Knowledge question and answer method and device, electronic equipment and readable storage medium

Abstract

The invention provides a knowledge question-answering method and device, electronic equipment and a readable storage medium. The method comprises the following steps: extracting at least one target professional noun in a user question; based on the at least one target professional noun and a pre-constructed professional noun tree, determining at least one associated professional noun having an association relationship with the at least one target professional noun and interpretation information of each professional noun; adjusting the user question based on the at least one associated professional noun to obtain at least one adjustment question; based on the user question and the at least one adjustment question, retrieving at least one target retrieval result matched with the user question and / or the at least one adjustment question from a preset target knowledge base; and inputting the user question, the explanation information of each professional noun and at least one target retrieval result into a trained large language model, and outputting target answer information corresponding to the user question. In this way, the accuracy of the output target answer and the user experience can be improved.

Description

Technical Field

[0001] The present application relates to the field of computers and technology, and in particular to a knowledge question answering method, device, electronic device, and readable storage medium. Background Art

[0002] In related technologies, users may have the need for consulting services. With the rapid development of science and technology, Retrieval-Augmented Generation (RAG) technology can be used to assist in answering users' questions. Specifically, after determining that the question input by the user has been received, relevant knowledge is retrieved in the vertical field knowledge base based on the user's question. The retrieved relevant knowledge is combined with the user's question to allow the large model to generate answers, which can significantly reduce the complexity of answer setting and the accuracy of the answer.

[0003] However, when searching for user questions, the knowledge base generally stores data on professional terms in vertical fields, and users may not describe the corresponding professional terms accurately when asking questions, which leads to a low match between the search results and user questions when searching for knowledge. At the same time, if the professional terms are not explained, the large language model may not be able to understand the user questions and related knowledge, which leads to low accuracy of the output target answers, affecting the user experience. Summary of the Invention

[0004] In view of this, the purpose of this application is to provide a knowledge question and answer method, device, electronic device and readable storage medium, which can extract the target professional terms in the user question after determining that the user question input by the user is received, and combine with the pre-built professional term tree to obtain related terms and explanations associated with the target professional terms, and then expand the user question. After matching with the target knowledge base, a target retrieval result with a higher matching degree and a wider matching range can be obtained, and then the user question, the explanation for the target professional term and the target retrieval result are input into the large language model to output the target answer, which helps to improve the accuracy of the output target answer and the user experience.

[0005] In a first aspect, an embodiment of the present application provides a knowledge question answering method, which is applied to a knowledge question answering system. The knowledge question answering method includes: Obtaining a user question input by a user, and extracting at least one target professional term contained in the user question; Based on the at least one target professional noun and a pre-constructed professional noun tree, determining at least one associated professional noun associated with the at least one target professional noun and explanation information for each associated professional noun; adjust the user question based on the at least one associated professional term, to obtain at least one adjusted question; retrieve at least one target search result matched with the user question and / or the at least one adjusted question from a pre-set target knowledge base based on the user question and the at least one adjusted question; input the user question, the explanation information of each associated professional term, and the at least one target search result into a trained large language model, and output target answer information corresponding to the user question.

[0006] In an optional implementation, the extracting the at least one target professional term contained in the user question comprises: input the user question into a trained professional term recognition model, and output an input professional term contained in the user question; calculate the similarity between each candidate professional term contained in the professional term tree and the input professional term; determine at least one candidate professional term with a similarity greater than a first preset similarity threshold to the input professional term as the at least one target professional term contained in the user question.

[0007] In an optional implementation, for each candidate professional term, the similarity between the candidate professional term and the input professional term is determined by the following steps: input the input professional term into a trained vector processing model, and output a first term vector corresponding to the input professional term; for each candidate professional term, input the candidate professional term into the vector processing model, and output a second term vector corresponding to the candidate professional term; calculate the similarity between the first term vector and the second term vector to determine the similarity between the candidate professional term and the input professional term.

[0008] In an optional implementation, the determining at least one associated professional term having an associated relationship with the at least one target professional term based on the at least one target professional term and a pre-constructed professional term tree comprises: normalize the at least one target professional term to determine at least one normalized standard professional term; for each standard professional term, determine at least one associated professional term having an associated relationship with the standard professional term based on a plurality of professional term paths contained in the professional term tree and a preset number of jumps.

[0009] In an optional embodiment, for each standard professional term, determining at least one associated professional term associated with the standard professional term based on multiple professional term paths included in the professional term tree and a preset jump number includes: For each standard professional noun, determining at least one target professional noun path containing the standard professional noun from a plurality of professional noun paths contained in the professional noun tree; For each target professional noun path, starting from the location of the corresponding standard professional noun, jumping on the target professional noun path according to the preset jump number, and determining the candidate professional noun corresponding to the jump end position as the associated professional noun.

[0010] In an optional embodiment, adjusting the user question based on the at least one associated professional term to obtain at least one adjusted question includes: The target professional term contained in the user question is replaced by each determined associated professional term to obtain at least one adjusted question.

