AI agent question and answer method based on intellectual property application process management system

Abstract

The application relates to the technical field of intellectual property services, and provides an AI intelligent agent question and answer method based on an intellectual property application process management system, which comprises the following steps: creating a knowledge base management module in the system, editing intellectual property process matter question and answer content to form an AI intelligent agent knowledge base; fine-tuning training is conducted on a large language model carried by the AI intelligent agent by using the knowledge base, so that the AI intelligent agent can learn the answering ability for intellectual property application process matter questions; a user identity recognition module is created in the AI intelligent agent to distinguish whether a questioner is a system internal user or an external customer; based on the recognized user identity, when the AI intelligent agent receives an intellectual property process matter question, different answering strategies are provided for questioners with different identities. The application can help users to obtain reliable answers in time, facilitate external customers to understand the reply content, meet the professional work requirements of internal users, avoid internal process details or management strategies from being leaked, and effectively guarantee information security.

Description

Technical Field

[0001] This invention relates to the technical fields of intellectual property services, and in particular to an AI intelligent agent question-answering method based on an intellectual property application process management system. Background Technology

[0002] In cross-border intellectual property application practice, there is a significant amount of frequent and highly specialized communication between clients (such as applicants) and process management personnel (such as process specialists). For example, clients often ask questions such as "What is a small entity and a large entity?" and "What is the priority period?" while process management personnel need to quickly consult internal operating procedures, such as "How to supplement materials?" and "How to calculate the official fee payment deadline?"

[0003] Traditional communication methods primarily rely on human customer service or email, which suffers from slow response times, inconsistent standards, and high labor costs. To address this issue, some intellectual property management systems have begun to introduce intelligent question-and-answer (Q&A) systems. However, existing intelligent Q&A systems (such as the hierarchical answering method disclosed in Patent Document 1 (CN121279451A)) mainly solve the problem of unclear credibility of the answer results. Their technical solution is to generate different levels of answers (e.g., high confidence, low confidence) based on the degree of matching between the user's question and the document knowledge base, to help users judge the reliability of the answer. The hierarchical nature of this solution is essentially a hierarchical answer format based on the degree of matching (e.g., a definitive answer when the matching degree is high, and a probabilistic answer or hint when the matching degree is low), and it does not support information security hierarchical classification based on user identity. Furthermore, the Q&A functions in existing intellectual property management systems are typically static and fragmented. On the one hand, once the knowledge base is established, it is difficult to self-correct and iterate based on unknown questions or incorrect answers that arise in actual Q&A, causing the system to become outdated over time. On the other hand, the system provides consistent answers to all questioners (whether internal professional managers or external clients). This raises two issues: First, the excessive use of internal terminology and complex process descriptions for external clients leads to difficulties in understanding and a poor user experience; second, overly simplified answers are provided to internal staff, failing to meet their professional work needs. More importantly, this may leak process details or management strategies that should only be known internally, posing a risk to information security. Summary of the Invention

[0004] To address the shortcomings of the existing technologies, this invention provides an AI-powered question-and-answer method based on an intellectual property application process management system. This method helps users obtain reliable answers in a timely and convenient manner while using the intellectual property application process management system, facilitating understanding of responses by external clients. It also meets the professional work needs of internal users of the intellectual property application process management system, prevents the leakage of internal process details or management strategies, and ensures information security.

[0005] Firstly, the AI ​​intelligent agent question-answering method based on an intellectual property application process management system provided by this invention includes: A knowledge base management module is created within the intellectual property application process management system. The knowledge base management module is used to edit the Q&A content of intellectual property process matters, forming an AI intelligent body knowledge base. By fine-tuning and training the large language model carried by the AI ​​agent through the AI ​​agent knowledge base, the AI ​​agent learns the ability to answer questions about intellectual property process matters. A user identification module is created within the AI ​​intelligent body to identify users who ask questions to the AI ​​intelligent body, distinguishing whether the questioning user is an internal user of the intellectual property application process management system or an external customer; Based on the identified identity of the user asking the question, when an AI agent that has learned to provide professional answers to questions about intellectual property procedures receives questions about intellectual property procedures, it provides different answering strategies for users with different identities.

[0006] Secondly, the intellectual property application process management system provided by this invention supports the dynamic generation of trademark application forms from multiple countries, specifically including: Analyze the content of trademark application forms in different target countries, break down the form content into different reusable logical units, set up corresponding standardized controls for each logical unit and integrate them into the editor; The editor combines different logical unit controls into trademark application form templates that adapt to the requirements of different target countries, and generates corresponding template configuration data based on the trademark application form templates. When a trademark application form template for a specific target country is accessed, a trademark application information collection interface adapted to the input habits of Chinese users is generated based on the template configuration data corresponding to the trademark application form template. The trademark application information collection interface contains four independent collection sub-units: an applicant information collection unit, a trademark basic information collection unit, a trademark supplementary information collection unit, and a trademark category information collection unit, used to collect basic trademark application data in Chinese format. While the user is inputting the basic trademark application data through the trademark application information collection interface, they can ask questions to the AI ​​agent to obtain answers to trademark application process questions from the AI ​​agent. The basic Chinese-format trademark application data collected from the applicant information collection unit, trademark basic information collection unit, trademark supplementary information collection unit, and trademark category information collection unit are used to automatically generate a trademark application form that conforms to the official specifications of the corresponding target country, according to the template configuration data.

