System and method for creating robot field natural language reasoning question-answering by using large model

Abstract

The invention relates to a robot field natural language reasoning question-answering system created by using a large model, and the system is constructed by the following steps: S1, exporting an entity table and an intention rule table from an existing knowledge graph, and formulating an NLU module in combination with a regular matching and similarity algorithm; s2, accessing an off-line large model chatglm and an on-line stellar fire large model, and constructing the use of a plurality of scenes; s3, constructing a question rewriting module to enable the question to better conform to the questioning standard of the large model; s4, a web socket communication module is constructed, and streaming output is realized; and S5, generating a question answering system in combination with the NLU, the large model and the question rewriting module. The invention further provides a question answering method using the question answering system. The accuracy and efficiency of the question-answering system in the field of robot equipment can be effectively improved, and the user experience is remarkably improved. The system not only can quickly respond to questions of the user, but also can provide detailed and accurate answers, so that diversified requirements of the user are met.

Description

Technical Field

[0001] The present invention relates to the field of robotic equipment, and in particular to a knowledge reasoning question and answer method and system that utilizes a large model and combines NLU natural language understanding. Background Art

[0002] With the rapid development of robotics, the complexity and automation of manufacturing equipment are increasing, which makes equipment operation, maintenance and fault diagnosis more difficult. Traditional operation manuals and maintenance guides are often unable to cope with the complex and changing manufacturing environment. Operators need to rely on rich experience and expertise when facing equipment failures. However, this method is time-consuming and error-prone, and cannot fully utilize the advantages of modern information technology.

[0003] In order to improve the efficiency and accuracy of operation and maintenance, the application of natural language understanding (NLU) technology in the field of robotics has gradually attracted attention. NLU technology can understand and process entities, relationships and attributes in user language, thereby effectively extracting the key content of the question. However, the development of question-answering systems relying solely on NLU technology is not comprehensive and in-depth enough, especially when dealing with complex professional field problems, the ability of NLU is limited.

[0004] In recent years, natural language processing technology based on large models (such as GPT-4) has made significant progress. These large models have strong semantic understanding and reasoning capabilities and can generate accurate and detailed answers. Combining large models with the field of robotics can better solve problems in equipment operation, maintenance, and fault diagnosis.

[0005] Specifically, NLU technology is used to pre-process the questions input by users, extract key entities and attributes, and then input these filtered keywords into the knowledge graph for query. The knowledge graph can provide detailed information and background knowledge related to these keywords. Then, the query results are combined with the user's original question to form a new, more specific question. This new question is fed into the big model for further semantic understanding and reasoning, thereby generating accurate answers for the field of robotic equipment.

[0006] This approach not only fully utilizes the advantages of NLU technology in language understanding, but also combines the structured information of knowledge graphs and the powerful reasoning ability of large models to provide an efficient and accurate question-answering system solution that can significantly improve the efficiency and accuracy of robot equipment operation and maintenance.

[0007] Through this robot equipment domain knowledge reasoning question-answering system that combines large models and NLU technology, users can quickly and accurately obtain the required operation and maintenance information through natural language interaction, greatly reducing errors and delays caused by lack of experience or incomplete information. At present, some question-answering systems based on natural language understanding have been applied to customer service, medical and other fields, but their application in the field of robot equipment is still relatively limited, especially in terms of combining large models for knowledge reasoning and question-answering, there is no mature solution. Therefore, there is an urgent need for a robot equipment domain knowledge reasoning question-answering system and method using large models and NLU technology to address the deficiencies in the existing technology. Summary of the invention

[0008] The technical problem to be solved by the present invention is how to achieve efficient processing of user questions and provide accurate and comprehensive answers, thereby improving the efficiency and accuracy of robot equipment operation and maintenance.

[0009] The present invention solves the above technical problems through the following technical means: a natural language reasoning question-answering system in the field of robotics is created using a large model, and the steps of constructing the system are as follows:

[0010] S1. Derive entity table and intent rule table from existing knowledge graph, combine regular matching and similarity algorithm, and formulate NLU module;

[0011] S2, access to the offline large model chatglm and the online Spark large model to build a variety of scenarios;

[0012] S3. Build a question rewriting module to make the questions more consistent with the questioning standards of the big model;

[0013] S4. Build a web socket communication module to achieve streaming output;

[0014] S5. Combine NLU with a large model and question rewriting module to generate a question-answering system.

[0015] Furthermore, the step S1 includes:

[0016] Export entity table: extract all relevant entities from the existing knowledge graph and use scripts to automatically export them;

[0017] Develop an intent rule table: Analyze the types of questions that users may ask, develop a corresponding intent rule table for each type of question, and clarify the identification methods of different intents. The intent rule table contains trigger conditions, keywords, and pattern matching rules.

