Multifunctional question and answer interaction method and system based on AI intelligence

Abstract

The invention provides a multifunctional question and answer interaction method and system based on AI intelligence, and relates to the technical field of question and answer interaction. Question collection information and category feature extraction data are obtained, the category feature extraction data are combined, question feature combination data are obtained, preset feature combination filling is carried out, and feature combination filling data are obtained; calculating a filling standard coefficient, triggering a question feedback mechanism, training a question and answer model, and outputting newly added feedback data with qualified filling standards; calculating a question-answer deviation coefficient, obtaining question state judgment information, and further obtaining question-answer state comprehensive judgment information; and calculating the question-answer deviation coefficient of each feature category, performing feature state judgment, and obtaining feature state judgment information, so that comprehensive understanding of question collection information and accurate training of the question-answer model are realized, and the performance of a question-answer interaction system and the user experience are effectively improved.

Description

Technical Field

[0001] The present invention proposes a multifunctional question-answering interaction method and system based on AI intelligence, which relates to the field of question-answering interaction technology, and specifically to the field of multifunctional question-answering interaction technology based on AI intelligence. Background Art

[0002] Traditional question-answering interactive systems face many technical challenges when dealing with complex and ever-changing question collection information. On the one hand, the diversity of question information and the dispersion of key information make it difficult for traditional systems to effectively extract and combine features to accurately understand user intent. On the other hand, question-answering models are prone to bias when dealing with specific questions, resulting in inaccurate answers and difficulty adapting to the ever-changing types and numbers of questions. In addition, traditional systems lack effective feedback mechanisms and feature status determination methods, and are unable to promptly discover and optimize feature categories that cause bias. Summary of the invention

[0003] The present invention provides a multifunctional question-answering interactive method and system based on AI intelligence to solve the above problems:

[0004] The present invention proposes a multifunctional question-answering interactive method and system based on AI intelligence, the method comprising:

[0005] S1. Obtaining problem collection information, obtaining category feature extraction data according to the problem collection information, combining the category feature extraction data, obtaining problem feature combination data, filling in a preset feature combination, and obtaining feature combination filling data;

[0006] S2. Calculate the filling standard coefficient of each question collection information, obtain filling standard comparison information, trigger the question feedback mechanism, obtain question feedback data, train the question-answering model, and output the newly added feedback data that meets the filling standard;

[0007] S3, calculating the question-answering deviation coefficient of the newly added question collection information, obtaining question status determination information, and then obtaining comprehensive question-answering status determination information;

[0008] S4. Trigger the question-answering status analysis of each preset feature category, calculate the question-answering deviation coefficient of each feature category, perform feature status determination, and obtain feature status determination information.

[0009] Further, the S1 includes:

[0010] Collect input problem information to obtain problem collection information;

[0011] Obtaining a preset feature category to extract category features of each preset feature category from the problem collection information, and obtaining category feature extraction data;

[0012] Combine all category feature extraction data of each problem collected data to obtain problem feature combination data;

[0013] Obtain a preset feature combination, and fill the preset feature combination accordingly according to the problem feature combination data to obtain feature combination filling data.

[0014] Further, the S2 includes:

[0015] Calculate the filling standard coefficient of each question collection information by filling data with feature combinations;

[0016] Comparing the filling standard coefficient with a preset standard threshold to obtain filling standard comparison information;

[0017] Triggering a question feedback mechanism based on the filling standard comparison information, and obtaining question feedback data of the question collection information through a preset knowledge graph database;

[0018] Train the question-answering model by collecting information from historical questions and combining it with question feedback data;

[0019] The new question collection information is outputted through the question-answering model to obtain new feedback data.

[0020] Further, the S3 includes:

[0021] The question-answering deviation coefficient of the newly added question collection information is calculated by filling in the standard coefficient and combining the newly added feedback data;

[0022] Compare the question-answer deviation coefficient with a preset deviation threshold to obtain a question-answer deviation comparison result;

[0023] Determine the question and answer status according to the question and answer deviation comparison result to obtain question and answer status determination information;

[0024] A comprehensive determination of the question and answer status is performed through the question and answer status determination information of all newly added question collection information to obtain the question and answer status comprehensive determination information.

[0025] Further, the S4 includes:

[0026] Triggering question and answer status analysis for each preset feature category based on comprehensive judgment information of question and answer status;

[0027] The question-answer deviation coefficient of each feature category is calculated by combining the weight information of the preset feature category with the question-answer deviation coefficient to obtain the feature deviation coefficient;

[0028] Comparing the characteristic deviation coefficient with a preset characteristic deviation threshold to obtain a characteristic deviation comparison result;

[0029] According to the feature deviation comparison result, feature state determination is performed to obtain feature state determination information.

[0030] Furthermore, the system comprises:

[0031] A feature filling module is used to obtain problem collection information, obtain category feature extraction data according to the problem collection information, combine the category feature extraction data, obtain problem feature combination data, fill in a preset feature combination, and obtain feature combination filling data;

[0032] The standard analysis output module is used to calculate the filling standard coefficient of each question collection information, obtain the filling standard comparison information, trigger the question feedback mechanism, obtain the question feedback data, train the question-answering model, and output the newly added feedback data that meets the filling standard;

[0033] The question-answering status analysis module is used to calculate the question-answering deviation coefficient of the newly added question collection information, obtain the question status determination information, and then obtain the comprehensive determination information of the question-answering status;

[0034] The feature deviation analysis module is used to trigger the question-answering status analysis of each preset feature category, calculate the question-answering deviation coefficient of each feature category, perform feature status determination, and obtain feature status determination information.

