Intelligent question and answer method

Abstract

The invention discloses an intelligent question and answer method. The method comprises the steps of obtaining question information; judging the format type of the question information; when determining that the format type of the question information is a character type, judging whether the question information is a mathematical question; when determining that the question information is a mathematical question, performing question answering based on a first reply library; and when determining that the question information is a non-mathematical question, performing question answering based on the second reply library. And the input format type is automatically identified, so that the user can conveniently input the questioning information. According to the method, the mathematical problems and the non-mathematical problems are distinguished for the problems of the user, and the mathematical problems and the non-mathematical problems are compared based on different databases, so that the data comparison rate and accuracy are improved, and the response fluency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent question answering, and in particular to an intelligent question answering method. Background Art

[0002] At present, with the rapid development of the Internet and artificial intelligence technology, intelligent question-answering systems have been applied. When students encounter learning problems that they do not understand, they can solve them through intelligent question-answering systems. In the existing intelligent question-answering field, it is impossible to automatically identify the input format type and perform corresponding format conversion, which is inconvenient for users to input question information. In the field of intelligent question-answering, mathematical problems are computationally intensive problems, so they involve a relatively large database of other questions. In the prior art, user questions are directly compared based on a fixed big data model, without distinguishing between mathematical problems and non-mathematical problems. The big data model includes a large question bank. For non-mathematical problems, the comparison rate is slow, which reduces the fluency of the response. Summary of the invention

[0003] The present invention aims to solve at least one of the technical problems in the above-mentioned technology to a certain extent. To this end, the purpose of the present invention is to propose an intelligent question-answering method, which automatically identifies the input format type and facilitates the user to input question information. The user's questions are distinguished between mathematical questions and non-mathematical questions, and mathematical questions and non-mathematical questions are compared based on different databases, which is convenient for improving the speed and accuracy of data comparison and improving the fluency of answering.

[0004] To achieve the above object, an embodiment of the present invention proposes an intelligent question-answering method, comprising:

[0005] Get question information;

[0006] Determine the format type of the question information;

[0007] When it is determined that the format type of the question information is a text type, determining whether the question information is a math question;

[0008] When it is determined that the question information is a math question, answering the question based on the first answer library;

[0009] When it is determined that the question information is a non-mathematical question, the question is answered based on the second answer library.

[0010] According to some embodiments of the present invention, the further step includes: when it is determined that the format type of the question information is a picture type, performing a type conversion process to convert the picture type into a text type.

[0011] According to some embodiments of the present invention, converting the image type into the text type includes:

[0012] Inputting the question information of the picture type into a pre-trained picture recognition model to determine the picture recognition information; the picture recognition information includes the text elements in the question information and the coordinate position information of each text element;

[0013] Input the image recognition information into a pre-trained semantic segmentation model to obtain a segmentation result, wherein the segmentation result includes a text segmentation part, a table segmentation part, a handwriting segmentation part, and an illustration segmentation part;

[0014] According to the segmentation results and the preset segmentation part-text tag mapping data table, it is converted into text-type question information.

[0015] According to some embodiments of the present invention, when it is determined that the question information is a math question, answering the question based on the first answer library includes:

[0016] Segment the question information into types to obtain plain text and formula text;

[0017] Perform formula analysis on the formula book to obtain a first analysis result;

[0018] Conduct question type analysis on the plain text to obtain the second analysis result;

[0019] Vectorizing the question set in the first answer library; each question in the question set includes a preset first analysis result and a preset second analysis result;

[0020] performing vectorization processing on the first analysis result and the second analysis result;

[0021] Performing similarity matching on the vectorized first analysis result and the preset first analysis result in the problem set to obtain a first matching result;

[0022] Performing similarity matching on the vectorized second analysis result and the preset second analysis result in the question set to obtain a second matching result;

[0023] Obtaining a third matching result according to the first matching result, the second matching result and a preset weighting coefficient; determining the question with the highest similarity in the third matching result as the target question;

[0024] The target answer corresponding to the target question is retrieved from the first answer library, and the target answer is formed into a complete answer using the encoder-decoder and word embedding form in deep learning and then output to obtain the answer result.

[0025] According to some embodiments of the present invention, formula analysis is performed on the formula book to obtain a first analysis result, including:

[0026] The formula book is analyzed based on the formula library to obtain a first analysis result.