[0018] Regular matching and similarity algorithm: Combine regular expressions to perform preliminary text matching to identify key elements in the question, use similarity algorithms to match user input with data in the knowledge base, and continuously optimize regular expressions and similarity calculation methods;

[0019] Formulate NLU module: Integrate entity table and intent rule table to build NLU module.

[0020] Furthermore, the step S2 comprises:

[0021] Access to the offline large model chatglm: Deploy the chatglm model on a local server or high-performance computing platform to provide necessary data support for the offline large model and regularly update training data and model parameters;

[0022] Access to the online Spark model: Access the online Spark model and communicate with the online model through the API interface;

[0023] Construction for multi-scenario use: Customized model application for different user scenarios. Each scenario has specific question-and-answer requirements and processing logic. The model is continuously optimized through scenario testing and user feedback.

[0024] Furthermore, the step S3 includes:

[0025] Normalization of questions: pre-process the questions input by users to make them normalized, and use text cleaning technology to automatically remove irrelevant information;

[0026] Semantic rewriting: Use NLU technology to perform semantic analysis on questions and identify core issues. Use rewriting rules to rewrite identified core issues to make them more consistent with the question standards of the large model. Combined with contextual information, rewrite questions in multiple rounds of dialogues in a related manner.

[0027] Optimize questioning methods: Analyze the response of the large model to different questioning methods, formulate optimization strategies, and continuously optimize questioning methods through experiments and feedback;

[0028] Use machine learning algorithms to automatically optimize and adjust the way you ask questions.

[0029] Further, the step S4 includes:

[0030] Building a web socket server: Building a web socket server on the server side and configuring security policies;

[0031] Implementation of client communication module: Implement the web socket communication module on the client. The client module needs to have the automatic reconnection function;

[0032] Streaming data processing: Implement streaming data processing technology to transmit the answers generated by the large model to the client in segments.

[0033] Furthermore, the step S5 comprises:

[0034] System integration: Integrate the NLU module, large model module, question rewriting module and communication module to build a complete question-answering system and conduct system testing;

[0035] Performance optimization: optimize the performance of the system;

[0036] User experience optimization: Collect user feedback, analyze user pain points and needs during use, optimize and improve in a targeted manner, and provide personalized setting options so that users can adjust system behavior according to their own needs;

[0037] Continuous improvement and iteration: Based on user feedback and system operation data, continuously improve and iterate the system;

[0038] Track the development of cutting-edge technologies and continuously introduce new technologies and methods.

[0039] The present invention also provides a question-answering method for creating a natural language reasoning question-answering system in the field of robotics using a large model using any of the above solutions, comprising the following steps:

[0040] Step 1: User access and question input, the user accesses the system through the user interface and enters questions in the question box;

[0041] Step 2: Analyze user questions and extract key information. The system uses NLU technology to analyze the questions input by users.

[0042] Step 3: Knowledge query based on graph database. After extracting the key information of the user's question, the system queries the graph database to obtain relevant knowledge.

[0043] Step 4: Knowledge integration and question generation: the system integrates the acquired relevant knowledge with the user's questions to generate new and more detailed questions;

[0044] Step 5: Answer generation and feedback. After the large model processes the input question, the system obtains the answer and performs the following processing: answer verification, answer optimization, and answer presentation.

[0045] Furthermore, the specific process of step 2 is as follows:

[0046] Text preprocessing: preprocess the text input by the user, including word segmentation, removal of stop words, part-of-speech tagging, and converting the original text into structured data;

[0047] Keyword extraction: Through word frequency statistics and TF-IDF algorithm, keywords are extracted from the text to represent the core content of the user's question;

[0048] Entity recognition: Use named entity recognition technology to identify specific entities and help the system understand the context of the question;

[0049] Intent recognition: Use deep learning models to identify the intent of user questions and determine specific needs.

[0050] Furthermore, the specific query process of step three is as follows:

[0051] Construct query statements: Construct query statements based on key information in user questions;

[0052] Execute query: send query statements to the graph database, retrieve relevant nodes and relationships, and return matching knowledge items;

[0053] Result filtering and sorting: Filter and sort query results based on the timestamp and source credibility rules of knowledge items.

[0054] Furthermore, the specific steps of step 4 are as follows:

[0055] Knowledge integration: Using NLU technology, the structured knowledge items obtained from the graph database are converted into natural language and embedded into the user's original question;

[0056] Question generation: Combine user questions and relevant knowledge to generate a new question;

[0057] Question input: The newly generated question is input into the big model, which uses deep learning algorithms to understand and reason and generate corresponding answers.