[0035] Furthermore, the feature filling module includes:

[0036] A data collection module is used to collect input problem information and obtain problem collection information;

[0037] A feature extraction module is used to obtain a preset feature category and extract category features of each preset feature category from the problem collection information to obtain category feature extraction data;

[0038] A feature combination module is used to combine all category feature extraction data of each problem collection data to obtain problem feature combination data;

[0039] The combination filling module is used to obtain a preset feature combination, and fill the preset feature combination accordingly according to the problem feature combination data to obtain feature combination filling data.

[0040] Furthermore, the standard analysis output module includes:

[0041] A filling calculation module is used to calculate the filling standard coefficient of each problem collection information by filling data with feature combinations;

[0042] A filling analysis module, used for comparing the filling standard coefficient with a preset standard threshold value to obtain filling standard comparison information;

[0043] A model training output module is used to trigger a question feedback mechanism according to the filling standard comparison information, and obtain question feedback data of the question collection information through a preset knowledge graph database;

[0044] Train the question-answering model by collecting information from historical questions and combining it with question feedback data;

[0045] The new question collection information is outputted through the question-answering model to obtain new feedback data.

[0046] Furthermore, the question-answering status analysis module includes:

[0047] A question-answering deviation calculation module, used to calculate the question-answering deviation coefficient of the newly added question collection information by filling in the standard coefficient and combining the newly added feedback data;

[0048] A question-answering state determination module is used to compare the question-answering deviation coefficient with a preset deviation threshold to obtain a question-answering deviation comparison result;

[0049] Determine the question and answer status according to the question and answer deviation comparison result to obtain question and answer status determination information;

[0050] The question status comprehensive determination module is used to perform a comprehensive determination on the question and answer status through the question and answer status determination information of all newly added question collection information to obtain the question and answer status comprehensive determination information.

[0051] Furthermore, the feature deviation analysis module includes:

[0052] A feature deviation calculation module is used to trigger the question-answering status analysis of each preset feature category according to the comprehensive judgment information of the question-answering status;

[0053] The question-answer deviation coefficient of each feature category is calculated by combining the weight information of the preset feature category with the question-answer deviation coefficient to obtain the feature deviation coefficient;

[0054] A feature state analysis module, used to compare the feature deviation coefficient with a preset feature deviation threshold to obtain a feature deviation comparison result;

[0055] According to the feature deviation comparison result, feature state determination is performed to obtain feature state determination information.

[0056] Beneficial effects of the present invention: Through feature extraction and combination, the system can understand the problem more comprehensively, thereby improving the accuracy of questions and answers. The question-answering model training and feedback mechanism enables the system to continuously learn and optimize to adapt to different question types and scenarios. The system can automatically process questions input by users, and intelligently classify and process them according to the characteristics of the questions. By calculating the filling standard coefficient and the question-answering deviation coefficient, the system can promptly discover and correct deviations in the data filling and question-answering process, thereby improving the adaptability of the system. Users can obtain timely and accurate answers through the system, improving user experience. The question feedback mechanism enables users to participate in the optimization process of the system, further improving the usability and satisfaction of the system. The system can perform a detailed analysis of the question-answering status of different feature categories, and can discover potential problems and improvement directions. This multi-feature category analysis capability enables the system to more comprehensively or specifically evaluate the impact of problem features. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 Schematic diagram of a multifunctional question-answering interactive method based on AI intelligence. DETAILED DESCRIPTION

[0058] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0059] In one embodiment of the present invention, a multifunctional question-answering interactive method and system based on AI intelligence is proposed by the present invention, and the method comprises:

[0060] S1. Obtaining problem collection information, obtaining category feature extraction data according to the problem collection information, combining the category feature extraction data, obtaining problem feature combination data, filling in a preset feature combination, and obtaining feature combination filling data;

[0061] S2. Calculate the filling standard coefficient of each question collection information, obtain filling standard comparison information, trigger the question feedback mechanism, obtain question feedback data, train the question-answering model, and output the newly added feedback data that meets the filling standard;

[0062] S3, calculating the question-answering deviation coefficient of the newly added question collection information, obtaining question status determination information, and then obtaining comprehensive question-answering status determination information;

[0063] S4, trigger the question-answering status analysis of each preset feature category, calculate the question-answering deviation coefficient of each feature category, perform feature status determination, and obtain feature status determination information, such as Figure 1 shown.

[0064] The working principle of the above technical solution is: obtain question collection information through user input or automatic collection. Based on this information, the system extracts category feature data, which may include keywords, semantic categories, context information, etc. of different feature types of the question. Combining these category feature data to form question feature combination data can more comprehensively understand the essence of the problem. Fill the feature combination data with preset features to ensure data integrity and obtain feature combination filling data. Calculate the filling standard coefficient of each question collection information, which reflects the integrity and accuracy of data filling. By comparing the filling standard coefficient, the system obtains filling standard comparison information. If the data filling is standard, the question feedback mechanism is triggered. The user or the system trains the question and answer model based on the question feedback data provided by the feedback mechanism to improve the accuracy and adaptability of the model. The trained model can output new feedback data that meets the filling standard. Calculate the question question and answer deviation coefficient of the newly added question collection information, which is used to evaluate the accuracy of the model answering questions. According to the question and answer deviation coefficient, the system obtains the question state determination information, and then comprehensively evaluates the question and answer state to obtain the question and answer state comprehensive determination information.

[0065] This comprehensive judgment information can help the system understand the accuracy of the current question-answering model, and also reflect whether the input question is standard or accurate. When the comprehensive judgment information is unqualified, the question-answering status analysis of each preset feature category is triggered, and each feature category is evaluated in detail. The question-answering deviation coefficient of each feature category is calculated, and the feature status judgment is performed based on these coefficients. The feature status judgment information is obtained, which reveals the performance impact of different feature categories in the question-answering process.