[0007] Compared with the prior art, the beneficial effects of this invention are as follows: This invention provides an AI agent question-answering method based on an intellectual property application process management system, comprising: creating a knowledge base management module within the intellectual property application process management system; editing question-and-answer content for intellectual property process matters through the knowledge base management module to form an AI agent knowledge base; fine-tuning and training a large language model mounted on the AI ​​agent using the AI ​​agent knowledge base, enabling the AI ​​agent to learn the ability to answer questions about intellectual property process matters professionally; creating a user identity recognition module within the AI ​​agent to identify the user who initiates the question, distinguishing whether the questioning user is an internal user of the intellectual property application process management system or an external customer; based on the identified questioning user's identity, when the AI ​​agent, having learned the ability to answer questions about intellectual property process matters, receives a question about intellectual property process matters, it provides different answering strategies for questioning users with different identities. This invention can help users obtain reliable answers in a timely and convenient manner while using the intellectual property application process management system, facilitate external customers' understanding of the answers, meet the professional work needs of internal users of the intellectual property application process management system, avoid leakage of internal process details or management strategies, and ensure information security. Attached Figure Description

[0008] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. Some specific embodiments of the invention will be described in detail below with reference to the accompanying drawings in an exemplary and non-limiting manner. The same reference numerals in the drawings designate the same or similar parts or components. It should be understood by those skilled in the art that these drawings are not necessarily drawn to scale. In the drawings: Figure 1 This is a flowchart illustrating an AI intelligent agent question-answering method based on an intellectual property application process management system according to an embodiment of the present invention; Figure 2 This is a schematic diagram illustrating a process for generating multi-country trademark application forms according to an embodiment of the present invention. Detailed Implementation

[0009] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are merely some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention. Example

[0010] See Figures 1-2 This embodiment provides an AI intelligent agent question-answering method based on an intellectual property application process management system, including the following steps: S101. Create a knowledge base management module within the intellectual property application process management system. Edit the Q&A content for intellectual property process matters through the knowledge base management module to form an AI intelligent body knowledge base. S102. Fine-tune and train the large language model carried by the AI ​​agent through the AI ​​agent knowledge base, so that the AI ​​agent can learn the ability to answer questions about intellectual property process matters. S103. Create a user identity recognition module within the AI ​​intelligent body to identify the user who asks a question to the AI ​​intelligent body and distinguish whether the questioning user is an internal user of the intellectual property application process management system or an external customer. S104. Based on the identified identity of the questioning user, when the AI ​​agent, which has learned to provide professional answers to questions about intellectual property procedures, receives questions about intellectual property procedures, it provides different answering strategies for questioning users with different identities.

[0011] It should be noted that the AI ​​intelligent agent question-answering method based on the intellectual property application process management system provided in this embodiment can help users obtain reliable answers in a timely and convenient manner during the use of the intellectual property application process management system, making it easier for external customers to understand the answers, while also meeting the professional work needs of internal users of the intellectual property application process management system, avoiding the leakage of internal process details or management strategies, and ensuring information security.

[0012] A knowledge base management module is created within the intellectual property application process management system. This module provides system administrators with a graphical or configuration file-based editing interface. Through this interface, administrators can create, edit, categorize, review, and publish Q&A content related to intellectual property process matters. Each Q&A typically includes a standard question, an authoritative answer, related keywords, the applicable process stage (e.g., application submission, response to examination opinions), an index of relevant legal provisions (e.g., pointing to specific provisions of trademark law or examination guidelines), and a user identity tag applicable to the Q&A content (e.g., set to be visible only to internal personnel or to clients). All Q&A content created through the knowledge base management module collectively constitutes the AI ​​intelligent agent knowledge base.

[0013] The AI ​​agent knowledge base is a structured database, such as a relational database or a vector database. It not only stores and manages all question-and-answer pairs generated by the knowledge base management module, but also stores related document fragments (such as the text content of internal operating procedures, summaries of historical case processing records), as well as anonymized common process patterns and rules extracted from the business database of the intellectual property application process management system. This knowledge base is the core data source for subsequent fine-tuning and training of the large language model, and for the AI ​​agent to retrieve relevant information at runtime.

