Agricultural voice question-answering system based on large model reasoning and intelligent auxiliary error correction

Abstract

The invention discloses an agricultural voice question answering system based on large model reasoning and intelligent auxiliary error correction, which relates to the technical field of agricultural knowledge intelligent question answering and comprises a robot terminal interaction module, a user interaction layer, a question answering processing layer, a performance evaluation optimization module and a data layer. According to the method, the GPT large model based on the Transform architecture is utilized, agricultural field adaptation is carried out, the powerful language understanding and generating capacity of the large model can be fully utilized, agricultural related problems can be better processed after fine tuning, initial answers with certain accuracy are generated, a powerful knowledge source and an inference engine are provided for the system, and the system has good application prospects. According to the method, the initial answer generated by the large model can be effectively evaluated and corrected by learning the characteristic difference of a large number of agricultural question and answer data pairs in cooperation with the established BERT error correction model, so that the accuracy and reliability of system answers are improved, and wrong answers possibly occurring in the large model are reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent question-answering of agricultural knowledge, and specifically to an agricultural voice question-answering system based on large model reasoning and intelligent auxiliary error correction. Background Art

[0002] Agricultural production involves a wide range of knowledge fields, including crop planting, pest and disease control, soil fertilizer, agricultural machinery and other aspects. Traditional ways of acquiring knowledge, such as reading books and consulting experts, often have limitations. The complexity and professionalism of agricultural knowledge make it difficult for ordinary farmers to quickly and accurately obtain the required information. With the continuous advancement of science and technology, intelligent agriculture has become an important direction for future agricultural development. In the development process of intelligent agriculture, a tool is needed to provide farmers with convenient and accurate agricultural knowledge consulting services to help them make better agricultural production decisions.

[0003] At present, traditional agricultural voice question-answering systems often rely on pre-set rules and a relatively small knowledge base that cannot cover the extensive and complex knowledge in the agricultural field. They may not be able to provide comprehensive and accurate answers due to the limitations of the knowledge base. In addition, due to the diversity and ambiguity of natural language, it is difficult for the rules to cover all possible expressions, which may cause the system to misunderstand the intention of the question and fail to correct errors in a timely manner, thereby giving wrong answers. Summary of the invention

[0004] The purpose of the present invention is to provide an agricultural voice question and answer system based on large model reasoning and intelligent assisted error correction, so as to solve the problem that the traditional agricultural voice question and answer system proposed in the above background technology may not be able to provide comprehensive and accurate answers due to the limitations of the knowledge base, and due to the diversity and ambiguity of natural language, it is difficult for the rules to cover all possible expressions, resulting in the system misunderstanding the intention of the question and being unable to correct errors in a timely manner, thereby giving incorrect answers to the questions.

[0005] To achieve the above-mentioned purpose, the present invention provides the following technical solutions: an agricultural voice question-answering system based on large model reasoning and intelligent auxiliary error correction, comprising a robot terminal interaction module, a user interaction layer, a question-answering processing layer, a performance evaluation and optimization module and a data layer;

[0006] The robot terminal interaction module is used to provide an interface for direct interaction between the agricultural voice question-and-answer system and the user, integrating voice input, voice output, visual recognition, and communication functions with the core of the question-and-answer system;

[0007] The user interaction layer is used to receive the user's question input information and send it to the question and answer processing layer, and present the system's answer to the user;

[0008] The question-answer processing layer is used to understand and process the questions raised by users, generate accurate answers, use the large model to infer the questions to generate initial answers, and establish an error correction model to judge the initial answers and correct incorrect judgments;

[0009] The performance evaluation and optimization module is used to evaluate the performance of each part of the system and take corresponding optimization measures according to the evaluation results;

[0010] The data layer is used to store and manage various agricultural data required by the system, including agricultural knowledge data, question and answer data, and model training data, providing the system with rich knowledge resources and data support.