[0066] The technical effect of the above technical solution is: through feature extraction and combination, the system can understand the problem more comprehensively, thereby improving the accuracy of questions and answers. The question-answering model training and feedback mechanism enables the system to continuously learn and optimize to adapt to different question types and scenarios. The system can automatically process questions input by users and intelligently classify and process them according to the characteristics of the questions. By calculating the filling standard coefficient and the question-answering deviation coefficient, the system can promptly discover and correct deviations in the data filling and question-answering process, thereby improving the adaptability of the system. Users can obtain timely and accurate answers through the system to improve user experience. The question feedback mechanism enables users to participate in the optimization process of the system, further improving the usability and satisfaction of the system. The system can perform a detailed analysis of the question-answering status of different feature categories, and can discover potential problems and improvement directions. This multi-feature category analysis capability enables the system to more comprehensively or specifically evaluate the impact of problem features.

[0067] In one embodiment of the present invention, the S1 includes:

[0068] Collect input problem information to obtain problem collection information;

[0069] Obtaining preset feature categories to extract category features of each preset feature category from the problem collection information to obtain category feature extraction data; the preset feature categories include information usage context field categories and information sentiment categories, etc.;

[0070] Combine all category feature extraction data of each problem collected data to obtain problem feature combination data;

[0071] Obtain a preset feature combination, and fill the preset feature combination accordingly according to the problem feature combination data to obtain feature combination filling data.

[0072] The working principle of the above technical solution is: receiving the question information input by the user, this process is called question information collection. The user can submit questions through text input, voice input or other interactive methods. The collected question information is feature extracted according to the preset feature categories. These preset feature categories include but are not limited to information usage context field categories (such as technology, education, entertainment, etc.) and information sentiment categories (such as positive, negative, neutral, etc.). For each question collection data, the system will perform detailed feature analysis according to the preset feature categories and extract the corresponding category feature data. The system combines all category feature extraction data of each question collection data to form a comprehensive question feature combination data. This process can help the system understand the nature and context information of the problem more deeply. A set of feature combination templates are pre-defined, which contain various feature categories that may need to be considered when dealing with problems. According to the problem feature combination data, the system fills the preset feature combination accordingly to ensure that each feature category has corresponding data support. The filled data is called feature combination filling data.

[0073] The technical effect of the above technical solution is: through category feature extraction and question feature combination, the system can more deeply understand the nature of the problem and contextual information, thereby providing more accurate answers. The design of preset feature categories and feature combination templates enables the system to handle multiple types of questions, including questions of different contextual fields and emotional tendencies. This design enhances the adaptability and flexibility of the system. The process of feature combination filling enables the system to quickly organize and analyze question data, thereby improving processing efficiency. By integrating information from multiple feature categories, the system is able to handle more complex questions. For example, when answering questions involving multiple fields or with complex emotional tendencies, the system is able to consider various factors more comprehensively, thereby providing more accurate answers. The process of feature extraction and combination is configurable, which means that the system can be adjusted and optimized according to actual needs. As technology continues to develop, the system can easily add new feature categories or adjust existing feature combination templates to adapt to changing question types and user needs.

[0074] In one embodiment of the present invention, S2 includes:

[0075] Calculate the filling standard coefficient of each question collection information by filling data with feature combinations;

[0076] The calculation formula of the filling standard coefficient is:

[0077]

[0078] Wherein, TC is the filling standard coefficient, t is the total number of preset features of the preset feature combination, a is the vector of filling data, b is the vector of preset data, a*b is the inner product of vector a and vector b, ||a|| and ||b|| are the modulus (length) of vector a and vector b respectively, is the similarity between the filled data and the preset data of the i-th preset feature, T sr is the total amount of filling data, T zr The total amount of data to be filled;

[0079] Comparing the filling standard coefficient with a preset standard threshold to obtain filling standard comparison information;

[0080] The problem feedback mechanism is triggered according to the filling standard comparison information, and the problem feedback data of the problem collection information is obtained through the preset knowledge graph database; when the filling standard coefficient is greater than the preset standard threshold, the problem feedback mechanism is triggered;

[0081] Train the question-answering model by collecting information from historical questions and combining it with question feedback data;

[0082] The new question collection information is outputted through the question-answering model to obtain new feedback data.

[0083] The working principle of the above technical solution is: the filling standard coefficient of each question collection information is calculated according to the feature combination filling data. This coefficient reflects the completeness and accuracy of the question collection information in the feature combination filling process. The system compares the calculated filling standard coefficient with the preset standard threshold. This standard threshold is preset according to the system requirements and data quality requirements. The comparison result generates filling standard comparison information, which is used to determine whether the question collection information meets the system requirements. When the filling standard coefficient is greater than the preset standard threshold, the system triggers the question feedback mechanism. Through the preset knowledge graph database, the system searches for feedback data related to the question collection information. The knowledge graph database contains a large number of question-answer pairs and related domain knowledge and rules, which can enable the system to provide more accurate feedback. The question-answering model is trained using historical question collection information and corresponding question feedback data. This process includes steps such as data preprocessing, feature selection, model construction and training. The trained question-answering model can understand the questions more accurately and give corresponding answers. For the newly added question collection information, the system outputs the question feedback data through the trained question-answering model. This process includes steps such as question understanding, answer generation and output. The output newly added feedback data will be used for subsequent question processing or user interaction.

[0084] The technical effect of the above technical solution is: by calculating the filling standard coefficient and comparing it with the preset standard threshold, the system can promptly discover and process the problem collection information whose data quality does not meet the requirements, thereby improving the integrity and accuracy of the data. By training the question-answering model with historical question collection information and question feedback data, the system can continuously optimize the performance of the model and improve the accuracy and efficiency of question-answering. With the support of the knowledge graph database, the system can handle more complex and diverse problems and enhance the adaptability and flexibility of the system. The system can provide more accurate and timely feedback data to improve the user experience and satisfaction. The system can continuously receive new question collection information and feedback data for model updating and optimization, so as to achieve continuous learning and progress of the system. This capability is essential to maintaining the competitiveness and adaptability of the system.