[0027] According to some embodiments of the present invention, question type analysis is performed on the plain text to obtain a second analysis result, including:

[0028] The plain text is segmented based on the option segmentation symbol and the sentence segmentation symbol to obtain several sentences;

[0029] Perform word segmentation and part-of-speech tagging on each sentence to determine the target keywords;

[0030] According to the position of each sentence in the plain text and the characteristic keywords contained therein, the attribute type of each sentence is determined, and the attribute type is divided into a narration sentence, a conditional sentence, a question sentence, and a redundant sentence;

[0031] The clauses with the attribute type of conditional clause or interrogative clause are regarded as key clauses;

[0032] The key sentences are semantically parsed to obtain the second analysis result.

[0033] According to some embodiments of the present invention, the first answer library is a network search library of the knowledge model, a mathematical knowledge library, and a custom question library;

[0034] The second answer library is a network search library of the knowledge model.

[0035] According to some embodiments of the present invention, when it is determined that the question information is a math question, answering the question based on the first answer library includes:

[0036] Taking the knowledge points currently contained in the question information as the prerequisite, searching for the corresponding mathematical operation rules in the first answer library;

[0037] Determine the instantiation rule according to the matched mathematical operation rules and the knowledge points currently contained in the question information;

[0038] Solve the problem according to the instantiation rules and get the answer.

[0039] According to some embodiments of the present invention, the further comprising:

[0040] The question information and the corresponding answer results are combined into question-answer pair data;

[0041] Calculate the question recognition error rate FPR and service satisfaction rate TPR based on the question-answer pair data;

[0042] When it is determined that the recognition error rate FPR is greater than the preset error rate threshold and the service satisfaction rate TPR is less than the preset satisfaction rate threshold, the first reply library and the second reply library are updated.

[0043] According to some embodiments of the present invention, determining whether the question information is a math question includes:

[0044] Calculate the characteristic value of the question information;

[0045]

[0046] Among them, a1 indicates whether it contains a numeric identifier; a2 indicates whether it contains a letter identifier; a3 indicates whether it contains a Chinese character identifier; a4 indicates whether it contains a special character identifier; p is the characteristic value of the question information;

[0047] The characteristic value of the question information is compared with the preset characteristic value. When it is determined that the characteristic value of the question information is greater than the preset characteristic value, it indicates that the question information is a math problem; otherwise, it indicates that the question information is not a math problem.

[0048] The present invention proposes an intelligent question-answering method, which automatically identifies the input format type, and facilitates the user to input question information. The user's questions are distinguished between mathematical questions and non-mathematical questions, and mathematical questions and non-mathematical questions are compared based on different databases, which is convenient for improving the speed and accuracy of data comparison and improving the fluency of answering.

[0049] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description and the accompanying drawings.

[0050] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0052] Figure 1 is a flow chart of an intelligent question-answering method according to an embodiment of the present invention;

[0053] Figure 2 The flowchart of converting a picture type into a text type according to an embodiment of the present invention. DETAILED DESCRIPTION

[0054] 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.

[0055] like Figure 1 As shown, the embodiment of the present invention proposes an intelligent question-answering method, including steps S1-S5:

[0056] S1. Obtain question information;

[0057] S2, determining the format type of the question information;

[0058] S3. When it is determined that the format type of the question information is a text type, determining whether the question information is a math question;

[0059] S4. When it is determined that the question information is a math question, answering the question based on the first answer library;

[0060] S5. When it is determined that the question information is a non-mathematical question, answer the question based on the second answer library.

[0061] The working principle of the above technical solution: Picture format, the system should be able to automatically detect and recognize the text content in the picture (such as using OCR technology). The data content of the first reply library is more than the data content of the second reply library, and the data content of the first reply library includes the data content of the second reply library.

[0062] The above technical solution has the following beneficial effects: it can effectively process different types of question information, provide accurate and timely answers, automatically identify the input format type, and facilitate users to input question information. It can distinguish between mathematical questions and non-mathematical questions for user questions, and compare mathematical questions and non-mathematical questions based on different databases, which is convenient for improving the speed and accuracy of data comparison and improving the fluency of answers.

[0063] According to some embodiments of the present invention, the further step includes: when it is determined that the format type of the question information is a picture type, performing a type conversion process to convert the picture type into a text type.

[0064] The working principle and beneficial effects of the above technical solution are: the text content in the image is extracted, thereby improving the practicality of the system and the user experience.