[0011] In an optional embodiment, the retrieving, based on the user question and the at least one adjustment question, at least one target retrieval result matching the user question and / or the at least one adjustment question from a preset target knowledge base includes: Determining a user question encoding vector corresponding to the user question and an adjustment question encoding vector corresponding to each adjustment question; Calculating the similarity between the retrieval result encoding vector contained in the target knowledge base and the user question encoding vector and / or the adjustment question encoding vector in the target knowledge base; At least one retrieval result encoding vector having a similarity with the user question encoding vector and / or the adjustment question encoding vector greater than a second preset similarity threshold is determined as the at least one target retrieval result.

[0012] In an optional embodiment, the professional term tree is constructed by the following steps: Input the acquired multiple historical question and answer text information into a pre-trained professional term recognition model, and output multiple extracted professional terms; For multiple extracted professional nouns, based on the semantic relationship between the extracted professional nouns, multiple professional noun nodes and the superordinate and subordinate relationships between the professional noun nodes are determined; The professional noun tree is constructed based on a plurality of professional noun nodes, the superordinate and subordinate relationships between the professional noun nodes, and the explanation information of each professional noun.

[0013] In an optional implementation, in the professional term tree, a professional term node corresponding to a superordinate professional term is a parent node of a professional term node corresponding to a subordinate professional term; professional term nodes corresponding to professional terms that are synonymous with each other or are antonyms of each other are located at the same level in the professional term tree.

[0014] In a second aspect, the embodiments of the present application further provide a knowledge question answering device, applied to a knowledge question answering system, the knowledge question answering method device comprising: a target professional term extraction module configured to obtain a user question input by a user and extract at least one target professional term contained in the user question; a related professional term determination module configured to determine, based on the at least one target professional term and a pre-constructed professional term tree, at least one related professional term having a correlation relationship with the at least one target professional term and explanation information of each related professional term; a user question adjustment module configured to adjust the user question based on the at least one related professional term to obtain at least one adjusted question; a knowledge base retrieval module configured to retrieve, based on the user question and the at least one adjusted question, at least one target retrieval result matching the user question and / or the at least one adjusted question from a pre-set target knowledge base; a target answer output module configured to input the user question, the explanation information of each related professional term and the at least one target retrieval result into a trained large language model to output target answer information corresponding to the user question.

[0015] In a third aspect, the embodiments of the present application further provide an electronic device, comprising a processor, a storage medium and a bus, the storage medium storing machine readable instructions executable by the processor, when the electronic device is running, the processor and the storage medium communicate through the bus, the processor executes the machine readable instructions to perform the steps of the knowledge question answering method according to any one of the first aspect.

[0016] In a fourth aspect, the embodiments of the present application further provide a computer readable storage medium, the computer readable storage medium storing a computer program, when the computer program is run by a processor, the steps of the knowledge question answering method according to any one of the first aspect are executed.

[0017] The knowledge question-answering method, device, electronic device and readable storage medium provided in the embodiments of the present application obtain a user question input by a user and extract at least one target professional term contained in the user question; based on at least one target professional term and a pre-constructed professional term tree, determine at least one associated professional term that has an association relationship with at least one target professional term and explanation information for each professional term; adjust the user question based on the at least one associated professional term to obtain at least one adjusted question; based on the user question and at least one adjusted question, retrieve at least one target retrieval result that matches the user question and / or at least one adjusted question from a pre-set target knowledge base; input the user question, the explanation information for each professional term and at least one target retrieval result into a trained large language model, and output target answer information corresponding to the user question. In this way, after confirming that the user question input by the user has been received, the target professional terms in the user question can be extracted, and combined with the pre-built professional term tree, the related terms and explanations associated with the target professional terms can be obtained, and then the user question can be expanded. After matching with the target knowledge base, a target retrieval result with a higher matching degree and a wider matching range can be obtained, and then the user question, the explanation for the target professional term and the target retrieval result can be input into the large language model to output the target answer, which helps to improve the accuracy of the output target answer and the user experience.

[0018] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without creative work.

[0020] Figure 1 A flowchart of a knowledge question answering method provided in an embodiment of the present application; Figure 2 A schematic diagram of the structure of the professional term tree provided in the embodiment of this application; Figure 3 A flowchart for constructing a professional term tree provided in an embodiment of the present application; Figure 4 A schematic diagram of the process of obtaining standard professional terms provided in the embodiment of this application; Figure 5 This is one of the structural diagrams of a knowledge question-answering device provided in an embodiment of the present application; Figure 6 This is a second structural diagram of a knowledge question-answering device provided in an embodiment of the present application; Figure 7 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. The components of the embodiments of the present application generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application for which protection is claimed, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, each other embodiment obtained by those skilled in the art without making creative work falls within the scope of protection of the present application.

[0022] First, the application scenarios to which this application is applicable are introduced. This application can be applied in the field of computer technology.