[0058] The present invention can effectively improve the accuracy and efficiency of the question-answering system in the field of robotic equipment and significantly improve the user experience. The system can not only quickly respond to user questions, but also provide detailed and accurate answers to meet the diverse needs of users. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 A flowchart of constructing a natural language reasoning question-answering system in the field of robotic devices created using a large model disclosed in an embodiment of the present invention;

[0060] Figure 2 A schematic diagram of a question-answering system for natural language reasoning in the field of robotic devices created using a large model disclosed in an embodiment of the present invention;

[0061] Figure 3A user question-answering flow chart of a natural language reasoning question-answering system in the field of robotic devices created using a large model disclosed in an embodiment of the present invention. DETAILED DESCRIPTION

[0062] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described in combination with the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0063] Example 1

[0064] Build a question-answering system based on NLU and large models

[0065] This example describes in detail the construction of a question-answering system based on natural language understanding (NLU) and a large model (such as the open source large model chatglm3.0 and the commercial Spark large model). The main difficulties and main solutions for building the system are as follows:

[0066] Difficulty 1: Accurate analysis of diverse user input

[0067] The diversity and complexity of user input questions pose challenges to the system's parsing, especially in the device field, where users may use a variety of terms and expressions, making the questions' semantics unclear or difficult to understand.

[0068] Solution:

[0069] Build a multi-level NLU module: By integrating the entity table, intent rule table, regular matching and similarity algorithm, a layered NLU module is built. The entity table extracts all device-related entities from the knowledge graph to ensure that the system can identify the key objects in the user's questions; the intent rule table formulates trigger conditions, keywords and pattern matching rules based on common question-and-answer scenarios to help the system accurately identify user intent.

[0070] Optimize similarity algorithm: To solve the problem of user input diversity, we improved the use of similarity algorithms (such as cosine similarity and Jaccard similarity) and made it more accurate to match user input and knowledge base content through repeated experiments and tuning. We also enhanced the flexibility and generalization ability of the system by analyzing and adjusting regular expressions.

[0071] Introducing a dynamic adjustment mechanism: Based on user feedback and system operation data, dynamically adjust the rules and algorithms of the NLU module to better adapt to the changes and diversity of user input and gradually improve the accuracy of parsing.

[0072] Difficulty 2: Collaboration and management of multi-model access

[0073] When multiple large models (such as the offline large model ChatGLM and the online large model Spark) are connected to the system, it is necessary to coordinate their functions and advantages in different usage scenarios to ensure the efficiency and accuracy of the system in various situations.

[0074] Solution:

[0075] Build a multi-model collaborative framework: In order to manage and coordinate the use of multiple models, a multi-model collaborative framework is designed. The functions of offline and online models are clearly distinguished in the system architecture, and the appropriate model is dynamically selected to answer different scenarios (such as technical support, product consultation), ensuring the balance of system resource utilization and response speed.

[0076] Real-time data synchronization and model optimization: Provide a regular update mechanism for offline models (such as chatglm) to keep the model up to date. For online models (such as Spark), real-time communication is carried out through the API interface to ensure the use of the latest data and parameters. The system will automatically select the optimal model for processing based on scenario requirements and model performance.

[0077] Optimize model switching logic: The system adjusts the logic and strategy of model switching based on real-time feedback and performance data, reduces the delay and computing overhead caused by model switching, and ensures the smoothness and continuity of answers.

[0078] Difficulty 3: Effectiveness and accuracy of question rewriting module

[0079] In order to make user questions meet the question standards of the big model, the question rewriting module needs to optimize the expression of the question while ensuring that the semantics remain unchanged, but different questions may have different effects on the input of the big model.

[0080] Solution:

[0081] Semantic rewriting and context management: Use NLU technology and semantic parsing algorithms to deeply analyze questions and identify the core content of questions. Combined with the rewriting rule library, questions can be rewritten while retaining their original meaning. For multiple rounds of dialogue, context information is combined to semantically associate and optimize questions to ensure coherence.

[0082] Continuous optimization and machine learning enhancement: Automatically optimize and adjust rewriting strategies through machine learning algorithms. Analyze the response effects of large models under different questioning methods, continuously train and improve question rewriting models, and gradually improve the accuracy and effectiveness of rewriting.

[0083] User feedback loop: In practical applications, a user feedback mechanism is introduced, where users evaluate and provide feedback on the generated answers. This data is used to further train and rewrite the model to improve its adaptability and accuracy.

[0084] Difficulty 4: Realizing streaming data output and real-time communication

[0085] In the process of achieving real-time interaction with users, the system needs to be able to respond quickly, transmit and display the answers generated by the large model in a timely manner. The design of streaming output needs to ensure that no information is lost during the data segment transmission process, while ensuring the order and continuity of the data.