[0065] like Figure 2 As shown, according to some embodiments of the present invention, converting the image type into a text type includes steps S61-S63:

[0066] S61, inputting the question information of the picture type into a pre-trained picture recognition model to determine the picture recognition information; the picture recognition information includes the text elements in the question information and the coordinate position information of each text element;

[0067] S62, inputting the image recognition information into a pre-trained semantic segmentation model to obtain a segmentation result, wherein the segmentation result includes a text segmentation part, a table segmentation part, a handwriting segmentation part, and a figure segmentation part;

[0068] S63, converting into text-type question information according to the segmentation result and the preset segmentation part-text mark mapping data table.

[0069] The working principle of the above technical solution is: use OCR (optical character recognition) technology to identify text elements in the picture and their coordinate position information. Use a semantic segmentation model to further subdivide the text elements in the picture, including text segmentation parts, table segmentation parts, handwriting segmentation parts and illustration segmentation parts. Based on the semantic segmentation model, it is possible to accurately distinguish different types of text elements and illustrations to avoid information confusion. Models based on attention mechanisms or graph convolutional networks (GCN) and other technologies are used to improve segmentation accuracy. The segmented text elements are sorted, reorganized and formatted to restore the original text information, and the text elements in the segmentation results are converted into readable text type question information. The preset segmentation part-text mark mapping data table should accurately reflect the correspondence between different types of text elements and the final text type question information. During the conversion process, natural language processing (NLP) technology is used to further correct typos, adjust word order, etc. to improve the accuracy and readability of text information.

[0070] The beneficial effects of the above technical solution are: effectively converting the image type into the text type, providing accurate and readable input information for the intelligent question-answering system, and enriching the format types of the user input question information.

[0071] According to some embodiments of the present invention, when it is determined that the question information is a math question, answering the question based on the first answer library includes:

[0072] Segment the question information into types to obtain plain text and formula text;

[0073] Perform formula analysis on the formula book to obtain a first analysis result;

[0074] Conduct question type analysis on the plain text to obtain the second analysis result;

[0075] Vectorizing the question set in the first answer library; each question in the question set includes a preset first analysis result and a preset second analysis result;

[0076] performing vectorization processing on the first analysis result and the second analysis result;

[0077] Performing similarity matching on the vectorized first analysis result and the preset first analysis result in the problem set to obtain a first matching result;

[0078] Performing similarity matching on the vectorized second analysis result and the preset second analysis result in the question set to obtain a second matching result;

[0079] Obtaining a third matching result according to the first matching result, the second matching result and a preset weighting coefficient; determining the question with the highest similarity in the third matching result as the target question;

[0080] The target answer corresponding to the target question is retrieved from the first answer library, and the target answer is formed into a complete answer using the encoder-decoder and word embedding form in deep learning and then output to obtain the answer result.

[0081] The working principle of the above technical solution is: the question information is divided into a plain text part and a formula part. The plain text part contains the description and conditions of the question, while the formula part contains mathematical expressions or equations. The formula library, i.e. the mathematical formula parsing library, is used to parse the formula and extract key information (such as variables, operators, functions, etc.). The plain text part is semantically analyzed to determine the type of question, the corresponding condition information and the target description of the solution.

[0082] The question set in the first answer library is converted into a vector form for subsequent similarity matching. Each question includes a preset first analysis result (formula analysis) and a preset second analysis result (question type analysis). Use word embedding technology (such as Word2Vec, BERT, etc.) to convert text and formulas into vectors. The first analysis result and the second analysis result of the current question information are also converted into vector form. Use measurement methods such as cosine similarity and Euclidean distance to calculate the similarity between the formula analysis of the current question information and the formula analysis in the question set in the first answer library. Similarly, calculate the similarity between the question type analysis of the current question information and the question type analysis in the question set in the first answer library. Combine the first matching result and the second matching result, as well as the preset weighting coefficient, to calculate the comprehensive similarity, and determine the question with the highest similarity as the target question. Determine the weighting coefficient based on the data ratio of the plain text and the formula book, retrieve the answer corresponding to the target question from the first answer library, and use the encoder-decoder and word embedding technology in deep learning to generate a complete and readable answer.

[0083] The beneficial effect of the above technical solution is: accurately answering mathematical problems based on the first answer library.

[0084] According to some embodiments of the present invention, formula analysis is performed on the formula book to obtain a first analysis result, including:

[0085] The formula book is analyzed based on the formula library to obtain a first analysis result.