[0014] The AI ​​agent is powered by a large language model as its core inference engine. To enable this large language model to learn professional answers to questions regarding intellectual property procedures, it needs to be supervised and fine-tuned using the content of the aforementioned AI agent knowledge base. The specific training steps are as follows: First, extract all question-answer pairs marked as high-quality from the AI ​​agent knowledge base as basic training data. Simultaneously, slice the text materials in the knowledge base, such as process rule documents and internal operation manuals, into text fragments suitable for model learning. Construct each question-answer pair or text fragment into a pre-formatted data object, removing duplicate or ambiguous data entries. To further enhance the model's generalization ability, synonym replacement and sentence transformation can be performed on some question data to expand the training dataset, ultimately forming a high-quality supervised fine-tuning dataset. Next, load the pre-trained weights of the large language model (e.g., Tencent's Mixed-Source Model). Employ parameter-efficient fine-tuning methods, such as low-rank adaptation, to reduce computational overhead and prevent the model from forgetting existing general knowledge. This method is achieved by injecting trainable rank decomposition matrices into the model's self-attention layer and feedforward network layer. In practice, the rank of the low-rank adaptation matrix can be set to eight. The training process uses a deep learning framework, with the following key hyperparameters set: learning rate of 0.002%, batch size of 16, number of training epochs of 3, and the AdamW optimizer. The training objective is to minimize the cross-entropy loss between the model-generated answers and the standard answers in the training data. During training, a reserved validation set is used to monitor the model's loss and generation quality to prevent overfitting. After training, the fine-tuned model is evaluated using an independent test set containing flow problems not encountered during training. Evaluation metrics include answer accuracy (measured by matching against standard answers in the knowledge base or by expert scoring), security, and language fluency. Models that pass the evaluation are deployed as the inference engine for the AI ​​agent, typically packaged as a model server providing application programming interface services.

[0015] A user identification module is created within the AI ​​agent, which integrates with the existing user authentication and permission system of the intellectual property application process management system. When a user asks a question to the AI ​​agent through a client (e.g., a webpage or mobile application), the user identification module obtains the user's credentials (e.g., a login token) for the current session. By querying the user database of the intellectual property application process management system, the user identification module identifies the user's attributes, with the key distinction based on the user type field. If the user type belongs to a predefined internal role such as administrator, process specialist, or reviewer, they are identified as an internal user. If the user type is an applicant, they are identified as an external customer. For anonymous access without logging in, they can be identified as an external customer by default, or only limited public Q&A services can be provided to them.

[0016] Based on the identified user's identity, when an AI agent, which has been fine-tuned and trained to provide professional answers, receives a question about intellectual property procedures from a user, it performs the following steps to provide answers with different strategies.

[0017] First, the AI ​​agent treats the user's natural language question as a query and performs semantic retrieval (e.g., using vector retrieval technology) in its knowledge base to obtain the most relevant knowledge fragments, including question-answer pairs and related document fragments. Simultaneously, if the user asking the question is identified as a logged-in external client, the AI ​​agent will retrieve basic status information of the client's current relevant cases from the business database of the intellectual property application process management system via a secure interface (with permissions limited to querying the client's own cases) as supplementary context.

[0018] The AI ​​agent then combines multiple elements into input prompts for a large language model. These elements include: system instructions that clearly define the AI ​​assistant's role, the current user's identity (e.g., "internal process specialist" or "applicant client"), relevant knowledge retrieved from the knowledge base as reference information, and relevant case status information for external clients. The system instructions are differentiated based on the user's identity: for internal users, instructions emphasize the comprehensiveness, professionalism, and operability of the answers, allowing the inclusion of internal code, unpublished review standards, or guidance to contact specific departments; for external clients, instructions emphasize the friendliness, clarity, and compliance of the answers, and strictly filter sensitive internal information, with answers based on general processes visible to the client and their own case status for personalized guidance.

[0019] Finally, the AI ​​agent, loaded with the finely tuned large language model, generates a natural language response that conforms to the preset strategy based on the complete prompts containing identity information and contextual knowledge, and presents it to the user through the front-end interface of the intellectual property application process management system.

[0020] Preferably, the AI ​​agent question-answering method further includes: An abnormal question and answer management module is created within the intellectual property application process management system. The abnormal question and answer management module is used to capture abnormal situations that occur during the question and answer interaction between the user and the AI ​​agent. The AI ​​agent knowledge base is modified and updated based on the captured anomalies, forming a new AI agent knowledge base. The large language model carried by the AI ​​agent is then continuously fine-tuned and trained using the new AI agent knowledge base.

[0021] It should be noted that by capturing abnormal situations during the interaction between the user and the AI ​​agent, and using this information to drive the correction, updating and continuous fine-tuning of the AI ​​agent's knowledge base and model, the system can achieve self-learning and evolution, thus solving the shortcomings of intelligent question-answering systems such as knowledge solidification, difficulty in adapting to dynamic changes in processes and new question scenarios.