[0086] Solution:

[0087] Build an efficient WebSocket communication module: Establish a high-performance WebSocket server on the server side to support streaming data transmission and ensure real-time communication with the client. Implement an automatic reconnection mechanism and efficient message processing logic on the client side to ensure the continuity and stability of data transmission.

[0088] Optimize streaming data processing algorithms: Design and optimize streaming data processing algorithms to transmit the answers generated by large models in segments, improve the system's response speed, and reduce user waiting time. Ensure that duplication and data loss are avoided during data transmission and maintain data integrity.

[0089] Data security and communication stability: Use encryption technology and security strategies to protect the security of data transmission. Optimize server configuration and load balancing strategies to ensure communication stability and response speed under high concurrency.

[0090] Difficulty 5: System Integration and Performance Optimization

[0091] During the system integration process, it is necessary to ensure efficient collaboration between various modules (such as NLU module, large model, question rewriting module and communication module). At the same time, it is necessary to optimize system performance to cope with high concurrent user requests.

[0092] Solution:

[0093] Modular design and testing: Modular design ensures flexible integration and independence between modules. Strict integration testing and system testing are carried out to ensure that each module functions properly and can communicate with each other, and to promptly discover and resolve potential problems and vulnerabilities.

[0094] Performance monitoring and dynamic adjustment: Use performance monitoring tools to monitor the system's operating status in real time, identify performance bottlenecks and areas with excessive resource consumption, and make dynamic adjustments and optimizations, such as increasing cache, optimizing query logic, and reducing data transmission delays.

[0095] Continuous improvement and user experience optimization: Based on user feedback and system performance data, continuous improvement and iteration are carried out. Continuously optimize system performance, improve user interface and interaction logic, and enhance system stability and user satisfaction.

[0096] The specific steps to build this solution are as follows Figure 1 The order of S1 to S5 shown in .

[0097] S1. Derive entity table and intent rule table from the existing knowledge graph, combine regular matching and similarity algorithm, and formulate NLU module, including the following steps:

[0098] Export entity table: Extract all relevant entities from the existing knowledge graph, such as product names, technical terms, functional modules, etc. The establishment of entity table can help the system identify key objects in user questions. Use scripts to automate the export process to ensure data integrity and consistency.

[0099] Formulate an intent rule table: Analyze the types of questions that users may ask, such as asking about functions, usage methods, maintenance, etc. Develop a corresponding intent rule table for each type of question and clarify the identification methods for different intents.

[0100] The intent rule table contains trigger conditions, keywords, and pattern matching rules to help the system accurately determine the user's intent.

[0101] Regular matching and similarity algorithms: Combine regular expressions to perform preliminary text matching to identify key elements in the question. Use similarity algorithms (such as cosine similarity and Jaccard similarity) to match user input with data in the knowledge base to improve matching accuracy. Improve the recognition accuracy of the system by continuously optimizing regular expressions and similarity calculation methods.

[0102] Formulate NLU module: Integrate entity table and intent rule table to build NLU module. NLU module is responsible for parsing natural language questions input by users, extracting entities and intents in the questions, and providing a basis for subsequent processing.

[0103] The NLU module adopts a layered architecture to ensure that the functions of each layer are clear and can be flexibly expanded and adjusted.

[0104] S2, access the offline large model chatglm and the online Spark large model, and build a variety of scenarios, including the following steps:

[0105] Access to the offline large model chatglm: Deploy the chatglm model on a local server or high-performance computing platform to ensure the efficient operation of the model and the security of the data. Provide necessary data support for the offline large model, and regularly update the training data and model parameters to ensure the accuracy and timeliness of the model.

[0106] Access to the online Spark Big Model: Access the online Spark Big Model and take advantage of its powerful computing power and the latest model parameters. The online model can obtain the latest data and updates in real time to provide more accurate answers. Communicate with the online model through the API interface to ensure the security and reliability of data transmission.

[0107] Construction of multi-scenario use: Customize the model application solutions for different user scenarios (such as technical support, product consultation, customer service). Each scenario has specific question-and-answer requirements and processing logic. Through scenario testing and user feedback, continuously optimize the application effect of the model and improve user satisfaction.

[0108] S3. Build a question rewriting module to make the question more consistent with the questioning standards of the large model, including the following steps:

[0109] Normalization of questions: Pre-process the questions input by users, including removing noise characters, standardizing the format, and word segmentation. Ensure the normalization of question input to lay a good foundation for subsequent processing. Use text cleaning technology to automatically remove irrelevant information and improve the efficiency of question processing.