[0023] In related technologies, users may have the need to consult business, and in the early days, answers may be given through manual customer service. However, in this model, the number of service personnel is limited, and customers need to wait in line for a long time during peak consultation periods, resulting in a poor customer experience. In order to solve the above problems, combined with the rapidly developing artificial intelligence technology, automated question-and-answer methods are used to partially replace manual customer service, guiding customers to use intelligent customer service first, and only switching to manual customer service when the intelligent customer service cannot answer the user's questions, thereby reducing customer waiting time and improving the user experience.

[0024] In an optional implementation, the earliest implementation of the intelligent question-answering system is FAQ (Frequently Asked Questions), which is a domain knowledge base formed by organizing a set of questions and answers of some possible common questions in advance, and publishing them on a web page to provide consulting services to users. When a user asks a question, a keyword matching method is adopted to split the user's question into keywords and then match the keywords with the common questions and the corresponding answers, and return the questions and corresponding answers that successfully hit the keywords in the user's question. The disadvantages of this solution are very obvious, mainly including: the need for manual continuous collection and maintenance of common questions; the expression of the customer's question must literally coincide with the existing expression in the domain knowledge base, otherwise no relevant content can be retrieved. The system does not understand the user's question. The system returns a question and answer pair related to the user's question, and cannot directly generate an answer based on the user's question. The user needs to understand and organize the answer based on the returned information.

[0025] Furthermore, Retrieval-Augmented Generation (RAG) technology can be used. After confirming that the question input by the user is received, relevant knowledge is retrieved in the vertical domain knowledge base based on the user's question. The retrieved relevant knowledge is combined with the user's question to allow the large model to generate the answer, which can significantly reduce the complexity of answer setting and the accuracy of the answer.

[0026] However, when searching for user questions, the knowledge base generally stores data on professional terms in vertical fields, and users may not describe the corresponding professional terms accurately when asking questions, which leads to a low match between the search results and user questions when searching for knowledge. At the same time, if the professional terms are not explained, the large language model may not be able to understand the user questions and related knowledge, which leads to low accuracy of the output target answers, affecting the user experience.

[0027] Based on this, an embodiment of the present application provides a knowledge question-answering method to improve the accuracy of the output target answers and user experience.

[0028] See also Figure 1 , Figure 1 This is a flow chart of a knowledge question answering method provided in an embodiment of the present application. Figure 1 As shown in , the knowledge question answering method provided in the embodiment of the present application includes: S101: Obtain a user question input by a user, and extract at least one target professional term contained in the user question.

[0029] S102: Based on the at least one target professional noun and a pre-constructed professional noun tree, determine at least one associated professional noun associated with the at least one target professional noun and explanation information for each associated professional noun.

[0030] S103: Adjust the user question based on the at least one associated professional term to obtain at least one adjusted question.

[0031] S104: Based on the user question and the at least one adjustment question, retrieve at least one target retrieval result that matches the user question and / or the at least one adjustment question from a preset target knowledge base.

[0032] S105: Input the user question, the explanation information of each of the related professional terms, and the at least one target retrieval result into the trained large language model, and output the target answer information corresponding to the user question.

[0033] The knowledge question-and-answer method provided in the embodiment of the present application can, after determining that a user question input by a user has been received, extract the target professional term in the user question, and combine it with a pre-built professional term tree to obtain related terms and explanations associated with the target professional term, thereby expanding the user question. After matching it with the target knowledge base, a target retrieval result with a higher matching degree and a wider matching range can be obtained, and then the user question, the explanation for the target professional term and the target retrieval result are input into the large language model to output the target answer, which helps to improve the accuracy of the output target answer and the user experience.

[0034] The following describes the exemplary steps of the embodiment of the present application: S101: Obtain a user question input by a user, and extract at least one target professional term contained in the user question.

[0035] In related technologies, users may have the need to consult business, and in the early days, answers may be given through manual customer service. However, in this model, the number of service personnel is limited, and customers need to wait in line for a long time during peak consultation periods, resulting in a poor customer experience. In order to solve the above problems, combined with the rapidly developing artificial intelligence technology, automated question-and-answer methods are used to partially replace manual customer service, guiding customers to use intelligent customer service first, and only switching to manual customer service when the intelligent customer service cannot answer the user's questions, thereby reducing customer waiting time and improving the user experience.

[0036] In an optional implementation, the earliest implementation of the intelligent question-answering system is FAQ (Frequently Asked Questions), which is a domain knowledge base formed by organizing a set of questions and answers of some possible common questions in advance, and publishing them on a web page to provide consulting services to users. When a user asks a question, a keyword matching method is adopted to split the user's question into keywords and then match the keywords with the common questions and the corresponding answers, and return the questions and corresponding answers that successfully hit the keywords in the user's question. The disadvantages of this solution are very obvious, mainly including: the need for manual continuous collection and maintenance of common questions; the expression of the customer's question must literally coincide with the existing expression in the domain knowledge base, otherwise no relevant content can be retrieved. The system does not understand the user's question. The system returns a question and answer pair related to the user's question, and cannot directly generate an answer based on the user's question. The user needs to understand and organize the answer based on the returned information.