[0086] The working principle of the above technical solution: The formula library contains various known, commonly used or related mathematical, physical, chemical and other formulas, and also contains information such as the standard form of the formula, the meaning of variables, and applicable conditions. Read out each formula in the formula book one by one, and match each read formula with the formula in the formula library. The matching criteria are the form of the formula, the use of variables, the applicable physical or mathematical background, etc. For each formula, record its matching status in the formula library. If the match is successful, record the matched formula information (such as standard form, variable meaning, etc.). The first analysis result includes: a list of successfully matched formulas and their detailed information.

[0087] The beneficial effect of the above technical solution is: a comprehensive and systematic analysis is performed on the formulas in the formula book, and a first analysis result is obtained.

[0088] According to some embodiments of the present invention, question type analysis is performed on the plain text to obtain a second analysis result, including:

[0089] The plain text is segmented based on the option segmentation symbol and the sentence segmentation symbol to obtain several sentences;

[0090] Perform word segmentation and part-of-speech tagging on each sentence to determine the target keywords;

[0091] According to the position of each sentence in the plain text and the characteristic keywords contained therein, the attribute type of each sentence is determined, and the attribute type is divided into a narration sentence, a conditional sentence, a question sentence, and a redundant sentence;

[0092] The clauses with the attribute type of conditional clause or interrogative clause are regarded as key clauses;

[0093] The key sentences are semantically parsed to obtain the second analysis result.

[0094] The working principle of the above technical solution is as follows: the plain text is preliminarily segmented based on the option segmentation symbols (such as commas, semicolons, periods, question marks, etc.) to obtain several preliminary sentences. Each preliminary sentence is further subdivided according to the sentence segmentation symbols (such as line breaks, paragraph separators, etc.) to ensure that each sentence is an independent and complete sentence. Each sentence is segmented and split into separate words. The results after word segmentation are tagged with parts of speech, and each word is assigned a corresponding part of speech (such as nouns, verbs, adjectives, etc.). According to the part-of-speech tagging results and the context, the target keywords in each sentence are determined. These keywords are related to the requirements, conditions, questions, etc. of the topic. Each sentence is marked with the part of speech and the target keywords are determined. According to the position of each sentence in the plain text (such as the beginning, middle, end, etc.) and the characteristic keywords contained (such as "if", "then", "please ask", "because", etc.), the attribute type of each sentence is determined. There are four types of attributes: narration clauses (describing background or situation), conditional clauses (posing conditions or restrictions), interrogative clauses (posing questions or doubts), and redundant clauses (information that is irrelevant to the question requirements or is repeated). Filter out clauses with the attribute type of conditional clauses or interrogative clauses from all clauses as key clauses. These key clauses contain the core requirements of the question or the questions that need to be answered. Perform semantic analysis on each key clause to understand its deep meaning and logical relationship. Semantic analysis includes analysis of the association between words, analysis of sentence structure, and understanding of the context. Based on the results of semantic analysis, the second analysis result is obtained, which includes the core requirements of the question, the questions that need to be answered, conditional restrictions, etc.

[0095] The beneficial effects of the above technical solution are as follows: the plain text is segmented to obtain a set of sentences, and then the sentences are segmented and POS tagged to determine the keywords, and then the attribute type is determined according to the position and characteristic keywords of the sentences, and the key sentences are screened out and semantic analysis is performed, and finally the second analysis result is obtained. The requirements and structure of the question are accurately determined, which facilitates the accurate comparison of the second analysis result.

[0096] According to some embodiments of the present invention, the first answer library is a network search library of the knowledge model, a mathematical knowledge library, and a custom question library;

[0097] The second answer library is a network search library of the knowledge model.

[0098] The working principle and beneficial effects of the above technical solution: The first answer library is a comprehensive knowledge repository that integrates multiple resources to meet a wide range of query needs. The first answer library includes: a network search library of knowledge models: this part is built based on a large amount of knowledge and information on the Internet, and contains content in various forms such as web pages, documents, and forum discussions. It uses search engine technology to quickly locate and return web pages or documents related to the query. Mathematical knowledge base: this is a knowledge base focused on the field of mathematics, containing mathematical formulas, theorems, problem-solving skills, mathematical concepts, etc. It aims to provide users with accurate and comprehensive mathematical knowledge and solutions. Customized question bank: this part is a question bank created by users or systems according to specific needs, including exercises, test questions, simulation questions, etc. It allows users to customize personalized learning plans and resources according to their own learning progress and needs. The second answer library is a network search library that focuses more on knowledge models. The main function of the second answer library is to provide search results based on knowledge models. It uses advanced information retrieval and artificial intelligence technologies to more accurately understand the user's query intent and return results that are highly matched with user needs.