[0022] Within the intellectual property application process management system, an integrated abnormal question-and-answer management module can monitor the entire process of user interaction with the AI ​​agent in real time. When pre-defined abnormal situations occur, such as the AI ​​agent returning a "cannot answer" message or the user giving a "negative review" to the answer, the abnormal question-and-answer management module will automatically intercept the complete dialogue content of the current round (including the user's question and the AI's answer) and store it as an abnormal question-and-answer record in a dedicated abnormal question-and-answer list. The system periodically (e.g., daily) or manually triggered by the administrator will perform analysis tasks on the abnormal question-and-answer list. The analysis process can combine natural language processing technology to cluster and perform root cause analysis on abnormal issues to determine whether the problem is due to errors or unclear expressions in existing question-and-answer content in the knowledge base, or the appearance of entirely new knowledge points missing from the knowledge base. Based on the judgment results, the system will generate corresponding knowledge base correction or supplement suggestions and push them to the system administrator for review. After the administrator approves the suggestions, they can use the knowledge base management module to correct (e.g., modify incorrect answers) or supplement (e.g., add new question-and-answer pairs) the corresponding content in the AI ​​agent's knowledge base. Once the knowledge base is updated, the system automatically initiates an incremental fine-tuning training process for the large language model mounted on the AI ​​agent. This process uses newly added or modified high-quality data to perform a small number of training rounds with a low learning rate, enabling the model to quickly learn new knowledge or correct existing errors while preserving its original capabilities to the greatest extent possible. After successful training evaluation, the new model is hot-uploaded to the production environment, thus completing a full optimization cycle and forming a continuously improving AI agent question-answering system.

[0023] Preferably, the step of editing the question-and-answer content of intellectual property process matters through the knowledge base management module to form an AI intelligent agent knowledge base includes: Collect knowledge points or procedural matters related to intellectual property procedures, including standard Q&A documents, operation manuals, and frequently asked questions; By using the knowledge base management module, for each knowledge point or process item, relevant questions and corresponding standard answers can be edited and created to obtain well-established question-and-answer pairs. The AI ​​agent's learning interface is invoked, enabling the AI ​​agent to vectorize the established question-and-answer pairs and store them in the knowledge base, thus forming the AI ​​agent's knowledge base.

[0024] It should be noted that this embodiment clarifies the specific process of knowledge source, editing operation, and vectorized storage. It transforms scattered unstructured knowledge such as standard documents and operation manuals into structured, machine-understandable question-and-answer pairs through systematic editing, and utilizes the AI ​​agent's learning interface for vectorized processing and storage. This achieves efficient and standardized input and storage of domain-specific knowledge, providing a high-quality, uniformly formatted data foundation for subsequent semantic retrieval and model fine-tuning. By converting natural language knowledge into vectors—a model-friendly form—the efficiency and accuracy of knowledge retrieval can be improved, thereby ensuring the professionalism and response speed of the AI ​​agent's answers.

[0025] Preferably, providing different answer strategies to users with different identities includes: The answers provided to external clients are those concerning intellectual property procedures that omit complex internal operational details and include conceptual explanations and macro-level process progress. The answers provided to internal users include professional terminology, specific operational guidelines, reminders of internal rules, and answers to intellectual property process matters related to deadline calculations.

[0026] It should be noted that, in this embodiment, based on user identification, the response content can be precisely customized in terms of information granularity, professional depth, and expression. For external customers, the response focuses on explaining concepts and macro-level processes, avoiding internal operational details and sensitive information. This helps improve customer experience and ensures that information is conveyed clearly, securely, and easily understood. For internal users, the response includes in-depth information such as professional terminology, specific operating steps, internal rule reminders, and even deadline calculations, directly supporting them in completing their work efficiently and accurately. This differentiated content strategy provided in this embodiment allows a single AI agent to simultaneously meet the distinct information needs of two user groups, improving system usability and user satisfaction while also strengthening the control over internal information.

[0027] Preferably, during the question-and-answer interaction between the user and the AI ​​agent, the abnormal question-and-answer management module monitors the question-and-answer process between the AI ​​agent and the user. If a preset abnormal question-and-answer situation occurs, the abnormal question-and-answer management module automatically intercepts the question-and-answer content between the user and the AI ​​agent as abnormal question-and-answer content and records it in the abnormal question-and-answer list.

[0028] It should be noted that the abnormal question and answer management module monitors in real time and automatically intercepts and records dialogue content when preset abnormal situations occur, thereby achieving proactive and real-time capture of abnormal interactions and avoiding the inefficiency and omissions of manual screening. Automatically recording abnormal content into a structured list provides a clear and centralized data source for subsequent systematic analysis.

[0029] Preferably, the abnormal question and answer content recorded in the abnormal question and answer list is analyzed to determine whether the original question and answer content is incorrect or whether new knowledge points related to intellectual property process matters are missing; based on different judgment results, the AI ​​intelligent agent knowledge base is corrected or supplemented accordingly through the knowledge base management module.