[0110] Semantic rewriting: Use NLU technology to perform semantic analysis on questions and identify core issues. Use rewriting rules to rewrite the identified core issues to make them more consistent with the question standards of the large model. Combined with contextual information, rewrite questions in multiple rounds of dialogue to ensure the coherence and accuracy of the questions.

[0111] Optimize questioning methods: Analyze the response of the big model to different questioning methods and formulate optimization strategies. Through experiments and feedback, continuously optimize the questioning methods so that the big model can better understand and answer questions.

[0112] Use machine learning algorithms to automatically optimize and adjust the way questions are asked to improve the intelligence level of the system.

[0113] S4. Build a web socket communication module to implement streaming output, including the following steps:

[0114] Web socket server construction: Build a web socket server on the server side to ensure real-time communication with the client. Select a high-performance server configuration to ensure the stability and response speed of data transmission. Configure security policies to ensure data security during communication and prevent information leakage and illegal access.

[0115] Implementation of client communication module: Implement the web socket communication module on the client to ensure real-time connection with the server. The client module needs to have an automatic reconnection function to ensure the continuity and stability of communication. Provide a friendly user interface to display information in the question-and-answer process in real time to improve user experience.

[0116] Streaming data processing: Implement streaming data processing technology to transmit the answers generated by the large model to the client in segments. Streaming output can reduce user waiting time and improve system response speed. Optimize the processing logic of streaming output to ensure orderly transmission and display of data and avoid data loss and duplication.

[0117] S5. Combining NLU, the large model, and the question rewriting module to generate a question-answering system includes the following steps:

[0118] System integration: Integrate the NLU module, large model module, question rewriting module and communication module to build a complete question-answering system. Ensure the flexibility and scalability of the system through modular design. Conduct system testing to ensure normal communication and collaboration between modules, and discover and resolve potential problems and vulnerabilities.

[0119] Performance optimization: Optimize system performance, including improving response speed, reducing resource consumption, and improving stability. Use performance monitoring tools to monitor the system's operating status in real time and make adjustments and optimizations in a timely manner. Perform regular system maintenance and upgrades to ensure that the system is always in the best condition and provides high-quality services.

[0120] User experience optimization: Collect user feedback, analyze user pain points and needs during use, and optimize and improve in a targeted manner. Provide personalized setting options so that users can adjust the system behavior according to their needs. Through user testing and data analysis, continuously improve the system's usability and user satisfaction.

[0121] Continuous improvement and iteration: Based on user feedback and system operation data, the system is continuously improved and iterated. Each iteration requires sufficient testing to ensure the stability and reliability of the system.

[0122] Track the development of cutting-edge technologies, continuously introduce new technologies and methods, and improve the intelligence level and service quality of the system.

[0123] Through the above five steps, this embodiment shows how to build an efficient and intelligent question-answering system. The system can accurately identify user questions, generate high-quality answers that meet user needs, and provide a smooth user experience. This process not only improves the practicality of the system, but also provides convenient services for users.

[0124] In order to further improve the system performance, the present invention can also make the following improvements:

[0125] Enhanced model training: Further improve the reasoning ability and answer generation quality of large models by introducing more training data and optimizing training algorithms.

[0126] Multimodal fusion: Combine multimodal data such as images and audio to improve the system's comprehensive analysis capabilities and user experience. For example, in equipment fault diagnosis, the system can combine equipment images for analysis and answers.

[0127] User feedback mechanism: Establish a user feedback mechanism to allow users to evaluate and provide feedback on answers. The system can continuously optimize and adjust the answer generation strategy based on user feedback to enhance the system's adaptive capabilities.

[0128] Dynamic knowledge update: Through automated data crawling and updating mechanisms, the knowledge content in the graph database is kept up to date, ensuring that the system can respond to changing user needs and external environments.

[0129] Example 2

[0130] Based on Example 1, Example 2 provides a method for creating a natural language reasoning question-answering system in the field of robotics using the large model created in Example 1 to perform reasoning question-answering, including the following steps:

[0131] Step 1: User access and question input

[0132] Users access the system through the user interface and enter questions in the question box. Questions can be in natural language form, and the system captures user questions and enters the next processing stage. This process has a high degree of freedom and is suitable for a variety of application scenarios such as equipment operation, maintenance, and fault diagnosis.

[0133] Step 2: Analyze user questions and extract key information

[0134] The system uses advanced natural language understanding (NLU) technology to analyze the questions entered by users. The specific process is as follows:

[0135] Text preprocessing: Preprocess the text input by the user, including word segmentation, removal of stop words, part-of-speech tagging, etc., to convert the original text into structured data.