[0099] According to some embodiments of the present invention, when it is determined that the question information is a math question, answering the question based on the first answer library includes:

[0100] Taking the knowledge points currently contained in the question information as the prerequisite, searching for the corresponding mathematical operation rules in the first answer library;

[0101] Determine the instantiation rule according to the matched mathematical operation rules and the knowledge points currently contained in the question information;

[0102] Solve the problem according to the instantiation rules and get the answer.

[0103] The working principle and beneficial effects of the above technical solution are as follows: segment and tag the question information. Use natural language processing technology to identify the mathematical terms and concepts in the question. According to the recognition results, determine the main knowledge points currently contained in the question information. According to the extracted knowledge points, search in the first answer library. Filter out the mathematical operation rules that match the knowledge points. Ensure that the selected rules are consistent with the context of the question information. After finding the matching mathematical operation rules, these rules need to be instantiated according to the specific values ​​or conditions in the question information. Instantiated rules refer to applying abstract mathematical operation rules to specific mathematical problems to form executable problem-solving steps. Specifically: analyze the values, variables and conditions in the question information. Substitute these values, variables and conditions into the found mathematical operation rules. Determine the specific problem-solving steps and strategies according to the substituted rules. Execute the problem-solving steps determined in the instantiated rules. Perform necessary mathematical operations and logical reasoning. Obtain the solution results, and verify and check them to ensure the correctness and rationality of the results. Accurately identify the mathematical knowledge points in the question information, find and apply the corresponding mathematical operation rules in the first answer library, and finally get the answer. This process makes full use of the mathematical knowledge resources in the first answer library, and combines technical means such as natural language processing and mathematical operations to achieve automatic answers to mathematical problems.

[0104] According to some embodiments of the present invention, the further comprising:

[0105] The question information and the corresponding answer results are combined into question-answer pair data;

[0106] Calculate the question recognition error rate FPR and service satisfaction rate TPR based on the question-answer pair data;

[0107] When it is determined that the recognition error rate FPR is greater than the preset error rate threshold and the service satisfaction rate TPR is less than the preset satisfaction rate threshold, the first reply library and the second reply library are updated.

[0108] The working principle and beneficial effects of the above technical solution: Whenever the system receives a question information and gives an answer result, the pair of information will be recorded to form question-answer pair data. FPR (False Positive Rate): FPR measures the proportion of non-mathematical questions that the system mistakenly recognizes as mathematical questions. FPR = the number of mathematical questions mistakenly recognized / the total number of questions. TPR (True Positive Rate): TPR measures the proportion of mathematical questions that the system correctly answers. TPR = the number of mathematical questions correctly answered / the total number of mathematical questions. After calculating FPR and TPR, they are compared with the preset error rate threshold and service satisfaction rate threshold. If FPR is greater than the preset error rate threshold and TPR is less than the preset satisfaction rate threshold, it means that the performance of the system needs to be improved and the answer library needs to be updated. The update includes adding new knowledge points, optimizing mathematical operation rules, updating custom question banks, etc. Question-answer pair data can be effectively used to evaluate and optimize the performance of the system, and the answer library can be updated as needed to improve the accuracy and satisfaction of the system.

[0109] According to some embodiments of the present invention, determining whether the question information is a math question includes:

[0110] Calculate the characteristic value of the question information;

[0111]

[0112] Among them, a1 indicates whether it contains a numeric identifier; a2 indicates whether it contains a letter identifier; a3 indicates whether it contains a Chinese character identifier; a4 indicates whether it contains a special character identifier; p is the characteristic value of the question information;

[0113] The characteristic value of the question information is compared with the preset characteristic value. When it is determined that the characteristic value of the question information is greater than the preset characteristic value, it indicates that the question information is a math problem; otherwise, it indicates that the question information is not a math problem.

[0114] The working principle and beneficial effects of the above technical solution: special characters are operation characters. For each variable, if the question information contains the corresponding identifier, it is assigned a value of 1; if not, it is assigned a value of 0. Multiply the value of each variable by the corresponding weight (power of 2), and then sum them up to get the eigenvalue p. After calculating the eigenvalue p, compare it with a preset eigenvalue threshold. This threshold is set based on historical data and experience, and is used to distinguish between math problems and non-mathematical problems. If p is greater than the preset eigenvalue threshold, the question information is judged to be a math problem. If p is less than or equal to the preset eigenvalue threshold, the question information is judged not to be a math problem. A preliminary judgment is made as to whether the question information is a math problem.