[0085] In one embodiment of the present invention, S3 includes:

[0086] The question-answering deviation coefficient of the newly added question collection information is calculated by filling in the standard coefficient and combining the newly added feedback data;

[0087] The calculation formula of the question-answering deviation coefficient is:

[0088]

[0089] Among them, WP is the question-answering bias coefficient, W dis the preset weight data for the newly added question collection data, xf is the vector of the newly added feedback data, yf is the vector of the preset feedback data, The similarity between the newly added feedback data and the preset feedback data of the newly added question collection data, TC d Filling standard coefficients for collecting data for newly added questions;

[0090] Compare the question-answer deviation coefficient with a preset deviation threshold to obtain a question-answer deviation comparison result;

[0091] Determine the question and answer status according to the question and answer deviation comparison result to obtain question and answer status determination information;

[0092] A comprehensive determination of the question and answer status is performed through the question and answer status determination information of all newly added question collection information to obtain the question and answer status comprehensive determination information.

[0093] Compare the total number of newly added question collection information whose question and answer deviation coefficient is greater than the preset deviation threshold with the preset number threshold to obtain a comprehensive comparison result of the question and answer status, and make a comprehensive judgment on the question and answer status based on the comprehensive comparison result of the question and answer status to obtain comprehensive judgment information of the question and answer status.

[0094] The working principle of the above technical solution is: first, combine the filling standard coefficient and the newly added feedback data (the newly added feedback data here refers to the newly added feedback data output by the question and answer model) to calculate the question and answer deviation coefficient of the newly added question collection information. The product reflects the comprehensive performance of the information in terms of filling standards and feedback deviation. In the formula, relatively speaking, the greater the similarity, the smaller the deviation coefficient. The system compares the calculated question-answer deviation coefficient with the preset deviation threshold. This deviation threshold is preset according to the system's requirements for question-answer accuracy. The comparison results generate a question-answer deviation comparison result, which is used to determine whether the question-answer model meets the accuracy requirements of the system when processing specific questions. Based on the question-answer deviation comparison result, the system determines the question-answer status of each newly added question collection information. This determination process involves classifying the degree of deviation (such as slight deviation, severe deviation, etc.) and generating corresponding question-answer status determination information. The system makes a comprehensive determination by summarizing the question-answer status determination information of all newly added question collection information. This process may involve statistical analysis, trend prediction, etc. of the determination information to generate comprehensive question-answer status determination information. The total number of newly added question collection information whose question-answer deviation coefficient is greater than the preset deviation threshold is also compared with the preset number threshold. This comparison result reflects the overall deviation level of the question-answer model when processing questions. Based on the comparison results, the system makes a comprehensive judgment on the question and answer status and generates final comprehensive judgment information on the question and answer status.

[0095] The technical effect of the above technical solution is: by calculating the question-answer deviation coefficient and comparing and judging, the system can promptly discover and deal with the deviations generated by the question-answer model when processing specific questions, thereby improving the accuracy of the question-answer. The system can determine whether to conduct an in-depth analysis of the question feature category deviation based on the comprehensive judgment information of the question-answer status. By reducing the question-answer deviation, the system can provide more accurate and reliable answers, thereby enhancing user satisfaction and trust. The system can continuously monitor the performance of the question-answer model and improve and optimize it based on the monitoring results. This capability is critical to maintaining the competitiveness and adaptability of the system. The comprehensive judgment information of the question-answer status can provide valuable reference information for system administrators or decision makers to help them make decisions, such as adjusting system parameters, optimizing model structure, etc.

[0096] In one embodiment of the present invention, the S4 includes:

[0097] Triggering question and answer status analysis for each preset feature category based on comprehensive judgment information of question and answer status;

[0098] The question-answer deviation coefficient of each feature category is calculated by combining the weight information of the preset feature category with the question-answer deviation coefficient, so as to obtain the feature deviation coefficient;

[0099] The calculation formula of the characteristic deviation coefficient is:

[0100]

[0101] Wherein, TP is the feature deviation coefficient, e is the number of features of the preset feature category, ta is the vector of actual data of the preset feature category, tb is the vector of preset data of the preset feature category, is the similarity between the actual data and the preset data of the oth feature of the preset feature category, WP o is the total value of the question-answer deviation coefficient of all features in the preset feature category, W o is the preset weight value of the oth feature;

[0102] Comparing the characteristic deviation coefficient with a preset characteristic deviation threshold to obtain a characteristic deviation comparison result;

[0103] According to the feature deviation comparison result, feature state determination is performed to obtain feature state determination information.

[0104] The working principle of the above technical solution is: trigger the question-answering status analysis of each preset feature category based on the comprehensive judgment information of the question-answering status. This means that when the system detects that there is a problem with the question-answering status, a detailed analysis is performed on each feature category to determine which features may have caused these problems. For each preset feature category, the system combines the weight information of the feature category and the question-answering deviation coefficient to calculate the feature deviation coefficient. The comprehensive similarity between all features of each feature category and the preset data can be calculated by (WP o *W o ) can calculate the weight ratio information of the feature category. In the formula, relatively speaking, the greater the similarity, the smaller the deviation coefficient. ; This coefficient reflects the degree of deviation generated by the question-answering model under a specific feature category. Compare the calculated feature deviation coefficient with the preset feature deviation threshold. This feature deviation threshold is preset according to the system's requirements for feature accuracy. The comparison result generates a feature deviation comparison result, which is used to determine whether a specific feature category meets the system's accuracy requirements during the question-answering process. Based on the feature deviation comparison result, the system makes a status judgment for each feature category. This judgment process may involve the classification of the degree of deviation (such as slight deviation, severe deviation, etc.) and generate corresponding feature status judgment information. This information is used to trigger subsequent operations such as feature optimization, model adjustment, or data preprocessing.