[0011] Preferably, the robot terminal interaction module includes a voice input submodule, a voice output submodule, a visual recognition submodule and a communication submodule;

[0012] The voice input submodule is used to receive the user's voice command using the microphone array equipped by the robot terminal, and convert the received voice signal into text through the voice recognition engine;

[0013] The voice output submodule is used to play the voice answer generated by the system using the robot's speaker;

[0014] The visual recognition submodule is used to provide image recognition function using a camera equipped with a robot terminal to meet the visual recognition requirements in different agricultural scenarios;

[0015] The communication submodule is used for stable communication between the robot terminal and the question-answering system core.

[0016] Preferably, the user interaction layer includes a voice input module and a voice output module;

[0017] The speech input module is used to perform preliminary noise reduction and filtering on the collected speech and convert the pre-processed speech into text;

[0018] The speech output module is used to convert the answer text generated by the system into speech, and output the speech to the speaker for playing.

[0019] Preferably, the question-answering processing layer includes a question preprocessing module, a model reasoning output module, an intelligent auxiliary error correction module and a text output module;

[0020] The question preprocessing module is used to clean the question text and perform lexical analysis and syntactic analysis on the processed text to obtain semantic information;

[0021] The model reasoning output module is used to use the powerful language understanding and generation capabilities of the pre-trained large language model to reason about the pre-processed agriculture-related questions and generate initial answers;

[0022] The intelligent auxiliary error correction module is used to establish a BERT error correction model to evaluate and correct the initial answer generated by the large model reasoning submodule;

[0023] The text output module is used to organize and output the corrected answer information text.

[0024] Preferably, the large model in the model reasoning output module is a GPT artificial intelligence model based on the Transformer architecture, and the model is adapted to the agricultural field. The large model reasoning process is: the preprocessed question is input into the large model, and the large model generates an initial answer based on the knowledge and language patterns it has learned.

[0025] Preferably, the evaluation and error correction process of the BERT error correction model in the intelligent assisted error correction module: collect, organize and annotate the agricultural question and answer data set used for error correction model training, use the annotated agricultural question and answer data to train the error correction model so that it learns the characteristic differences between correct answers and incorrect answers, input the initial answer generated by the large model and the relevant question text into the error correction model together, evaluate the initial answer generated by the large model to determine whether it is correct, and if the initial answer is judged to be wrong, correct it to generate a final accurate answer.

[0026] Preferably, the performance evaluation and optimization module includes an evaluation index setting module, a data collection and analysis module and a performance optimization module;

[0027] The evaluation index setting module includes question-answering accuracy index, interaction performance index and model performance index;

[0028] The data collection and analysis module is used to collect user feedback information on the system through feedback buttons or channels on the user interface and store it in a dedicated database, and use data mining and analysis tools to analyze system logs and extract useful performance data for analysis;

[0029] The performance optimization module includes question-answer accuracy optimization, interaction performance optimization, and model performance optimization.

[0030] Preferably, the data layer includes an agricultural knowledge database and a data update and maintenance module;

[0031] The agricultural knowledge database uses MySQL, PostgreSQL or MongoDB database to store data, and the collected agricultural knowledge data is entered into the database;

[0032] The data update and maintenance module is used to regularly update data, set scheduled tasks, crawl the latest agricultural knowledge data on the Internet again, compare and merge it with the data in the existing database, and update expired or inaccurate information.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] 1. In the present invention, by utilizing the GPT large model based on the Transformer architecture and adapting it to the agricultural field, the powerful language understanding and generation capabilities of the large model can be fully utilized. After fine-tuning, it can better handle agricultural-related issues and generate initial answers with a certain degree of accuracy, providing the system with a powerful knowledge source and reasoning engine. The established BERT error correction model can effectively evaluate and correct the initial answers generated by the large model by learning the feature differences of a large number of agricultural question-and-answer data pairs, thereby improving the accuracy and reliability of the system's answers and reducing the possible erroneous answers of the large model.

[0035] 2. In the present invention, by optimizing the accuracy of question and answer, interactive performance and model performance according to the evaluation results, the overall performance of the system can be continuously improved, so that the system can better meet user needs and improve user satisfaction. By regularly updating data, it is ensured that the agricultural knowledge in the database always remains timely and accurate, and can promptly reflect new knowledge and new technologies in the agricultural field, so that the system's answers are more in line with the actual situation, thereby improving the practicality and reliability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 The present invention is a system block diagram of an agricultural voice question-answering system based on large model reasoning and intelligent assisted error correction.