[0030] It should be noted that in this embodiment, intelligent judgment of the causes of anomalies (distinguishing between content errors and knowledge gaps) and targeted measures (correction or supplementation) based on different judgment results enable the system's self-optimization to move beyond simple data stacking and acquire preliminary diagnostic capabilities. This allows for precise identification of the root causes of knowledge base problems, enabling the adoption of the most effective update strategy and avoiding blind operations. For example, correcting existing errors can quickly rectify system deviations, while supplementing new knowledge points can expand the system's capability boundaries. This differentiated processing strategy based on cause analysis provided in this embodiment can significantly improve the efficiency and accuracy of the knowledge base optimization process, making the system's continuous learning process more intelligent and effective.

[0031] Preferably, the abnormal question-and-answer situations include: the current AI agent returning a prompt message indicating that it cannot answer, or the user actively giving negative feedback on the current AI agent's answer.

[0032] It's important to note that treating the AI ​​agent's inability to answer as an anomaly effectively identifies knowledge gaps in the system; similarly, treating negative user reviews as an anomaly directly reflects users' subjective dissatisfaction with the quality of answers, serving as a crucial channel for discovering inaccurate, unclear, or irrelevant responses. Combining these two approaches, from the perspectives of system self-awareness and user feedback respectively, constructs a relatively comprehensive and reliable anomaly detection mechanism. Example

[0033] See Figures 1-2 In this embodiment, the intellectual property application process management system supports the dynamic generation of trademark application forms from multiple countries, specifically including: S301. Analyze the form content of trademark application forms in different target countries, break down the form content into different reusable logical units, set up corresponding standardized controls for each logical unit and integrate them into the editor; S302. Using the editor, different logical unit controls are combined into trademark application form templates that adapt to the requirements of different target countries, and corresponding template configuration data is generated based on the trademark application form templates. S303. When the trademark application form template of a specific target country is accessed, a trademark application information collection interface adapted to the input habits of Chinese users is generated based on the template configuration data corresponding to the trademark application form template. The trademark application information collection interface includes four independent collection sub-units: applicant information collection unit, trademark basic information collection unit, trademark supplementary information collection unit, and trademark category information collection unit, which are used to collect basic trademark application data in Chinese format. When the user inputs the basic trademark application data through the trademark application information collection interface, they can ask questions to the AI ​​agent to obtain answers from the AI ​​agent regarding trademark application process issues. S304. The basic data of Chinese-format trademark applications collected by the applicant information collection unit, trademark basic information collection unit, trademark supplementary information collection unit and trademark category information collection unit are configured according to the template to automatically generate a trademark application form that conforms to the official specifications of the corresponding target country.

[0034] In this embodiment, by splitting logical units and integrating them into controls, form elements of trademark application forms from different countries can be reused, avoiding the repetitive work of developing separate forms for each country and improving template configuration efficiency. Furthermore, by configuring data through templates, flexible adaptation to form specifications of different countries can be achieved. When adjusting the form rules of the target country, no modification to the underlying system code is required; only configuration adjustments are needed for iteration. Simultaneously, by adapting the data collection interface to the habits of Chinese users, the barrier of directly selecting complex forms for the target country can be avoided, effectively reducing the error rate. In addition, by automatically converting data to generate application forms for the target country, no manual formatting or field reorganization is needed, effectively improving the efficiency of trademark application form creation.

[0035] It should be noted that in traditional trademark application processes, users need to switch scenarios to search for process information when filling in information, leading to operational interruptions, high information entry error rates, and low application efficiency. In this embodiment, users can initiate questions without exiting the intellectual property application process management system, avoiding operational interruptions caused by switching systems and searching for information, thus improving the smoothness of information entry and user experience. The AI ​​agent provides real-time answers based on a professional trademark application knowledge base, and can provide scenario-based professional guidance for the data collection sub-units currently filled in by the user (such as applicant information and trademark category). For example, when a user asks about the classification of a certain product while filling in trademark category information, the AI ​​can provide accurate suggestions based on the currently collected trademark purpose and the classification standards of the target country, avoiding application rejection or correction due to classification errors, and effectively improving the accuracy of the basic application data. For users lacking experience in multi-country trademark applications (such as SMEs and individual applicants), there is no need to systematically learn the complex application rules of different target countries in advance. They can obtain targeted answers instantly during the filling process, lowering the professional knowledge threshold for trademark applications and enabling more entities to easily carry out multi-country trademark layout.

[0036] Furthermore, to facilitate user access, a trademark application form template management interface can be configured. This interface provides access interfaces for trademark application templates from multiple target countries. Users can initiate access to the trademark application form templates for a specific target country simply by selecting the desired target country's access interface.