[0105] The technical effect of the above technical solution is: by conducting a detailed analysis of the question-answering status of each feature category, the system can locate the problem more accurately, thereby providing a more targeted solution. According to the feature deviation coefficient and feature state determination information, the system can adjust the weight of the feature category to improve the accuracy and efficiency of the question-answering model. By identifying and optimizing the feature categories that cause deviations, the system can improve the overall performance of the question-answering model and reduce deviations and errors. The system can dynamically adjust and optimize the model based on the feature state determination information to adapt to the changing types of questions and user needs. The feature state determination information can provide valuable reference information for system administrators or decision makers, such as adjusting feature weights, optimizing model structure, or improving data preprocessing processes. By continuously monitoring and analyzing the feature deviation coefficient and feature state determination information, the system can continuously learn and optimize its own performance to maintain competitiveness and adaptability.

[0106] In one embodiment of the present invention, the system comprises:

[0107] A feature filling module is used to obtain problem collection information, obtain category feature extraction data according to the problem collection information, combine the category feature extraction data, obtain problem feature combination data, fill in a preset feature combination, and obtain feature combination filling data;

[0108] The standard analysis output module is used to calculate the filling standard coefficient of each question collection information, obtain the filling standard comparison information, trigger the question feedback mechanism, obtain the question feedback data, train the question-answering model, and output the newly added feedback data that meets the filling standard;

[0109] The question-answering status analysis module is used to calculate the question-answering deviation coefficient of the newly added question collection information, obtain the question status determination information, and then obtain the comprehensive determination information of the question-answering status;

[0110] The feature deviation analysis module is used to trigger the question-answering status analysis of each preset feature category, calculate the question-answering deviation coefficient of each feature category, perform feature status determination, and obtain feature status determination information.

[0111] The working principle of the above technical solution is: obtain question collection information through user input or automatic collection. Based on this information, the system extracts category feature data, which may include keywords, semantic categories, context information, etc. of different feature types of the question. Combining these category feature data to form question feature combination data can more comprehensively understand the essence of the problem. Fill the feature combination data with preset features to ensure data integrity and obtain feature combination filling data. Calculate the filling standard coefficient of each question collection information, which reflects the integrity and accuracy of data filling. By comparing the filling standard coefficient, the system obtains filling standard comparison information. If the data filling is standard, the question feedback mechanism is triggered. The user or the system trains the question and answer model based on the question feedback data provided by the feedback mechanism to improve the accuracy and adaptability of the model. The trained model can output new feedback data that meets the filling standard. Calculate the question question and answer deviation coefficient of the newly added question collection information, which is used to evaluate the accuracy of the model answering questions. According to the question and answer deviation coefficient, the system obtains the question state determination information, and then comprehensively evaluates the question and answer state to obtain the question and answer state comprehensive determination information.

[0112] This comprehensive judgment information can help the system understand the accuracy of the current question-answering model, and also reflect whether the input question is standard or accurate. When the comprehensive judgment information is unqualified, the question-answering status analysis of each preset feature category is triggered, and each feature category is evaluated in detail. The question-answering deviation coefficient of each feature category is calculated, and the feature status judgment is performed based on these coefficients. The feature status judgment information is obtained, which reveals the performance impact of different feature categories in the question-answering process.

[0113] The technical effect of the above technical solution is: through feature extraction and combination, the system can understand the problem more comprehensively, thereby improving the accuracy of questions and answers. The question-answering model training and feedback mechanism enables the system to continuously learn and optimize to adapt to different question types and scenarios. The system can automatically process questions input by users and intelligently classify and process them according to the characteristics of the questions. By calculating the filling standard coefficient and the question-answering deviation coefficient, the system can promptly discover and correct deviations in the data filling and question-answering process, thereby improving the adaptability of the system. Users can obtain timely and accurate answers through the system to improve user experience. The question feedback mechanism enables users to participate in the optimization process of the system, further improving the usability and satisfaction of the system. The system can perform a detailed analysis of the question-answering status of different feature categories, and can discover potential problems and improvement directions. This multi-feature category analysis capability enables the system to more comprehensively or specifically evaluate the impact of problem features.

[0114] In one embodiment of the present invention, the feature filling module includes:

[0115] A data collection module is used to collect input problem information and obtain problem collection information;

[0116] A feature extraction module is used to obtain preset feature categories and extract category features of each preset feature category from the problem collection information to obtain category feature extraction data; the preset feature categories include information usage context field categories and information sentiment categories, etc.;

[0117] A feature combination module is used to combine all category feature extraction data of each problem collection data to obtain problem feature combination data;

[0118] The combination filling module is used to obtain a preset feature combination, and fill the preset feature combination accordingly according to the problem feature combination data to obtain feature combination filling data.

[0119] The working principle of the above technical solution is: receiving the question information input by the user, this process is called question information collection. The user can submit questions through text input, voice input or other interactive methods. The collected question information is feature extracted according to the preset feature categories. These preset feature categories include but are not limited to information usage context field categories (such as technology, education, entertainment, etc.) and information sentiment categories (such as positive, negative, neutral, etc.). For each question collection data, the system will perform detailed feature analysis according to the preset feature categories and extract the corresponding category feature data. The system combines all category feature extraction data of each question collection data to form a comprehensive question feature combination data. This process can help the system understand the nature and context information of the problem more deeply. A set of feature combination templates are pre-defined, which contain various feature categories that may need to be considered when dealing with problems. According to the problem feature combination data, the system fills the preset feature combination accordingly to ensure that each feature category has corresponding data support. The filled data is called feature combination filling data.