[0136] Keyword extraction: Keywords are extracted from the text through word frequency statistics, TF-IDF and other algorithms to represent the core content of the user's question.

[0137] Entity recognition: Use named entity recognition (NER) technology to identify specific entities such as device names and component names to help the system understand the context of the problem.

[0138] Intent recognition: Use deep learning models (such as BERT, GPT, etc.) to identify the intent of user questions and determine specific needs, such as the cause of equipment failure, operation instructions, or maintenance suggestions.

[0139] Step 3: Knowledge query based on graph database

[0140] After extracting the key information of the user's question, the system queries the graph database to obtain relevant knowledge. Graph databases (such as Neo4j) provide the following advantages:

[0141] Efficient relational query: Able to quickly store and query complex relations, supporting multi-step reasoning and cross-domain knowledge integration.

[0142] Flexible data model: knowledge content can be easily expanded and updated, supporting continuous learning and optimization of the system.

[0143] The specific query process is as follows:

[0144] Construct query statements: Construct query statements based on the key information in the user's question to ensure the accuracy of the query.

[0145] Execute query: send the query statement to the graph database, retrieve relevant nodes and relationships, and return matching knowledge items.

[0146] Result filtering and sorting: Filter and sort query results based on rules such as the timestamp of the knowledge entry and the credibility of the source to ensure that they are highly relevant to the user's question.

[0147] Step 4: Knowledge integration and problem generation

[0148] The system integrates the acquired relevant knowledge with the user's questions to generate new and more detailed questions. The specific steps are as follows:

[0149] Knowledge integration: Use natural language processing technology to convert structured knowledge items obtained from the graph database into natural language and embed them into the user's original question.

[0150] Question generation: Combine user questions and relevant knowledge to generate a new question, making it more complete and detailed, to help the large model perform accurate reasoning.

[0151] Question input: The newly generated question is input into the big model. The big model uses deep learning algorithms to understand and reason and generate corresponding answers.

[0152] Step 5: Answer generation and feedback

[0153] After the large model processes the input question, the system takes the answer and does the following:

[0154] Answer verification: Preliminary verification of the answers generated by the large model to ensure logical consistency and factual accuracy. Verification methods include comparison with knowledge base data and checking using predefined rules.

[0155] Answer optimization: The language and structure of verified answers are polished to make them more in line with natural language habits and easier for users to understand.

[0156] Answer presentation: Display optimized answers to users, allowing them to ask further questions or request clarification.

[0157] Through the above steps, the present invention can effectively improve the accuracy and efficiency of the question-answering system in the field of robotic equipment and significantly improve the user experience. The system can not only quickly respond to user questions, but also provide detailed and accurate answers to meet the diverse needs of users.

[0158] See also Figure 3 ,The following takes the SDR00 sweeping robot as an example to explain in detail the method of using a large model to create a natural language reasoning question-answering system in the field of robots for reasoning question-answering, including the following steps:

[0159] Step 1: User access and input issues

[0160] Users access the system through computers, mobile phones or other terminal devices and enter the main interface of the system. On the main interface, users can see a question box, and the system prompts users to enter their questions. Users enter questions in the question box: "What are the parts of the SDR00 sweeping robot? What are their functions?" After the system receives the user's question, it automatically enters the next step of the processing flow.

[0161] Step 2: Analyze user questions and extract key information

[0162] The system uses natural language understanding (NLU) technology to analyze the questions entered by users. Figure 2 In the interface shown, the user asks a question to the SDR00 robot vacuum cleaner. NLU technology processes and understands natural language text and breaks down the user's question into several key parts, including:

[0163] Main body: SDR00 sweeping robot

[0164] Attributes: Parts

[0165] Function: The function of the part

[0166] Based on the analysis of user questions, the system further extracts key information. This key information includes:

[0167] SDR00 sweeping robot: as the main query object

[0168] Parts: Specific content that users are interested in

[0169] Function: The specific purpose of the part that the user wants to know

[0170] The system marks these key information and generates a query request.

[0171] Step 3: Query the graph database and obtain relevant knowledge

[0172] The system submits the generated query request to the internal graph database. The graph database stores a large amount of detailed information about the SDR00 sweeping robot, including the structural diagram of its parts, the names of the parts, and their functional descriptions. The system searches the graph database and matches all the information related to the query request.

[0173] For example, the system might retrieve the following entries in a graph database:

[0174] LiDAR: measures distance and identifies objects, helping robots perceive the environment and avoid obstacles.

[0175] Camera: Captures images and is used in conjunction with LiDAR for environment recognition and navigation.

[0176] Main brush assembly: located at the bottom, responsible for cleaning ground garbage and dust.