[0037] In the figure: 1. Robot terminal interaction module; 2. User interaction layer; 21. Voice input module; 22. Voice output module; 3. Question and answer processing layer; 31. Question preprocessing module; 32. Model reasoning output module; 33. Intelligent assisted error correction module; 34. Text output module; 4. Performance evaluation and optimization module; 5. Data layer; 51. Agricultural knowledge database; 52. Data update and maintenance module. DETAILED DESCRIPTION

[0038] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the implementation regulations described are only 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.

[0039] Reference Figure 1As shown: An agricultural voice question-answering system based on large model reasoning and intelligent auxiliary error correction, including a robot terminal interaction module 1, a user interaction layer 2, a question-answering processing layer 3, a performance evaluation and optimization module 4 and a data layer 5;

[0040] The robot terminal interaction module 1 is used to provide an interface for direct interaction between the agricultural voice question-and-answer system and the user, integrating voice input, voice output, visual recognition, and communication functions with the core of the question-and-answer system. Through the equipped camera and corresponding visual recognition algorithm, it can meet the visual recognition needs in different agricultural scenarios, provide more abundant auxiliary information for the question-and-answer system, and enhance the accuracy and practicality of the system's answers; specifically, it includes a voice input submodule, a voice output submodule, a visual recognition submodule, and a communication submodule;

[0041] The voice input submodule is used to receive the user's voice commands using the microphone array equipped on the robot terminal, and convert the received voice signals into text through the voice recognition engine. Through the cooperation of the microphone array and the voice recognition engine, the user's voice commands can be better received in complex agricultural environments and accurately converted into text; the voice output submodule is used to use the robot's speaker to play the voice answers generated by the system; the visual recognition submodule is used to use the camera equipped on the robot terminal to provide image recognition function to meet the visual recognition needs in different agricultural scenarios; the communication submodule is used for stable communication between the robot terminal and the core of the question-answering system;

[0042] The user interaction layer 2 is used to receive the user's question input information and send it to the question and answer processing layer 3, and present the system's answer to the user; specifically, it includes a voice input module 21 and a voice output module 22;

[0043] The speech input module 21 is used to perform preliminary noise reduction and filtering on the collected speech, and convert the pre-processed speech into text. In an agricultural environment, it can effectively reduce the impact of background noise on speech recognition, so that the system can understand the user's questions more accurately; the speech output module 22 is used to convert the answer text generated by the system into speech, and output the speech to the speaker for playback;

[0044] The question-answer processing layer 3 is used to understand and process the questions raised by users, generate accurate answers, use the big model to infer the questions to generate initial answers, and establish an error correction model to judge the initial answers and correct the wrong judgments. By using the GPT big model based on the Transformer architecture and adapting it to the agricultural field, the powerful language understanding and generation capabilities of the big model can be fully utilized. After fine-tuning, the big model can better handle agricultural-related issues and generate initial answers with a certain degree of accuracy, providing the system with a powerful knowledge source and reasoning engine. The established BERT error correction model can effectively evaluate and correct the initial answers generated by the big model by learning the feature differences of a large number of agricultural question-answer data pairs, thereby improving the accuracy and reliability of the system's answers and reducing the possible erroneous answers of the big model. Specifically, it includes a question preprocessing module 31, a model reasoning output module 32, an intelligent auxiliary error correction module 33 and a text output module 34;

[0045] The question preprocessing module 31 is used to clean the question text and perform lexical analysis and syntactic analysis on the processed text to obtain semantic information;

[0046] The model reasoning output module 32 is used to use the powerful language understanding and generation capabilities of the pre-trained large language model to reason about the pre-processed agriculture-related questions and generate initial answers;

[0047] The big model is a GPT artificial intelligence model based on the Transformer architecture, and the model is adapted to the agricultural field. The big model reasoning process: the preprocessed questions are input into the big model, and the big model generates an initial answer based on the knowledge and language patterns it has learned;

[0048] The intelligent auxiliary error correction module 33 is used to establish a BERT error correction model to evaluate and correct the initial answer generated by the large model reasoning submodule;