[0120] The technical effect of the above technical solution is: through category feature extraction and question feature combination, the system can more deeply understand the nature of the problem and contextual information, thereby providing more accurate answers. The design of preset feature categories and feature combination templates enables the system to handle multiple types of questions, including questions of different contextual fields and emotional tendencies. This design enhances the adaptability and flexibility of the system. The process of feature combination filling enables the system to quickly organize and analyze question data, thereby improving processing efficiency. By integrating information from multiple feature categories, the system is able to handle more complex questions. For example, when answering questions involving multiple fields or with complex emotional tendencies, the system is able to consider various factors more comprehensively, thereby providing more accurate answers. The process of feature extraction and combination is configurable, which means that the system can be adjusted and optimized according to actual needs. As technology continues to develop, the system can easily add new feature categories or adjust existing feature combination templates to adapt to changing question types and user needs.

[0121] In one embodiment of the present invention, the standard analysis output module includes:

[0122] A filling calculation module is used to calculate the filling standard coefficient of each problem collection information by filling data with feature combinations;

[0123] The calculation formula of the filling standard coefficient is:

[0124]

[0125] Wherein, TC is the filling standard coefficient, t is the total number of preset features of the preset feature combination, a is the vector of filling data, b is the vector of preset data, a*b is the inner product of vector a and vector b, ||a|| and ||b|| are the modulus (length) of vector a and vector b respectively, is the similarity between the filled data and the preset data of the i-th preset feature, T sr is the total amount of filling data, T zr The total amount of data to be filled;

[0126] A filling analysis module, used for comparing the filling standard coefficient with a preset standard threshold value to obtain filling standard comparison information;

[0127] A model training output module is used to trigger a problem feedback mechanism according to the filling standard comparison information, and obtain problem feedback data of the problem collection information through a preset knowledge graph database; when the filling standard coefficient is greater than a preset standard threshold, the problem feedback mechanism is triggered;

[0128] The question-answering model is trained by combining historical question collection information with question feedback data; the question-answering model training method is a commonly used training method in the prior art, and the training data used in the present invention is not conventional training data, but training data processed by filling standards;

[0129] The new question collection information is outputted through the question-answering model to obtain new feedback data.

[0130] The working principle of the above technical solution is: the filling standard coefficient of each question collection information is calculated according to the feature combination filling data. This coefficient reflects the completeness and accuracy of the question collection information in the feature combination filling process. The system compares the calculated filling standard coefficient with the preset standard threshold. This standard threshold is preset according to the system requirements and data quality requirements. The comparison result generates filling standard comparison information, which is used to determine whether the question collection information meets the system requirements. When the filling standard coefficient is greater than the preset standard threshold, the system triggers the question feedback mechanism. Through the preset knowledge graph database, the system searches for feedback data related to the question collection information. The knowledge graph database contains a large number of question-answer pairs and related domain knowledge and rules, which can enable the system to provide more accurate feedback. The question-answering model is trained using historical question collection information and corresponding question feedback data. This process includes steps such as data preprocessing, feature selection, model construction and training. The trained question-answering model can understand the questions more accurately and give corresponding answers. For the newly added question collection information, the system outputs the question feedback data through the trained question-answering model. This process includes steps such as question understanding, answer generation and output. The output newly added feedback data will be used for subsequent question processing or user interaction.

[0131] The technical effect of the above technical solution is: by calculating the filling standard coefficient and comparing it with the preset standard threshold, the system can promptly discover and process the problem collection information whose data quality does not meet the requirements, thereby improving the integrity and accuracy of the data. By training the question-answering model with historical question collection information and question feedback data, the system can continuously optimize the performance of the model and improve the accuracy and efficiency of question-answering. With the support of the knowledge graph database, the system can handle more complex and diverse problems and enhance the adaptability and flexibility of the system. The system can provide more accurate and timely feedback data to improve the user experience and satisfaction. The system can continuously receive new question collection information and feedback data for model updating and optimization, so as to achieve continuous learning and progress of the system. This capability is essential to maintaining the competitiveness and adaptability of the system.

[0132] In one embodiment of the present invention, the question-answering status analysis module includes:

[0133] A question-answering deviation calculation module, used to calculate the question-answering deviation coefficient of the newly added question collection information by filling in the standard coefficient and combining the newly added feedback data;

[0134] The calculation formula of the question-answering deviation coefficient is:

[0135]

[0136] Among them, WP is the question-answering bias coefficient, W d is the preset weight data for the newly added question collection data, xf is the vector of the newly added feedback data, yf is the vector of the preset feedback data, The similarity between the newly added feedback data and the preset feedback data of the newly added question collection data, TC d Filling standard coefficients for collecting data for newly added questions;

[0137] A question-answering state determination module is used to compare the question-answering deviation coefficient with a preset deviation threshold to obtain a question-answering deviation comparison result;

[0138] Determine the question and answer status according to the question and answer deviation comparison result to obtain question and answer status determination information;

[0139] The question status comprehensive determination module is used to perform a comprehensive determination on the question and answer status through the question and answer status determination information of all newly added question collection information to obtain the question and answer status comprehensive determination information.

[0140] Compare the total number of newly added question collection information whose question and answer deviation coefficient is greater than the preset deviation threshold with the preset number threshold to obtain a comprehensive comparison result of the question and answer status, and make a comprehensive judgment on the question and answer status based on the comprehensive comparison result of the question and answer status to obtain comprehensive judgment information of the question and answer status.