[0037] Furthermore, Retrieval-Augmented Generation (RAG) technology can be used. After confirming that the question input by the user is received, relevant knowledge is retrieved in the vertical domain knowledge base based on the user's question. The retrieved relevant knowledge is combined with the user's question to allow the large model to generate answers, which can significantly reduce the complexity of answer setting and the accuracy of the answer.

[0038] Specifically, the workflow of RAG is as follows: the user inputs a question, retrieval is performed based on the user input question, and the retrieved content is input into the generation model (such as GPT-4, LLaMA, DeepSeek-V3) together with the user question to generate the final answer.

[0039] In an optional embodiment, the application scenarios of RAG may include fields such as medicine / law, generating compliance answers by retrieving the latest guidelines or precedents; it may also include fields such as news / finance, and generating analytical summaries by retrieving market data or news reports in real time.

[0040] Exemplarily, in the related art, the mainstream intelligent question-answering system adopts the LLM+knowledge retrieval module architecture, wherein the knowledge retrieval module is used to retrieve the TopK most relevant knowledge blocks from the vertical domain knowledge base based on the user question, and the LLM model generates answers based on the user question and the retrieved relevant knowledge blocks. Specifically, the user inputs the query question (generally plain text input), and the vertical domain knowledge base stores high-frequency question-answering data (QA pairs); here, the vertical domain knowledge base also stores text data; further, it refers to text encoding of user questions and knowledge base knowledge, which is a prerequisite for subsequent knowledge retrieval. Through text encoding, the two texts (user questions and knowledge in the knowledge base) can be converted into two vectors, and then the similarity of the two texts is reflected by calculating the similarity index of the two vectors. Based on the user question, the top k knowledge ranked by similarity are retrieved from the vertical domain knowledge base, and a prompt word template is given when calling the large model. The prompt word template, the top k knowledge ranked by similarity, and the query question entered by the user are input into the large model to obtain the final answer and feedback to the user.

[0041] In an optional implementation, when searching for user questions, the knowledge base generally stores data on professional terms in vertical fields, and users may not accurately describe the corresponding professional terms when asking questions, which leads to a low match between the search results and user questions when searching for knowledge. At the same time, if the professional terms are not explained, the large language model may not be able to understand the user questions and related knowledge, which leads to a low accuracy of the output target answers, affecting the user experience.

[0042] In an embodiment of the present application, after determining that a user question input by a user has been received, the target professional term in the user question can be extracted, and combined with a pre-built professional term tree, related terms and explanations associated with the target professional term can be obtained, thereby expanding the user question. After matching it with the target knowledge base, a target retrieval result with a higher matching degree and a wider matching range can be obtained, and then the user question, the explanation for the target professional term and the target retrieval result can be input into the large language model, and a target answer with higher accuracy can be output.

[0043] In an optional implementation, after determining that a user question input by a user is received, it is necessary to extract professional terms from the user question to determine at least one target professional term included in the user question.

[0044] Specifically, the step of "retrieving at least one target professional term contained in the user question" includes: a1: Input the user question into the trained professional term recognition model, and output the input professional term contained in the user question.

[0045] a2: Calculating the similarity between each candidate professional noun contained in the professional noun tree and the input professional noun.

[0046] a3: Determine at least one candidate professional noun whose similarity to the input professional noun is greater than a first preset similarity threshold as at least one target professional noun included in the user question.

[0047] In an embodiment of the present application, the recognition of target professional terms in user questions is a named entity problem. Therefore, a mature named entity model, such as the BERT model, can be used, combined with a domain annotation dataset (generally requiring more than 10,000 samples) to perform model fine-tuning training to obtain a professional term recognition model, and then the professional term recognition model is used to recognize the input professional terms in the user questions.

[0048] For example, if the question input by the user is "What are the living habits of British shorthair cats", after identification by the professional term recognition model, it can be determined that the input professional term contained in the user question input by the user is "British shorthair cat".

[0049] Furthermore, users may have different ways of expressing different professional terms when entering user questions. For the above example, if the user needs to inquire about the habits of British shorthair cats, when entering the user question, the description of "British shorthair cat" may be "British shorthair", "British shorthair cat", "British shorthair cat", etc. Therefore, after extracting the input professional terms contained in the user question through the professional term recognition model, it is also necessary to calculate the similarity between each candidate professional term contained in the professional term tree and the input professional term, and then determine at least one target professional term contained in the user question.

[0050] In an optional embodiment, when calculating the similarity between the candidate professional nouns in the professional noun tree and the input professional nouns, the candidate professional nouns and the input professional nouns can be converted into noun vectors, and then the similarity between the noun vectors is calculated to determine the similarity between the candidate professional nouns and the input professional nouns.