[0037] In some examples, suppose we want to facilitate a Chinese company's trademark registration application in target country A. First, the system pre-collects trademark application forms officially released by the target country, breaking down the form content into four reusable logical units: applicant information, basic trademark information, supplementary trademark information, and trademark category information. For each logical unit, a corresponding standardized input control is developed and integrated into a visual editor. Second, users can drag and drop these four logical unit controls in the visual editor to create a dedicated template for trademark applications in country A. The system automatically generates corresponding template configuration data, which can pre-define the mapping and conversion rules between the Chinese user's input fields and the target country's required fields. Third, when a Chinese company accesses the country A trademark application template, the system recreates a fully Chinese trademark application information collection interface based on the configuration data. The user selects basic data such as applicant name, address, application date, trademark image, and application category according to Chinese conventions, without needing to understand the target country's form specifications. Finally, the system automatically reorganizes and converts the Chinese-formatted data selected by the user according to the configuration rules. For example, the address is adjusted to the order required by the target country, and the date is converted to country A's format, automatically generating a trademark application form that conforms to the target country's official specifications.

[0038] Preferably, the template configuration data pre-records mapping rules and data format conversion rules between Chinese input fields and target country form fields that are adapted to the requirements of the target country's form. When automatically generating a trademark application form that conforms to the official specifications of the corresponding target country according to the template configuration data, the specific steps are as follows: the fields are reorganized and the format is converted according to the mapping rules and format conversion rules recorded in the template configuration data to automatically generate a trademark application form that conforms to the official specifications of the corresponding target country.

[0039] Preferably, the step of splitting the form content of trademark application forms from different target countries into different reusable logical units includes: hierarchically splitting according to the field attributes, layout logic, and required / optional attributes of the official forms of the target countries, dividing functionally related continuous fields into the same logical unit, with each logical unit corresponding to independent field validation rules, display logic, and data storage path, and synchronously recording the correspondence between each logical unit and the fields of the official forms of the target countries during the splitting process.

[0040] It's important to note that the logical unit splitting rules can resolve issues like field mismatches and omissions that easily occur during template combination, ensuring the functional independence and field correspondence of each logical unit and improving the accuracy of template configuration. Continuing with the example of a Chinese company applying for trademark registration in target country A, the system splits the target country's trademark application form hierarchically according to the form's field attributes, layout logic, and required / optional attributes. It groups consecutive fields such as applicant's name / title, applicant's address, contact email, and contact phone number into the same applicant information logical unit. This logical unit is configured with independent field validation rules (e.g., contact email must conform to email format, Chinese address must include province, city, and district information), Chinese display logic (all field prompts must be in Chinese), and data storage path (all applicant information is stored in the applicant information database table). The splitting process simultaneously records the correspondence between the logical unit and the official form fields of the target country. When combining templates, the system automatically checks whether the applicant information logical unit has been added; if missing, a prompt appears to prevent the generated template from lacking necessary fields.

[0041] Preferably, the template configuration data is stored in a distributed structure, including two independent storage partitions: a general rule base and a country-specific personalized rule base. The general rule base stores field mapping and format conversion rules common to all target countries, while the country-specific personalized rule base stores form adaptation rules unique to the corresponding target country. When the official form rules of the target country are adjusted, the content in the corresponding country-specific personalized rule base is updated to complete the template iteration.

[0042] It should be noted that the distributed storage structure of the template configuration data can solve the problem of needing to modify the entire template and incurring high iteration costs when adjusting the form rules of the target country. It improves the efficiency of template updates, and rule iteration can be completed without pausing system services, greatly reducing system maintenance costs.

[0043] The template configuration data can be stored in a cloud-distributed system, divided into two independent storage partitions: a general rule base and a country-specific personalized rule base. The general rule base stores rules for special character escaping, required field validation, and sensitive word filtering that are common to all target countries. The country-specific personalized rule base stores form adaptation rules unique to different target countries. When a target country modifies the option rules for the applicant type, the operations and maintenance personnel only need to update the corresponding content in the country-specific personalized rule base. All trademark application templates will take effect immediately without modifying the editor's underlying logic or the content of the general rule base, and without suspending system services.

[0044] Preferably, in the applicant information collection unit, the trademark basic information collection unit, the trademark supplementary information collection unit, and the trademark category information collection unit, each collection unit supports step-by-step selection, real-time verification, and automatic temporary storage functions. During the collection process, Chinese prompts and examples of common Chinese formats are used to guide the user's input.