[0141] The working principle of the above technical solution is: first, the question and answer deviation coefficient of the newly added question collection information is calculated by combining the filling standard coefficient and the newly added feedback data (the newly added feedback data here refers to the newly added feedback data output by the question and answer model). This coefficient reflects the degree of deviation generated by the question and answer model when processing a specific question. The system compares the calculated question and answer deviation coefficient with the preset deviation threshold. This deviation threshold is preset according to the system's requirements for question and answer accuracy. The comparison result generates a question and answer deviation comparison result, which is used to determine whether the question and answer model meets the accuracy requirements of the system when processing a specific question. According to the question and answer deviation comparison result, the system determines the question and answer status of each newly added question collection information. This determination process involves the classification of the degree of deviation (such as slight deviation, severe deviation, etc.), and generates corresponding question and answer status determination information. The system makes a comprehensive determination by summarizing the question and answer status determination information of all newly added question collection information. This process may involve statistical analysis, trend prediction, etc. of the determination information to generate comprehensive determination information of the question and answer status. The total number of newly added question collection information whose question and answer deviation coefficient is greater than the preset deviation threshold is also compared with the preset number threshold. This comparison result reflects the overall deviation level of the question-answering model when processing questions. Based on the comparison result, the system makes a comprehensive judgment on the question-answering status and generates the final comprehensive judgment information on the question-answering status.

[0142] The technical effect of the above technical solution is: by calculating the question-answer deviation coefficient and comparing and judging, the system can promptly discover and deal with the deviations generated by the question-answer model when processing specific questions, thereby improving the accuracy of the question-answer. The system can determine whether to conduct an in-depth analysis of the question feature category deviation based on the comprehensive judgment information of the question-answer status. By reducing the question-answer deviation, the system can provide more accurate and reliable answers, thereby enhancing user satisfaction and trust. The system can continuously monitor the performance of the question-answer model and improve and optimize it based on the monitoring results. This capability is critical to maintaining the competitiveness and adaptability of the system. The comprehensive judgment information of the question-answer status can provide valuable reference information for system administrators or decision makers to help them make decisions, such as adjusting system parameters, optimizing model structure, etc.

[0143] In one embodiment of the present invention, the feature deviation analysis module includes:

[0144] A feature deviation calculation module is used to trigger the question-answering status analysis of each preset feature category according to the comprehensive judgment information of the question-answering status;

[0145] The question-answer deviation coefficient of each feature category is calculated by combining the weight information of the preset feature category with the question-answer deviation coefficient to obtain the feature deviation coefficient;

[0146] The calculation formula of the characteristic deviation coefficient is:

[0147]

[0148] Wherein, TP is the feature deviation coefficient, e is the number of features of the preset feature category, ta is the vector of actual data of the preset feature category, tb is the vector of preset data of the preset feature category, is the similarity between the actual data and the preset data of the oth feature of the preset feature category, WP o is the total value of the question-answer deviation coefficient of all features in the preset feature category, W o is the preset weight value of the oth feature;

[0149] A feature state analysis module, used to compare the feature deviation coefficient with a preset feature deviation threshold to obtain a feature deviation comparison result;

[0150] According to the feature deviation comparison result, feature state determination is performed to obtain feature state determination information.

[0151] The working principle of the above technical solution is: trigger the question and answer status analysis of each preset feature category according to the comprehensive judgment information of the question and answer status. This means that when the system detects that there is a problem with the question and answer status, each feature category is analyzed in detail to determine which features may have caused these problems. For each preset feature category, the system calculates the feature deviation coefficient in combination with the weight information of the feature category and the question and answer deviation coefficient. This coefficient reflects the degree of deviation generated by the question and answer model under a specific feature category. The calculated feature deviation coefficient is compared with the preset feature deviation threshold. This feature deviation threshold is preset according to the system's requirements for feature accuracy. The comparison result generates a feature deviation comparison result, which is used to determine whether a specific feature category meets the accuracy requirements of the system during the question and answer process. Based on the feature deviation comparison result, the system performs a status judgment on each feature category. This judgment process may involve the classification of the degree of deviation (such as slight deviation, severe deviation, etc.) and generate corresponding feature status judgment information. This information is used to trigger subsequent operations such as feature optimization, model adjustment or data preprocessing.

[0152] The technical effect of the above technical solution is: by conducting a detailed analysis of the question-answering status of each feature category, the system can locate the problem more accurately, thereby providing a more targeted solution. According to the feature deviation coefficient and feature state determination information, the system can adjust the weight of the feature category to improve the accuracy and efficiency of the question-answering model. By identifying and optimizing the feature categories that cause deviations, the system can improve the overall performance of the question-answering model and reduce deviations and errors. The system can dynamically adjust and optimize the model based on the feature state determination information to adapt to the changing types of questions and user needs. The feature state determination information can provide valuable reference information for system administrators or decision makers, such as adjusting feature weights, optimizing model structure, or improving data preprocessing processes. By continuously monitoring and analyzing the feature deviation coefficient and feature state determination information, the system can continuously learn and optimize its own performance to maintain competitiveness and adaptability.

[0153] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A multifunctional question-answering interactive method based on AI intelligence, characterized in that: The method comprises: S1. Obtaining problem collection information, obtaining category feature extraction data according to the problem collection information, combining the category feature extraction data, obtaining problem feature combination data, filling in a preset feature combination, and obtaining feature combination filling data; S2. Calculate the filling standard coefficient of each question collection information, obtain filling standard comparison information, trigger the question feedback mechanism, obtain question feedback data, train the question-answering model, and output the newly added feedback data that meets the filling standard; S3, calculating the question-answering deviation coefficient of the newly added question collection information, obtaining question status determination information, and then obtaining comprehensive question-answering status determination information; S4. Trigger the question-answering status analysis of each preset feature category, calculate the question-answering deviation coefficient of each feature category, perform feature status determination, and obtain feature status determination information.