[0049] The evaluation and error correction process of the BERT error correction model includes: collecting, organizing and annotating agricultural question-answering data sets used for error correction model training; using the annotated agricultural question-answering data to train the error correction model so that it can learn the characteristic differences between correct and incorrect answers; inputting the initial answer generated by the large model and the relevant question text into the error correction model; evaluating the initial answer generated by the large model to determine whether it is correct; if the initial answer is judged to be wrong, correcting it to generate the final accurate answer;

[0050] The text output module 34 is used to organize and output the corrected answer information text;

[0051] The performance evaluation and optimization module 4 is used to evaluate the performance of each part of the system and take corresponding optimization measures according to the evaluation results; specifically, it includes an evaluation index setting module, a data collection and analysis module and a performance optimization module;

[0052] The evaluation index setting module includes question-answering accuracy index, interactive performance index and model performance index, which comprehensively measures system performance, provides clear direction and standards for system optimization, and helps to discover the advantages and disadvantages of the system;

[0053] The data collection and analysis module is used to collect user feedback on the system through the feedback button or channel on the user interface and store it in a special database. It uses data mining and analysis tools to analyze system logs and extract useful performance data for analysis. By collecting user feedback information and analyzing system logs, it is possible to obtain detailed performance data of the system in actual use. These data provide a basis for subsequent optimization measures, making the optimization more targeted and effective;

[0054] The performance optimization module includes question-answering accuracy optimization, interactive performance optimization, and model performance optimization. It optimizes question-answering accuracy, interactive performance, and model performance based on the evaluation results, which can continuously improve the overall performance of the system and enable the system to better meet user needs.

[0055] The data layer 5 is used to store and manage various agricultural data required by the system, including agricultural knowledge data, question-answering data, and model training data, providing the system with rich knowledge resources and data support; specifically, it includes an agricultural knowledge database 51 and a data update and maintenance module 52;

[0056] The agricultural knowledge database 51 uses MySQL, PostgreSQL or MongoDB database to store data, and the collected agricultural knowledge data is entered into the database;

[0057] The data update and maintenance module 52 is used to regularly update data, set scheduled tasks, crawl the latest agricultural knowledge data on the Internet again, compare and integrate it with the data in the existing database, and update expired or inaccurate information. By regularly updating data, it is ensured that the agricultural knowledge in the database always remains timely and accurate, and can promptly reflect new knowledge and new technologies in the agricultural field, so that the system's answers are more in line with actual conditions, thereby improving the practicality and reliability of the system.

[0058] The usage and working principle of the present invention are as follows: the user inputs agricultural problems via voice at the robot terminal, the voice input module 21 converts it into text form, the question preprocessing module 31 performs cleaning, lexical analysis and syntactic analysis preprocessing operations on the text problem, and then inputs the preprocessed text problem into the large model, the large model generates an initial answer, the initial answer of the large model and the question text are input into the error correction model, the error correction model evaluates and corrects the initial answer to obtain a final accurate answer, and the final answer is sent to the voice output module 22, the voice output module 22 converts it into voice form and plays it to the user.

[0059] Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An agricultural voice question-answering system based on large model reasoning and intelligent assisted error correction, characterized by: It includes a robot terminal interaction module (1), a user interaction layer (2), a question-answering processing layer (3), a performance evaluation and optimization module (4) and a data layer (5); The robot terminal interaction module (1) is used to provide an interface for direct interaction between the agricultural voice question-and-answer system and the user, integrating voice input, voice output, visual recognition and communication functions with the core of the question-and-answer system; The user interaction layer (2) is used to receive the user's question input information and send it to the question and answer processing layer (3), and present the system's answer to the user; The question-answer processing layer (3) is used to understand and process the questions raised by the user, generate accurate answers, use the large model to infer the questions to generate initial answers, and establish an error correction model to judge the initial answers and correct incorrect judgments; The performance evaluation and optimization module (4) is used to evaluate the performance of each part of the system and take corresponding optimization measures according to the evaluation results; The data layer (5) is used to store and manage various agricultural data required by the system, including agricultural knowledge data, question-and-answer data, and model training data, providing the system with rich knowledge resources and data support.