[0051] Specifically, for each candidate professional noun, the similarity between the candidate professional noun and the input professional noun is determined by the following steps: b1: Input the input professional noun into the trained vector processing model, and output the first noun vector corresponding to the input professional noun.

[0052] b2: For each candidate professional noun, input the candidate professional noun into the vector processing model, and output a second noun vector corresponding to the candidate professional noun.

[0053] b3: Calculate the similarity between the first noun vector and the second noun vector to determine the similarity between the candidate professional noun and the input professional noun.

[0054] In an optional embodiment, the input professional noun and the candidate professional noun can be simultaneously input into a trained vector processing model, and the input professional noun and the candidate professional noun are embedded to obtain a first noun vector corresponding to the input professional noun and a second noun vector corresponding to the candidate professional noun.

[0055] Embedding maps discrete, high-dimensional data into a low-dimensional, continuous vector space. This process not only reduces the data's dimensionality but also preserves its key features and underlying relationships. Embedding vectors abstract and encode features, mapping similar objects to similar locations in the vector space, thereby capturing the data's semantic information and underlying relationships.

[0056] In an optional embodiment, when calculating the similarity between the first noun vector and the second noun vector, the similarity between the first noun vector and the second noun vector can be calculated jointly by a dense vector (for example: BGE) and a sparse vector (for example: BM25), thereby improving the accuracy of vector calculation and matching.

[0057] Here, dense vector calculation may be performed by determining the similarity between vectors through methods such as cosine similarity, and sparse vector calculation may be performed by matching keywords to determine the similarity between vectors.

[0058] Furthermore, after calculating the similarity between the input professional noun and the candidate professional nouns included in the professional noun tree, at least one candidate professional noun whose similarity with the input professional noun is greater than a first preset similarity threshold is determined as at least one target professional noun included in the user question.

[0059] In an optional embodiment, the first preset similarity threshold can be set according to the number of candidate professional nouns contained in the professional noun tree and the noun screening requirements. The specific setting method is not specifically limited here. For example, the first preset similarity threshold can be set to 0.95.

[0060] In another optional embodiment, after calculating the similarity between each candidate professional noun and the input professional noun, the candidate professional nouns can be sorted in descending order of similarity, and the candidate professional nouns ranked before a preset position (for example, ranked in the top three) can be determined as the target professional noun included in the user question.

[0061] Furthermore, after extracting at least one target professional term from the user question, the at least one target professional term can be matched with the candidate professional terms included in the pre-constructed professional terms to determine at least one associated professional term that has an association relationship with the target professional term and the explanatory information for each of the associated professional terms.

[0062] S102: Based on the at least one target professional noun and a pre-constructed professional noun tree, determine at least one associated professional noun associated with the at least one target professional noun and explanation information for each associated professional noun.

[0063] In an embodiment of the present application, the professional term tree is a collection of different professional terms in historical corpus information and is constructed based on the semantic relationship between professional terms. It can provide semantic relationships and corresponding explanatory information between different professional terms, thereby assisting the subsequent knowledge retrieval process and the process of outputting target answers by the large language model. The construction process of the professional term tree will be explained below.

[0064] Specifically, the professional term tree is constructed through the following steps: c1: Input the obtained multiple historical question and answer text information into the pre-trained professional term recognition model and output multiple extracted professional terms.

[0065] c2: For multiple extracted professional nouns, based on the semantic relationships between the extracted professional nouns, multiple professional noun nodes and the superordinate and subordinate relationships between the professional noun nodes are determined.

[0066] c3: constructing the professional noun tree based on multiple professional noun nodes, the superordinate and subordinate relationships between the professional noun nodes, and the explanation information of each professional noun.

[0067] In the embodiments of the present application, as mentioned above, the identification of extracted professional terms is a named entity problem. Therefore, a mature named entity model, such as the BERT model, can be used in combination with a domain annotation dataset (generally requiring more than 10,000 samples) to perform model fine-tuning training to obtain a professional term recognition model, and then the extracted professional terms are identified through the professional term recognition model.

[0068] Here, the process of extracting professional terms can be obtained by extracting professional terms from historical question and answer text information. Before this, manual data annotation is required. After the professional terms are annotated, the professional recognition model is trained to obtain a professional term recognition model that can identify and extract professional terms. Then, the professional terms in the historical question and answer text information are extracted through the trained professional term recognition model.

[0069] Furthermore, after determining multiple extracted professional nouns, it is necessary to determine the semantic relationship between each extracted professional noun, determine multiple professional noun nodes and the superordinate and subordinate relationships between professional noun nodes, and then construct a professional noun tree based on multiple professional noun nodes and the superordinate and subordinate relationships between professional noun nodes and the explanatory information of each professional noun.

[0070] Here, the semantic relationships between the extracted professional nouns may include: synonyms, hypernyms, hyponyms, and appositives. After the professional nouns are extracted and the semantic relationships between nouns are constructed, an abstract placeholder "Object" is used as the final parent node of the entire professional noun tree, and all nouns without hypernyms are connected to Object to complete the construction of the professional noun tree.