2. According to claim 1, a multifunctional question-answering interactive method based on AI intelligence is characterized in that: The S1 includes: Collect input problem information to obtain problem collection information; Obtaining a preset feature category to extract category features of each preset feature category from the problem collection information, and obtaining category feature extraction data; Combine all category feature extraction data of each problem collected data to obtain problem feature combination data; Obtain a preset feature combination, and fill the preset feature combination accordingly according to the problem feature combination data to obtain feature combination filling data.

3. According to claim 1, a multifunctional question-answering interactive method based on AI intelligence is characterized in that: The S2 includes: Calculate the filling standard coefficient of each question collection information by filling data with feature combinations; Comparing the filling standard coefficient with a preset standard threshold to obtain filling standard comparison information; Triggering a question feedback mechanism based on the filling standard comparison information, and obtaining question feedback data of the question collection information through a preset knowledge graph database; Train the question-answering model by collecting information from historical questions and combining it with question feedback data; The new question collection information is outputted through the question-answering model to obtain new feedback data.

4. According to claim 1, a multifunctional question-answering interactive method based on AI intelligence is characterized in that: The S3 includes: The question-answering deviation coefficient of the newly added question collection information is calculated by filling in the standard coefficient and combining the newly added feedback data; Compare the question-answer deviation coefficient with a preset deviation threshold to obtain a question-answer deviation comparison result; Determine the question and answer status according to the question and answer deviation comparison result to obtain question and answer status determination information; A comprehensive determination of the question and answer status is performed through the question and answer status determination information of all newly added question collection information to obtain the question and answer status comprehensive determination information.

5. According to claim 1, a multifunctional question-answering interactive method based on AI intelligence is characterized in that: The S4 includes: Triggering question and answer status analysis for each preset feature category based on comprehensive judgment information of question and answer status; The question-answer deviation coefficient of each feature category is calculated by combining the weight information of the preset feature category with the question-answer deviation coefficient to obtain the feature deviation coefficient; Comparing the characteristic deviation coefficient with a preset characteristic deviation threshold to obtain a characteristic deviation comparison result; According to the feature deviation comparison result, feature state determination is performed to obtain feature state determination information.

6. A multifunctional question-answering interactive system based on AI intelligence, characterized in that: The system comprises: A feature filling module is used to obtain problem collection information, obtain category feature extraction data according to the problem collection information, combine the category feature extraction data, obtain problem feature combination data, fill in preset feature combinations, and obtain feature combination filling data; The standard analysis output module is used to calculate the filling standard coefficient of each question collection information, obtain the filling standard comparison information, trigger the question feedback mechanism, obtain the question feedback data, train the question-answering model, and output the newly added feedback data that meets the filling standard; The question-answering status analysis module is used to calculate the question-answering deviation coefficient of the newly added question collection information, obtain the question status determination information, and then obtain the comprehensive determination information of the question-answering status; The feature deviation analysis module is used to trigger the question-answering status analysis of each preset feature category, calculate the question-answering deviation coefficient of each feature category, perform feature status determination, and obtain feature status determination information.

7. The multifunctional question-answering interactive system based on AI intelligence according to claim 6 is characterized in that: The feature filling module includes: A data collection module is used to collect input problem information and obtain problem collection information; A feature extraction module is used to obtain a preset feature category and extract category features of each preset feature category from the problem collection information to obtain category feature extraction data; A feature combination module is used to combine all category feature extraction data of each problem collection data to obtain problem feature combination data; The combination filling module is used to obtain a preset feature combination, and fill the preset feature combination accordingly according to the problem feature combination data to obtain feature combination filling data.

8. The multifunctional question-answering interactive system based on AI intelligence according to claim 6 is characterized in that: The standard analysis output module includes: A filling calculation module is used to calculate the filling standard coefficient of each problem collection information by filling data with feature combinations; A filling analysis module, used for comparing the filling standard coefficient with a preset standard threshold value to obtain filling standard comparison information; A model training output module is used to trigger a question feedback mechanism according to the filling standard comparison information, and obtain question feedback data of the question collection information through a preset knowledge graph database; Train the question-answering model by collecting information from historical questions and combining it with question feedback data; The new question collection information is outputted through the question-answering model to obtain new feedback data.

9. The multifunctional question-answering interactive system based on AI intelligence according to claim 6 is characterized in that: The question-answering status analysis module includes: A question-answering deviation calculation module, used to calculate the question-answering deviation coefficient of the newly added question collection information by filling in the standard coefficient and combining the newly added feedback data; A question-answering state determination module is used to compare the question-answering deviation coefficient with a preset deviation threshold to obtain a question-answering deviation comparison result; Determine the question and answer status according to the question and answer deviation comparison result to obtain question and answer status determination information; The question status comprehensive determination module is used to perform a comprehensive determination on the question and answer status through the question and answer status determination information of all newly added question collection information to obtain the question and answer status comprehensive determination information.

10. The multifunctional question-answering interactive system based on AI intelligence according to claim 6, characterized in that: The characteristic deviation analysis module comprises: A feature deviation calculation module is used to trigger the question-answering status analysis of each preset feature category according to the comprehensive judgment information of the question-answering status; The question-answer deviation coefficient of each feature category is calculated by combining the weight information of the preset feature category with the question-answer deviation coefficient to obtain the feature deviation coefficient; A feature state analysis module, used to compare the feature deviation coefficient with a preset feature deviation threshold to obtain a feature deviation comparison result; According to the feature deviation comparison result, feature state determination is performed to obtain feature state determination information.

Images