2. The agricultural voice question-answering system based on large model reasoning and intelligent assisted error correction according to claim 1 is characterized in that: The robot terminal interaction module (1) comprises a voice input submodule, a voice output submodule, a visual recognition submodule and a communication submodule; The voice input submodule is used to receive the user's voice command using the microphone array equipped by the robot terminal, and convert the received voice signal into text through the voice recognition engine; The voice output submodule is used to play the voice answer generated by the system using the robot's speaker; The visual recognition submodule is used to provide image recognition function using a camera equipped with a robot terminal to meet the visual recognition requirements in different agricultural scenarios; The communication submodule is used for stable communication between the robot terminal and the question-answering system core.

3. The agricultural voice question-answering system based on large model reasoning and intelligent assisted error correction according to claim 1 is characterized in that: The user interaction layer (2) comprises a voice input module (21) and a voice output module (22); The speech input module (21) is used to perform preliminary noise reduction and filtering processing on the collected speech, and convert the pre-processed speech into text; The speech output module (22) is used to convert the answer text generated by the system into speech, and output the speech to a speaker for playing.

4. The agricultural voice question-answering system based on large model reasoning and intelligent assisted error correction according to claim 1 is characterized in that: The question-answer processing layer (3) includes a question preprocessing module (31), a model reasoning output module (32), an intelligent auxiliary error correction module (33) and a text output module (34); The question preprocessing module (31) is used to clean the question text and perform lexical analysis and syntactic analysis on the processed text to obtain semantic information; The model reasoning output module (32) is used to use the powerful language understanding and generation capabilities of the pre-trained large language model to reason about the pre-processed agricultural related questions and generate initial answers; The intelligent auxiliary error correction module (33) is used to establish a BERT error correction model to evaluate and correct the initial answer generated by the large model reasoning submodule; The text output module (34) is used to organize and output the error-corrected answer information text.

5. The agricultural voice question-answering system based on large model reasoning and intelligent assisted error correction according to claim 4 is characterized in that: The large model in the model reasoning output module (32) is a GPT artificial intelligence model based on the Transformer architecture, and the model is adapted to the agricultural field. The large model reasoning process is as follows: the preprocessed question is input into the large model, and the large model generates an initial answer based on the knowledge and language patterns it has learned.

6. The agricultural voice question-answering system based on large model reasoning and intelligent assisted error correction according to claim 4 is characterized in that: The evaluation and error correction process of the BERT error correction model in the intelligent auxiliary error correction module (33) is as follows: collecting, organizing and annotating agricultural question-and-answer data pairs used for error correction model training, using the annotated agricultural question-and-answer data to train the error correction model so that it can learn the characteristic differences between correct answers and incorrect answers, inputting the initial answer generated by the large model and the relevant question text into the error correction model, evaluating the initial answer generated by the large model to determine whether it is correct, and if the initial answer is determined to be wrong, correcting it to generate a final accurate answer.

7. The agricultural voice question-answering system based on large model reasoning and intelligent assisted error correction according to claim 1 is characterized in that: The performance evaluation and optimization module (4) includes an evaluation index setting module, a data collection and analysis module and a performance optimization module; The evaluation index setting module includes question-answering accuracy index, interaction performance index and model performance index; The data collection and analysis module is used to collect user feedback information on the system through feedback buttons or channels on the user interface and store it in a dedicated database, and use data mining and analysis tools to analyze system logs and extract useful performance data for analysis; The performance optimization module includes question-answer accuracy optimization, interaction performance optimization, and model performance optimization.

8. The agricultural voice question-answering system based on large model reasoning and intelligent auxiliary error correction according to claim 1 is characterized in that: The data layer (5) includes an agricultural knowledge database (51) and a data updating and maintenance module (52); The agricultural knowledge database (51) uses MySQL, PostgreSQL or MongoDB database to store data, and the collected agricultural knowledge data is entered into the database; The data updating and maintenance module (52) is used to regularly update data, set scheduled tasks, crawl the latest agricultural knowledge data on the Internet again, compare and merge it with the data in the existing database, and update expired or inaccurate information.

Images