[0177] Side brush: Located on the side, sweeps dust and debris from corners and walls.

[0178] Roller assembly: includes driving wheels and auxiliary wheels to help the robot move and turn.

[0179] Dust box: collects dust and garbage generated during cleaning.

[0180] Battery: Provides power support for the robot.

[0181] Electronic control panel: controls various operations and functions of the robot.

[0182] The system extracts all relevant knowledge points from the graph database and summarizes and organizes them. For each part, the system not only obtains its name, but also includes a detailed functional description and its role in the overall work of the robot. This information will be integrated to form a comprehensive answer to the user's question.

[0183] Step 4: Integrate the problem and input into the big model

[0184] The system integrates all the extracted part information and functional descriptions with the user's original question. This step ensures the completeness and accuracy of the answer while enabling the answer to be clearly communicated to the user. The integrated answer may look like this:

[0185] LiDAR: Used to measure distance and identify objects, helping robots perceive their surroundings and avoid obstacles. LiDAR’s high-precision measurement capabilities enable robots to respond flexibly in complex environments.

[0186] Camera: used to capture images of the surrounding environment and combined with the ranging information of the LiDAR to help the robot accurately identify and navigate the environment. The camera can also identify specific markers to further improve the robot's navigation accuracy.

[0187] Main brush assembly: The main brush assembly is the core component of the robot's cleaning system and is located at the bottom of the robot. The main brush rotates to sweep dust and garbage on the ground into the suction port. The main brush is designed to effectively handle different types of ground garbage, including hair, paper scraps and dust.

[0188] Side brush: The side brush is located on the side of the robot and is used to clean corners and walls. The side brush rotates to sweep dust and garbage from the walls and corners into the cleaning range of the main brush. The presence of the side brush ensures that the robot can clean the dead corners of the walls and furniture edges.

[0189] Roller assembly: The roller assembly includes drive wheels and auxiliary wheels, which are used to support and move the robot. The drive wheels provide forward and steering power, while the auxiliary wheels help the robot move smoothly. The design of the roller assembly ensures that the robot can run smoothly on different types of surfaces.

[0190] Dust box: The dust box is an important part of the robot's garbage collection system. During the cleaning process, the robot sucks dust and garbage into the dust box. The dust box has a large capacity and is designed to be easy to disassemble and clean. Users can easily dump garbage and clean the dust box.

[0191] Battery: The battery provides the necessary power support for the robot. High-capacity batteries ensure that the robot can operate for a long time. The battery design supports fast charging, reducing the robot's downtime caused by charging.

[0192] Electronic control board: The electronic control board is the control center of the robot, responsible for processing sensor data, executing cleaning tasks and managing battery usage. The processor and software on the electronic control board ensure that the robot can complete cleaning tasks efficiently and intelligently.

[0193] The system inputs the integrated detailed questions and relevant knowledge points into the big model. The big model has strong semantic understanding and reasoning capabilities, which can conduct in-depth analysis of complex inputs and generate high-quality answers. The big model uses its huge training data and complex algorithms to accurately explain the function of each part and give detailed instructions.

[0194] Step 5: Generate and present answers

[0195] The big model generates detailed answers based on the input information. The system formats the answers to make them clearer and easier to read, and presents them to the user through the user interface.

[0196] Through the above steps, the system can efficiently answer users' detailed questions about the parts and functions of the SDR00 sweeping robot and provide comprehensive and accurate information. The system not only improves the user experience, but also greatly reduces the time and cost of users obtaining information.

[0197] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A natural language reasoning question-answering system for creating a robot domain using a large model, characterized by: The steps to build the system are as follows: S1. Derive entity table and intent rule table from existing knowledge graph, combine regular matching and similarity algorithm, and formulate NLU module; S2, access to the offline large model chatglm and the online Spark large model to build a variety of scenarios; S3. Build a question rewriting module to make the questions more consistent with the questioning standards of the big model; S4. Build a web socket communication module to achieve streaming output; S5. Combine NLU with a large model and question rewriting module to generate a question-answering system.

2. A natural language reasoning question-answering system for creating a robot domain using a large model as claimed in claim 1, characterized in that: The step S1 comprises: Export entity table: extract all relevant entities from the existing knowledge graph and use scripts to automatically export them; Develop an intent rule table: Analyze the types of questions that users may ask, develop a corresponding intent rule table for each type of question, and clarify the identification methods of different intents. The intent rule table contains trigger conditions, keywords, and pattern matching rules. Regular matching and similarity algorithm: Combine regular expressions to perform preliminary text matching to identify key elements in the question, use similarity algorithms to match user input with data in the knowledge base, and continuously optimize regular expressions and similarity calculation methods; Formulate NLU module: Integrate entity table and intent rule table to build NLU module.

