteria meets the predetermined quality compliance condition.

[0151] The specific implementation manner of this response text generation device is basically the same as the specific embodiments of the above response text generation method, and will not be elaborated here.

[0152] An embodiment of the present application also provides an electronic device. The electronic device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the above response text generation method is implemented. The electronic device can be any intelligent terminal including a tablet computer, an in-vehicle computer, etc.

[0153] Please refer to Figure 8 , Figure 8 which shows the hardware structure of an electronic device in another embodiment. The electronic device includes:

[0154] The processor 801 can be implemented in ways such as a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present application;

[0155] The memory 802 can be implemented in forms such as a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 802 can store processing systems and other application programs. When implementing the technical solutions provided in the embodiments of this specification through software or firmware, the relevant program codes are stored in the memory 802 and are called by the processor 801 to execute the reply text generation method of the embodiments of the present application;

[0156] The input / output interface 803 is used to implement information input and output;

[0157] The communication interface 804 is used to implement communication interaction between this device and other devices, and can achieve communication through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.);

[0158] The bus 805 transmits information between the various components of the device (such as the processor 801, the memory 802, the input / output interface 803, and the communication interface 804);

[0159] Among them, the processor 801, the memory 802, the input / output interface 803, and the communication interface 804 achieve communication connections with each other inside the device through the bus 805.

[0160] The embodiments of the present application also provide a computer-readable storage medium, which stores a computer program, and when the computer program is executed by a processor, the above-mentioned reply text generation method is implemented.

[0161] As a non-transitory computer-readable storage medium, the memory can be used to store non-transitory software programs and non-transitory computer-executable programs. In addition, the memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state storage devices. In some embodiments, the memory may optionally include a memory remotely disposed relative to the processor, and these remote memories may be connected to the processor through a network. Examples of the above networks include but are not limited to the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0162] The reply text generation method, reply text generation device, electronic device, and storage medium provided by the embodiments of the present application first use a domain scheduling agent to perform domain identification on the target examination opinion text (specifically including the examination opinion notice text, the comparison text, and the application text), can identify the target technical field, and determine the corresponding target domain agent based on the target technical field, realizing the dynamic role assignment of agents for different technical fields; secondly, through the collaborative work of at least two cascaded draft agents and quality review agents in the target domain agent, patent analysis, examination opinion analysis, and reply text generation can be achieved in one step, and the quality of the reply content is gradually improved by combining a plurality of preset reply book scoring criteria and quality compliance conditions, realizing the continuous optimization of the reply opinion, which helps to improve the accuracy of the reply content generation; finally, the target reply text is generated through the reply text agent and the preset reply text format. It can be seen that the present application can automate all links of patent analysis, examination opinion analysis, and reply opinion generation in the reply text generation process, without relying on manual continuous input of prompt words to participate in the reply generation process, significantly improving the efficiency of reply text generation.

[0163] The embodiments described in the embodiments of the present application are for more clearly illustrating the technical solutions of the embodiments of the present application, and do not constitute a limitation on the technical solutions provided by the embodiments of the present application. Those skilled in the art know that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of the present application are equally applicable to similar technical problems.

[0164] Those skilled in the art can understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than shown in the figures, or combine certain steps, or different steps.

[0165] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, that is, they may be located in one place, or may be distributed to multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0166] Those of ordinary skill in the art can understand that all or some of the steps in the methods disclosed above, and the functional modules / units in systems and devices, can be implemented as software, firmware, hardware, and appropriate combinations thereof.

[0167] The terms "first", "second", "third", "fourth", etc. (if any) in the description of this application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily have to be used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of this application described here can be implemented in an order other than those illustrated or described here. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products, or devices.

[0168] It should be understood that in this application, "at least one (item)" means one or more, and "a plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships can exist. For example, "A and / or B" can mean: only A exists, only B exists, and both A and B exist at the same time. Among them, A and B can be singular or plural. The character " / " generally means that the associated objects before and after are in an "or" relationship. "At least one (one) of the following" or its similar expressions refer to any combination of these items, including any combination of single items (ones) or plural items (ones). For example, at least one (one) of a, b, or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0169] In several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only illustrative. For example, the above-mentioned unit division is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. The displayed or discussed coupling, direct coupling, or communication connection to each other can be through some interfaces. The indirect coupling or communication connection of devices or units can be in electrical, mechanical, or other forms.

[0170] The units described above as separate components may or may not be physically separated. The components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0171] In addition, the functional units in each embodiment of this application can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above integrated units can be implemented in the form of hardware or in the form of software functional units.

[0172] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods in each embodiment of this application. The aforementioned storage medium includes: USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs and other various media that can store programs.

[0173] The preferred embodiments of the embodiments of this application have been described above with reference to the accompanying drawings, but this does not limit the scope of the rights of the embodiments of this application. Any modification, equivalent replacement, and improvement made by those skilled in the art without departing from the scope and essence of the embodiments of this application shall be within the scope of the rights of the embodiments of this application.