[0071] Specifically, in the professional noun tree, the professional noun node corresponding to the superordinate professional noun is the parent node of the professional noun node corresponding to the subordinate professional noun; the professional noun nodes corresponding to professional nouns that are synonyms or appositives are located at the same level in the professional noun tree.

[0072] Specifically, synonyms: different ways of expressing that professional nouns A and B belong to the same concept. For example, British shorthair and British shorthair cat represent the same concept and are synonyms. Hypernym / hyponym: If professional noun A is the hypernym of professional noun B, it means that professional noun A is the parent concept of professional noun B. For example: British shorthair cat belongs to the family cat, so cat family is the hypernym of British shorthair cat. Similarly, British shorthair cat is the hyponym of cat family. Appositive: If professional nouns A and B have a common hypernym C, then A and B are called appositives.

[0073] For example, see Figure 2 , Figure 2 The schematic diagram of the structure of the professional term tree provided in the embodiment of this application is as follows: Figure 2 As shown, the parent node of the entire professional noun tree is "animals", and the subordinates of "animals" include "felines" and "canines"; for the branch of "felines", the hyponyms of "felines" include "British Shorthair Cat", "Chinese Pastoral Cat" and "American Shorthair Cat", and "British Shorthair Cat", "Chinese Pastoral Cat" and "American Shorthair Cat" are appositives to each other and are located at the same level of the professional noun tree, and "British Shorthair Cat" also includes the synonym "British Shorthair"; for the branch of "canines", the hyponyms of "canines" include "Husky" and "Shiba Inu", and "Husky" and "Shiba Inu" are appositives to each other and are located at the same level of the professional noun tree.

[0074] In an optional implementation, multiple professional noun trees may be constructed for professional nouns of different attributes. For example, a professional noun tree for "animals" and a professional noun tree for "home furnishings" may be constructed.

[0075] In an optional implementation, after constructing the professional term tree, explanation information is provided for each professional term in the tree, explaining what the term means. For example, "Chinese rural cat" is a general term for domestic cats native to China, which are divided into multiple breeds based on coat color, such as tabby cats, orange cats, calico cats, white cats, and black cats. Configuring an explanation for each professional term helps the large language model better understand the specific meaning of the term when it is used to generate answers later, thereby ensuring the accuracy and quality of the target answer generation.

[0076] For example, see Figure 3 , Figure 3 The flowchart for constructing the professional term tree provided in the embodiment of this application is as follows: Figure 3As shown, data annotation is performed, and a professional noun recognition model is trained through the annotated data to obtain a professional noun recognition model. Professional nouns are batch extracted from historical question and answer text information, and the professional noun structure is sorted out. After the explanation information of the professional nouns is annotated, a constructed professional noun tree is obtained.

[0077] Furthermore, after determining at least one target professional term and a professional term tree in the user question, the at least one target professional term and the professional terms in the professional term tree may be matched to obtain at least one associated professional term that has an associated relationship with the target professional term.

[0078] Specifically, the step of “determining at least one associated professional term that has an associated relationship with the at least one target professional term based on the at least one target professional term and the pre-constructed professional term tree” includes: d1: normalizing the at least one target professional term to determine at least one standard professional term after the normalization process.

[0079] d2: For each standard professional term, based on multiple professional term paths included in the professional term tree and a preset jump number, determine at least one associated professional term that has an associated relationship with the standard professional term.

[0080] In an optional embodiment, the target professional terms are normalized in order to solve the problem of inaccurate expressions when users ask questions. For example, "British Shorthair cat" is a standard expression and has a synonym "British shorthair". However, customers may use arbitrary expressions in their questions, such as "British shorthair cat". Therefore, it is necessary to "map" the vocabulary used by the user to the standard expression on the professional term tree.

[0081] Specifically, each professional noun on the professional noun tree can be embedded and stored in a vector database. Then, the same model can be used to embed the target professional noun entered by the user. The similarity between the target professional noun entered by the user and all the professional nouns on the professional noun tree is calculated, and samples with high similarity are taken as standard professional nouns corresponding to the target professional noun.

[0082] Here, the similarity between the target professional term input by the user and all professional terms in the professional term tree may be calculated by vector similarity (for example, cosine similarity calculation method, etc.).

[0083] For example, see Figure 4 , Figure 4A schematic diagram of the process of obtaining standard professional terms provided in the embodiment of the present application is provided. Each professional term on the professional term tree ("British Shorthair Cat", "Chinese Rural Cat", "American Shorthair Cat") is embedded to obtain multiple candidate professional term vectors. At the same time, the target professional term "British Shorthair Cat" entered in the user's question is embedded to obtain the target professional term vector. The similarity between the target professional term vector and each candidate professional term vector is calculated to determine the standard professional term "British Shorthair" with a similarity of 0.95.