[0045] It should be noted that the hierarchical sub-unit structure of the data collection interface can solve the problems of information confusion and easy omissions when users select information, improve the accuracy of information collection, shorten the selection time, and lower the selection threshold for users. Specifically, the trademark application information collection interface adapted for Chinese users includes four independent collection sub-units: the applicant information collection unit prompts users to select the applicant type (enterprise / individual), company name, Chinese address, and contact information according to Chinese usage; the trademark basic information collection unit prompts users to upload a trademark image, select the trademark type (text / graphic / combination), and select a trademark design description; the trademark supplementary information collection unit prompts users to upload evidence of trademark use and select the application type (already used / intended use); and the trademark category information collection unit prompts users to check the trademark category to be applied for and select the corresponding goods / services description. Each collection unit supports step-by-step selection, real-time verification, and automatic saving. After exiting and re-entering, users can continue from the last selected position. All prompts are in Chinese, eliminating the need for users to understand the form requirements of the target country.

[0046] Preferably, when N joint applicants need to be added, the applicant information collection unit generates an independent applicant information collection subpage for each joint applicant. After collection, the information of multiple applicants is sorted, merged, and format converted according to the requirements of the target country's form, and automatically filled into the corresponding joint applicant field position in the target country's form.

[0047] It should be noted that dynamic adaptation for multiple joint applicants can solve the problem of manually organizing information, adjusting sorting and formatting of multiple applicants in multi-applicant scenarios, improving the processing efficiency of multi-applicant applications and avoiding application risks caused by manual sorting errors. Taking a Chinese company and two related companies jointly applying for a trademark in Country A as an example, when the user clicks the applicant editing icon on the data collection interface, the system automatically generates a separate subpage for collecting applicant information. The user selects the applicant's Chinese address and other information. After submission, the system automatically sorts the information according to the order of first applicant, joint applicant 1, and joint applicant 2, converts all the applicant information in Chinese format to Country A format, and automatically fills it into the corresponding joint applicant field in the target country's form, without requiring the user to manually adjust the information order or format.

[0048] Preferably, the process of reorganizing and converting fields according to the mapping rules and format conversion rules recorded in the template configuration data includes: identifying the field types and content attributes of the collected Chinese format trademark application basic data, matching the corresponding mapping rules in the template configuration data, and directly completing the field content migration if there is a one-to-one mapping relationship, and completing the content reorganization according to the preset field merging and splitting rules if there is a many-to-one or one-to-many mapping relationship.

[0049] It should be noted that matching field mapping and format conversion can solve the problem of content mismatch that easily occurs when fields are mapped in a many-to-one or one-to-many manner, improving the accuracy of data conversion and reducing the application risk caused by conversion errors. For example, the system recognizes the four independent fields of "China" selected by the user, matches them with the many-to-one mapping rules of the target country's format fields in the template configuration data, and automatically merges the contents of the four fields into the target country's format fields and fills them into the corresponding positions.

[0050] Preferably, after the step of automatically generating a trademark application form that conforms to the official specifications of the target country, a dual verification step is also included. The first verification is a format verification, which automatically checks whether the fields, format, and layout of the generated form fully conform to the official requirements of the target country. The second verification is a content verification, which automatically compares whether the content of the generated form is consistent with the original content entered by the user. After all verifications pass, the form can be downloaded.

[0051] It should be noted that the dual verification after the application form is generated can resolve application risks caused by conversion errors and content omissions, thereby improving the compliance rate of the application form. For example, after the system automatically generates a trademark application form for country A, it first initiates a first-level format verification, automatically checking whether the number of fields, layout, date format, address format, and evidence format of the generated form comply with the latest official requirements of the target country. If there are format errors, they are automatically corrected and a correction log is generated. Then, a second-level content verification is initiated, automatically comparing whether the converted applicant name, trademark image, category information, and evidence of use are completely consistent with the content originally entered by the user. If there are content differences, a difference prompt will pop up to guide the user to confirm. Only after both verifications pass can the user's client download the application form.

[0052] It should be noted that the above embodiments are merely preferred embodiments of the present invention, and the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention, and the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. An AI-powered question-answering method based on an intellectual property application process management system, characterized in that, include: A knowledge base management module is created within the intellectual property application process management system. The knowledge base management module is used to edit the Q&A content of intellectual property process matters, forming an AI intelligent body knowledge base. By fine-tuning and training the large language model carried by the AI ​​agent through the AI ​​agent knowledge base, the AI ​​agent learns the ability to answer questions about intellectual property process matters. A user identification module is created within the AI ​​intelligent body to identify users who ask questions to the AI ​​intelligent body, distinguishing whether the questioning user is an internal user of the intellectual property application process management system or an external customer; Based on the identified identity of the user asking the question, when an AI agent that has learned to provide professional answers to questions about intellectual property procedures receives questions about intellectual property procedures, it provides different answering strategies for users with different identities.

2. The AI ​​intelligent agent question-answering method based on the intellectual property application process management system according to claim 1, characterized in that, The AI ​​intelligent agent question-answering method also includes: An abnormal question and answer management module is created within the intellectual property application process management system. The abnormal question and answer management module is used to capture abnormal situations that occur during the question and answer interaction between the user and the AI ​​agent. The AI ​​agent knowledge base is modified and updated based on the captured anomalies, forming a new AI agent knowledge base. The large language model carried by the AI ​​agent is then continuously fine-tuned and trained using the new AI agent knowledge base.