3. A natural language reasoning question-answering system for creating a robot domain using a large model as claimed in claim 1, characterized in that: The step S2 comprises: Access to the offline large model chatglm: Deploy the chatglm model on a local server or high-performance computing platform to provide necessary data support for the offline large model and regularly update training data and model parameters; Access to the online Spark model: Access the online Spark model and communicate with the online model through the API interface; Construction for multi-scenario use: Customized model application for different user scenarios. Each scenario has specific question-and-answer requirements and processing logic. The model is continuously optimized through scenario testing and user feedback.

4. A natural language reasoning question-answering system for creating a robot domain using a large model as claimed in claim 1, characterized in that: The step S3 comprises: Normalization of questions: pre-process the questions input by users to make them normalized, and use text cleaning technology to automatically remove irrelevant information; Semantic rewriting: Use NLU technology to perform semantic analysis on questions and identify core issues. Use rewriting rules to rewrite identified core issues to make them more consistent with the question standards of the large model. Combined with contextual information, rewrite questions in multiple rounds of dialogues in a related manner. Optimize questioning methods: Analyze the response of the large model to different questioning methods, formulate optimization strategies, and continuously optimize questioning methods through experiments and feedback; Use machine learning algorithms to automatically optimize and adjust the way you ask questions.

5. The method of creating a natural language reasoning question-answering system in the field of robotics using a large model as claimed in claim 1, characterized in that: The step S4 comprises: Building a web socket server: Building a web socket server on the server side and configuring security policies; Implementation of client communication module: Implement the web socket communication module on the client. The client module needs to have the automatic reconnection function; Streaming data processing: Implement streaming data processing technology to transmit the answers generated by the large model to the client in segments.

6. A natural language reasoning question-answering system for creating a robot domain using a large model as claimed in claim 1, characterized in that: The step S5 comprises: System integration: Integrate the NLU module, large model module, question rewriting module and communication module to build a complete question-answering system and conduct system testing; Performance optimization: optimize the performance of the system; User experience optimization: Collect user feedback, analyze user pain points and needs during use, optimize and improve in a targeted manner, and provide personalized setting options so that users can adjust system behavior according to their own needs; Continuous improvement and iteration: Based on user feedback and system operation data, continuously improve and iterate the system; Track the development of cutting-edge technologies and continuously introduce new technologies and methods.

7. A question-answering method for creating a natural language reasoning question-answering system in the field of robotics using a large model as claimed in any one of claims 1 to 6, characterized in that: The following steps are involved: Step 1: User access and question input, the user accesses the system through the user interface and enters questions in the question box; Step 2: Analyze user questions and extract key information. The system uses NLU technology to analyze the questions input by users. Step 3: Knowledge query based on graph database. After extracting the key information of the user's question, the system queries the graph database to obtain relevant knowledge. Step 4: Knowledge integration and question generation: the system integrates the acquired relevant knowledge with the user's questions to generate new and more detailed questions; Step 5: Answer generation and feedback. After the large model processes the input question, the system obtains the answer and performs the following processing: answer verification, answer optimization, and answer presentation.

8. A method for creating a natural language reasoning question-answering method in the field of robots using a large model as claimed in claim 7, characterized in that: The specific process of step 2 is as follows: Text preprocessing: preprocess the text input by the user, including word segmentation, removal of stop words, part-of-speech tagging, and converting the original text into structured data; Keyword extraction: Through word frequency statistics and TF-IDF algorithm, keywords are extracted from the text to represent the core content of the user's question; Entity recognition: Use named entity recognition technology to identify specific entities and help the system understand the context of the problem; Intent recognition: Use deep learning models to identify the intent of user questions and determine specific needs.

9. A method for creating a natural language reasoning question-answering method in the field of robots using a large model as claimed in claim 7, characterized in that: The specific query process of step three is as follows: Construct query statements: Construct query statements based on key information in user questions; Execute query: send query statements to the graph database, retrieve relevant nodes and relationships, and return matching knowledge items; Result filtering and sorting: Filter and sort query results based on the timestamp and source credibility rules of knowledge items.

10. A method for creating a natural language reasoning question-answering method in the field of robots using a large model as claimed in claim 7, characterized in that: The specific steps of step 4 are as follows: Knowledge integration: Using NLU technology, the structured knowledge items obtained from the graph database are converted into natural language and embedded into the user's original question; Question generation: Combine user questions and relevant knowledge to generate a new question; Question input: The newly generated question is input into the big model, which uses deep learning algorithms to understand and reason and generate corresponding answers.

Images