[0084] For example, a user asks, "What is the lifespan of a British shorthair cat?" The professional noun recognition model identifies the professional noun contained in the sentence as "British shorthair cat." "British shorthair cat" is normalized to obtain the standard professional noun "British shorthair." "British shorthair" is used to search for related professional nouns in the constructed professional noun tree, locating the branch of the professional noun tree to which "British shorthair" belongs.

[0085] Furthermore, after the standard professional term is determined, at least one associated professional term that has an associated relationship with the standard professional term is determined based on a plurality of professional term paths included in the professional term tree and a preset number of jumps.

[0086] Specifically, the step of "for each standard professional term, determining at least one associated professional term that has an associated relationship with the standard professional term based on multiple professional term paths included in the professional term tree and a preset number of jumps" includes: e1: For each standard professional noun, determine at least one target professional noun path containing the standard professional noun from a plurality of professional noun paths contained in the professional noun tree.

[0087] e2: For each target professional noun path, starting from the location of the corresponding standard professional noun, jump on the target professional noun path according to the preset jump number, and determine the candidate professional noun corresponding to the jump end position as the associated professional noun.

[0088] In an embodiment of the present application, the nth-degree nearest neighbor method can be used to determine the associated professional nouns associated with the target professional noun. Specifically, a preset number of jumps can be set according to the professional noun acquisition requirements. For example, while ensuring the accuracy of acquiring the associated professional nouns, the probability of acquiring irrelevant professional nouns is reduced, and the preset number of jumps can be set to 2-4 times.

[0089] Specifically, for each standard professional noun, at least one target professional noun path containing the standard professional noun is determined from the multiple professional noun paths contained in the professional noun tree; for each target professional noun path, starting from the location of the corresponding standard professional noun, jumping on the target professional noun path according to a preset number of jumps, and determining the candidate professional noun corresponding to the jump end position as the associated professional noun.

[0090] For example, see Figure 2 The standard professional term "British Shorthair" is determined, and the preset jump count is set to 3. The target professional term paths obtained are: Target Professional Term Path 1: British Shorthair -> British Shorthair -> Cats; Target Professional Term Path 2: British Shorthair -> British Shorthair -> Chinese Rural Cat; Target Professional Term Path 3: British Shorthair -> British Shorthair -> American Shorthair. The associated professional terms associated with the standard professional term "British Shorthair" are determined to be "Cat Family," "Chinese Rural Cat," and "American Shorthair."

[0091] Furthermore, after matching the target professional noun according to the professional noun tree to obtain at least one related professional noun, the user question is rewritten and adjusted according to the related professional noun to obtain at least one adjusted question.

[0092] S103: Adjust the user question based on the at least one associated professional term to obtain at least one adjusted question.

[0093] Specifically, the step of “adjusting the user question based on the at least one associated professional term to obtain at least one adjusted question” includes: f1: Replace the target professional term contained in the user question with each determined associated professional term to obtain at least one adjusted question.

[0094] In an embodiment of the present application, adjusting the user question according to the associated professional terms is to replace the target professional terms in the user question with the associated professional terms to obtain at least one adjusted question.

[0095] For the above example, the user question input by the user is "How long is the lifespan of a British shorthair cat?". After extraction, it is determined that the target professional noun in the user question is "British shorthair cat". After normalizing the target professional noun, the standard professional noun obtained is "British shorthair". After matching with the professional noun tree, the related professional nouns obtained are "cats", "Chinese rural cats" and "American shorthair cats". The adjusted questions after adjustment based on the related professional nouns are "How long is the lifespan of cats", "How long is the lifespan of Chinese rural cats" and "How long is the lifespan of American shorthair cats".

[0096] Furthermore, after determining at least one adjustment question, based on the at least one adjustment question and the user question input by the user, a match can be performed in the target knowledge base based on the user question and the at least one adjustment question to obtain at least one target retrieval result that matches the user question and / or the at least one adjustment question.

[0097] Specifically, the step of “retrieve, based on the user question and the at least one adjustment question, at least one target retrieval result that matches the user question and / or the at least one adjustment question from a preset target knowledge base” includes: g1: Determine a user question encoding vector corresponding to the user question and an adjustment question encoding vector corresponding to each adjustment question.

[0098] g2: Calculate the similarity between the retrieval result encoding vector contained in the target knowledge base and the user question encoding vector and / or the adjustment question encoding vector in the target knowledge base.

[0099] g3: Determine at least one retrieval result encoding vector whose similarity with the user question encoding vector and / or the adjustment question encoding vector is greater than a second preset similarity threshold as the at least one target retrieval result.

[0100] In an embodiment of the present application, the user question and at least one adjustment question are encoded, and the text is converted into the corresponding user question encoding vector and adjustment question encoding vector, and then the similarity (such as cosine similarity, etc.) is calculated based on the retrieval result encoding vector contained in the target knowledge center to determine at least one target retrieval result that matches the user question encoding vector and / or the adjustment question encoding vector.