Claims

1. A method for generating a response text, characterized in that, The method includes: Obtaining a target examination opinion text; wherein, the target examination opinion text includes an examination opinion notice text, a comparison text, and an application text; Performing domain recognition on the target examination opinion text through a pre-trained domain scheduling agent to obtain a target technical domain, and determining a target domain agent from pre-trained candidate domain agents based on the target technical domain; wherein, the target domain agent includes at least two cascaded draft agents, a target quality review agent, and a target response text agent; Generating response opinions for the target technical domain, the examination opinion notice text, the comparison text, and the application text through the at least two cascaded draft agents to obtain a response content; Performing content quality scoring on the response content through the target quality review agent to obtain quality scores for multiple response book scoring criteria, and if at least one of the quality scores for the multiple response book scoring criteria does not meet a predetermined quality compliance condition, optimizing at least one of the at least two cascaded draft agents based on the response book scoring criteria that do not meet the predetermined quality compliance condition to regenerate the response content; If each of the quality scores for the multiple response book scoring criteria meets the predetermined quality compliance condition, generating a response text for a preset response text format and the response content through the target response text agent to obtain a target response text.

2. The method according to claim 1, wherein The at least two cascaded draft agents include a target domain draft generation agent and a target draft polishing agent; The generating response opinions for the target technical domain, the examination opinion notice text, the comparison text, and the application text through the at least two cascaded draft agents to obtain a response content includes: Generating response opinions for the target technical domain, the examination opinion notice text, the comparison text, and the application text through the target domain draft generation agent to obtain an initial response content; wherein, the target domain draft generation agent is used to respond to the viewpoints in the examination opinion based on the comparison text and the application text; Optimizing the response opinions of the initial response content through the target draft polishing agent to obtain an optimized response content.

3. The method according to claim 2, wherein The at least two cascaded draft agents further include a target draft reflection agent between the target domain draft generation agent and the target draft polishing agent; The optimizing the response opinions of the initial response content through the target draft polishing agent to obtain an optimized response content includes: Performing content logic analysis on the initial response content through the target draft reflection agent to obtain content optimization ideas; Optimizing the response opinions of the initial response content based on the content optimization ideas through the target draft polishing agent to obtain the optimized response content.

4. The method according to claim 3, characterized in that, The optimizing at least one of the at least two cascaded draft agents based on the response book scoring criteria that do not meet the predetermined quality compliance condition includes: If the scoring criterion for the response letter that does not meet the predetermined quality compliance condition is the accuracy quality score of the response content, optimize the target domain draft generation agent and the target draft reflection agent in the at least two cascaded draft agents; If the scoring criterion for the response letter that does not meet the predetermined quality compliance condition is the compliance quality score of the response content, optimize the target domain draft generation agent and the target draft reflection agent in the at least two cascaded draft agents; If the scoring criterion for the response letter that does not meet the predetermined quality compliance condition is the logical quality score of the response content, optimize the target domain draft generation agent and the target draft reflection agent in the at least two cascaded draft agents.

5. The method according to claim 2, wherein The optimization of at least one of the at least two cascaded draft agents based on the scoring criterion for the response letter that does not meet the predetermined quality compliance condition further includes: If the scoring criterion for the response letter that does not meet the predetermined quality compliance condition is the language quality score of the response content, optimize the target domain draft generation agent and the target draft polishing agent in the at least two cascaded draft agents.

6. The method according to any one of claims 1 to 5, characterized in that, The obtaining of the target technical field by performing domain recognition on the target examination opinion text through the pre-trained domain scheduling agent includes: Performing text domain feature extraction on the target examination opinion text through the domain scheduling agent to obtain text domain features; Performing domain parsing on the text domain features to obtain the target technical field.

7. The method according to any one of claims 1 to 5, characterized in that The generation of the target response text by generating the response text through the target response text agent for the preset response text format and the response content includes: Performing format feature extraction on the response text format through the target response text agent to obtain text format features, and performing content feature extraction on the response content through the target response text agent to obtain response content features; Inserting the response content into the response text format in sequence according to the text format features and the response content features to obtain the target response text.

8. A response text generation device, characterized in that, The device includes: An examination opinion text acquisition module for acquiring a target examination opinion text; wherein, the target examination opinion text includes an examination opinion notice text, a comparison text, and an application text; A domain scheduling agent module for performing domain recognition on the target examination opinion text through a pre-trained domain scheduling agent to obtain a target technical field, and determining a target domain agent from pre-trained candidate domain agents based on the target technical field; wherein, the target domain agent includes at least two cascaded draft agents, a target quality review agent, and a target response text agent; A response opinion generation module for generating a response opinion on the target technical field, the examination opinion notice text, the comparison text, and the application text through the at least two cascaded draft agents to obtain a response content; The reply opinion optimization module is used to obtain quality scores of multiple reply book scoring criteria by scoring the content quality of the reply content through the target quality review intelligent agent, and if at least one of the quality scores of the multiple reply book scoring criteria does not meet the predetermined quality compliance condition, at least one of the at least two cascaded draft intelligent agents is optimized based on the reply book scoring criteria that do not meet the predetermined quality compliance condition to regenerate the reply content; The reply text generation module is used to generate a target reply text by generating a reply text for a preset reply text format and the reply content through the target reply text intelligent agent if each of the quality scores of the multiple reply book scoring criteria meets the predetermined quality compliance condition.

9. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory stores a computer program, and when the processor executes the computer program, the reply text generation method according to any one of claims 1 to 7 is implemented.

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

Images