3. The AI ​​intelligent agent question-answering method based on the intellectual property application process management system according to claim 1, characterized in that, The process of editing Q&A content related to intellectual property procedures through the knowledge base management module to form an AI intelligent agent knowledge base includes: Collect knowledge points or procedural matters related to intellectual property procedures, including standard Q&A documents, operation manuals, and frequently asked questions; By using the knowledge base management module, for each knowledge point or process item, relevant questions and corresponding standard answers can be edited and created to obtain well-established question-and-answer pairs. The AI ​​agent's learning interface is invoked, enabling the AI ​​agent to vectorize the established question-and-answer pairs and store them in the knowledge base, thus forming the AI ​​agent's knowledge base.

4. The AI ​​intelligent agent question-answering method based on the intellectual property application process management system according to claim 1, characterized in that, The provision of different answering strategies to users with different identities includes: The answers provided to external clients are those concerning intellectual property procedures that omit complex internal operational details and include conceptual explanations and macro-level process progress. The answers provided to internal users include professional terminology, specific operational guidelines, reminders of internal rules, and answers to intellectual property process matters related to deadline calculations.

5. The AI ​​intelligent agent question-answering method based on the intellectual property application process management system according to claim 2, characterized in that, During the question-and-answer interaction between the user and the AI ​​agent, the abnormal question-and-answer management module monitors the question-and-answer process between the AI ​​agent and the user. If a preset abnormal question-and-answer situation occurs, the abnormal question-and-answer management module automatically intercepts the question-and-answer content between the user and the AI ​​agent as abnormal question-and-answer content and records it in the abnormal question-and-answer list.

6. The AI ​​intelligent agent question-answering method based on the intellectual property application process management system according to claim 5, characterized in that, The abnormal Q&A content recorded in the abnormal Q&A list is analyzed to determine whether the original Q&A content is incorrect or whether new knowledge points related to intellectual property procedures are missing. Based on the different judgment results, the AI ​​agent knowledge base is corrected or supplemented accordingly through the knowledge base management module.

7. The AI ​​intelligent agent question-answering method based on the intellectual property application process management system according to claim 5, characterized in that, The abnormal question-and-answer situations include: the current AI agent returning a message indicating that it cannot answer, or the user actively giving negative feedback on the current AI agent's answer.

8. The AI ​​intelligent agent question-answering method based on an intellectual property application process management system according to any one of claims 1-7, characterized in that, The intellectual property application process management system supports the dynamic generation of trademark application forms for multiple countries, specifically including: Analyze the content of trademark application forms in different target countries, break down the form content into different reusable logical units, set up corresponding standardized controls for each logical unit and integrate them into the editor; The editor combines different logical unit controls into trademark application form templates that adapt to the requirements of different target countries, and generates corresponding template configuration data based on the trademark application form templates. When a trademark application form template for a specific target country is accessed, a trademark application information collection interface adapted to the input habits of Chinese users is generated based on the template configuration data corresponding to the trademark application form template. The trademark application information collection interface contains four independent collection sub-units: an applicant information collection unit, a trademark basic information collection unit, a trademark supplementary information collection unit, and a trademark category information collection unit, used to collect basic trademark application data in Chinese format. While the user is inputting the basic trademark application data through the trademark application information collection interface, they can ask questions to the AI ​​agent to obtain answers to trademark application process questions from the AI ​​agent. The basic Chinese-format trademark application data collected from the applicant information collection unit, trademark basic information collection unit, trademark supplementary information collection unit, and trademark category information collection unit are used to automatically generate a trademark application form that conforms to the official specifications of the corresponding target country, according to the template configuration data.

9. The AI ​​intelligent agent question-answering method based on the intellectual property application process management system according to claim 8, characterized in that, The template configuration data pre-records mapping rules and data format conversion rules between Chinese input fields and target country form fields that are adapted to the requirements of the target country's form. When automatically generating a trademark application form that conforms to the official specifications of the corresponding target country according to the template configuration data, the specific steps are as follows: the fields are reorganized and the format is converted according to the mapping rules and format conversion rules recorded in the template configuration data to automatically generate a trademark application form that conforms to the official specifications of the corresponding target country.

10. The AI ​​intelligent agent question-answering method based on the intellectual property application process management system according to claim 8, characterized in that, The step of splitting the form content of trademark application forms from different target countries into different reusable logical units includes: hierarchically splitting according to the field attributes, layout logic, and required / optional attributes of the official forms of the target countries; dividing functionally related continuous fields into the same logical unit; each logical unit corresponds to independent field validation rules, display logic, and data storage path; and synchronously recording the correspondence between each logical unit and the fields of the official forms of the target countries during the splitting process.

Images