[0115] 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. An intelligent question-answering method, characterized in that: include: Get question information; Determine the format type of the question information; When it is determined that the format type of the question information is a text type, determining whether the question information is a math question; When it is determined that the question information is a math question, answering the question based on the first answer library; When it is determined that the question information is a non-mathematical question, the question is answered based on the second answer library.

2. The intelligent question-answering method according to claim 1, characterized in that: Also includes: When it is determined that the format type of the question information is a picture type, a type conversion process is performed to convert the picture type into a text type.

3. The intelligent question-answering method according to claim 2, characterized in that: Convert image types to text types, including: Inputting the question information of the picture type into a pre-trained picture recognition model to determine the picture recognition information; the picture recognition information includes the text elements in the question information and the coordinate position information of each text element; Input the image recognition information into a pre-trained semantic segmentation model to obtain a segmentation result, wherein the segmentation result includes a text segmentation part, a table segmentation part, a handwriting segmentation part, and an illustration segmentation part; According to the segmentation results and the preset segmentation part-text tag mapping data table, it is converted into text-type question information.

4. The intelligent question-answering method according to claim 1, wherein: When it is determined that the question information is a math question, the question is answered based on the first answer library, including: Segment the question information into types to obtain plain text and formula text; Perform formula analysis on the formula book to obtain a first analysis result; Conduct question type analysis on the plain text to obtain the second analysis result; Vectorizing the question set in the first answer library; each question in the question set includes a preset first analysis result and a preset second analysis result; performing vectorization processing on the first analysis result and the second analysis result; Performing similarity matching on the vectorized first analysis result and the preset first analysis result in the problem set to obtain a first matching result; Performing similarity matching on the vectorized second analysis result and the preset second analysis result in the question set to obtain a second matching result; Obtaining a third matching result according to the first matching result, the second matching result and a preset weighting coefficient; determining the question with the highest similarity in the third matching result as the target question; The target answer corresponding to the target question is retrieved from the first answer library, and the target answer is formed into a complete answer using the encoder-decoder and word embedding form in deep learning and then output to obtain the answer result.

5. The intelligent question-answering method according to claim 4, characterized in that: The formula book is analyzed to obtain a first analysis result, including: The formula book is analyzed based on the formula library to obtain a first analysis result.

6. The intelligent question-answering method according to claim 4, characterized in that: The question type analysis is performed on the plain text to obtain the second analysis result, including: The plain text is segmented based on the option segmentation symbol and the sentence segmentation symbol to obtain several sentences; Perform word segmentation and part-of-speech tagging on each sentence to determine the target keywords; According to the position of each sentence in the plain text and the characteristic keywords contained therein, the attribute type of each sentence is determined, and the attribute type is divided into a narration sentence, a conditional sentence, a question sentence, and a redundant sentence; The clauses with the attribute type of conditional clause or interrogative clause are regarded as key clauses; The key sentences are semantically parsed to obtain the second analysis result.

7. The intelligent question-answering method according to claim 1, characterized in that: The first answer library is a network search library of the knowledge model, a mathematical knowledge library and a custom question library; The second answer library is a network search library of the knowledge model.

8. The intelligent question-answering method according to claim 1, characterized in that: When it is determined that the question information is a math question, the question is answered based on the first answer library, including: Taking the knowledge points currently contained in the question information as the prerequisite, searching for the corresponding mathematical operation rules in the first answer library; Determine the instantiation rule according to the matched mathematical operation rules and the knowledge points currently contained in the question information; Solve the problem according to the instantiation rules and get the answer.

9. The intelligent question-answering method according to claim 1, wherein: Also includes: The question information and the corresponding answer results are combined into question-answer pair data; Calculate the question recognition error rate FPR and service satisfaction rate TPR based on the question-answer pair data; When it is determined that the recognition error rate FPR is greater than the preset error rate threshold and the service satisfaction rate TPR is less than the preset satisfaction rate threshold, the first reply library and the second reply library are updated.

10. The intelligent question-answering method according to claim 1, wherein: Determine whether the question is a math question, including: Calculate the characteristic value of the question information; Among them, a1 indicates whether it contains a numeric identifier; a2 indicates whether it contains a letter identifier; a3 indicates whether it contains a Chinese character identifier; a4 indicates whether it contains a special character identifier; p is the characteristic value of the question information; The characteristic value of the question information is compared with the preset characteristic value. When it is determined that the characteristic value of the question information is greater than the preset characteristic value, it indicates that the question information is a math problem; otherwise, it indicates that the question information is not a math problem.

Images