[0101] Specifically, after calculating the similarity between the user question encoding vector and / or the adjusted question encoding vector and the retrieval result encoding vector contained in the target knowledge base, at least one retrieval result encoding vector whose similarity is greater than a second preset similarity threshold is determined as at least one target retrieval result.

[0102] In an optional implementation, the second preset similarity threshold may be set according to the number of search result encoding vectors contained in the target knowledge base and the search result screening requirements, and the specific setting method is not specifically limited herein.

[0103] In another optional embodiment, after calculating the similarity between the user question encoding vector and / or the adjusted question encoding vector and the retrieval result encoding vector contained in the target knowledge base, the candidate retrieval results can be sorted in descending order of similarity, and the candidate retrieval results ranked before a preset position (for example, ranked in the top three) can be determined as the target retrieval results.

[0104] Furthermore, after determining the target search result, the user question input by the user, the explanation information of each professional term and at least one target search result can be simultaneously input into the large language model to obtain the target answer information matching the user question.

[0105] S105: Input the user question, the explanation information of each of the related professional terms, and the at least one target retrieval result into the trained large language model, and output the target answer information corresponding to the user question.

[0106] In an optional embodiment, if the model processing capability of the large language model is weak, in order to improve the accuracy of the target answer information obtained, before inputting at least one target retrieval result into the large language model, the similarity between at least one target retrieval result and the user question can be calculated again (cosine vector similarity calculation, etc.), and the retrieval results in the target retrieval results that obviously do not match the user question can be removed, thereby improving the accuracy of the output target answer information.

[0107] In another optional implementation, if the large language model has strong model processing capabilities, there is no need to filter the target search results. The user question input by the user, the explanation information for each professional term, and at least one target search result are directly input into the large language model at the same time to obtain the target answer information that matches the user question. In this way, the data processing steps can be reduced, thereby improving the efficiency of outputting the target answer information.

[0108] In an optional implementation, in order for the large language model to more accurately fit the current conversation scenario, while inputting the user question entered by the user, the explanation information for each professional term, and at least one target retrieval result into the large language model, the designed prompt words can also be input into the large language model, thereby improving the accuracy of the output target answer information.

[0109] For example, the prompt template is as follows: You are an intelligent customer service assistant, skilled at answering user questions based on the provided knowledge. The user's question is as follows: {XXXX}; The meanings of the various professional terms involved in the user's question are as follows: [Professional term 1 and its meaning], [Professional term 2 and its meaning]; Please carefully understand the professional terms and their meanings, and then combine the information provided with the professional terms to understand the reference knowledge provided. The reference knowledge is as follows: {YYYYYY}; Note: Please answer strictly based on the provided reference knowledge and do not diverge. If the provided knowledge cannot answer the user's question, please directly respond with "Sorry, I don't know, please transfer to manual customer service."

[0110] The knowledge question-answering method provided in the embodiment of the present application obtains a user question input by a user and extracts at least one target professional term contained in the user question; based on the at least one target professional term and a pre-constructed professional term tree, determines at least one associated professional term that has an association relationship with the at least one target professional term and explanation information for each professional term; adjusts the user question based on the at least one associated professional term to obtain at least one adjusted question; based on the user question and the at least one adjusted question, retrieves at least one target retrieval result that matches the user question and / or at least one adjusted question from a pre-set target knowledge base; inputs the user question, the explanation information for each professional term and the at least one target retrieval result into a trained large language model, and outputs target answer information corresponding to the user question. In this way, after confirming that the user question input by the user has been received, the target professional terms in the user question can be extracted, and combined with the pre-built professional term tree, the related terms and explanations associated with the target professional terms can be obtained, and then the user question can be expanded. After matching with the target knowledge base, a target retrieval result with a higher matching degree and a wider matching range can be obtained, and then the user question, the explanation for the target professional term and the target retrieval result can be input into the large language model to output the target answer, which helps to improve the accuracy of the output target answer and the user experience.

[0111] Based on the same inventive concept, a knowledge question and answer device corresponding to the knowledge question and answer method is also provided in the embodiment of the present application. Since the principle of solving the problem by the device in the embodiment of the present application is similar to the above-mentioned knowledge question and answer method in the embodiment of the present application, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be repeated.

[0112] See also Figure 5 , Figure 5 This is one of the structural diagrams of a knowledge question-answering device provided in an embodiment of the present application. Figure 6 This is a second structural diagram of a knowledge question answering device provided in an embodiment of the present application. Figure 5 As shown in , the knowledge question answering device 500 includes: The target professional term extraction module 510 is used to obtain a user question input by a user and extract at least one target professional term contained in the user question; The associated professional term determination module 520 is configured to determine, based on the at least one target professional term and a pre-built professional term tree, at least one associated professional term associated with the at least one target professional term and explanation information for each associated professional term; A user question adjustment module 530 is configured to adjust the user question based on the at least one associated professional term to obtain at least one adjusted question; A knowledge base retrieval module 540 is configured to retrieve, based on the user question and the at least one adjustment question, at least one target retrieval result that matches the user question and / or the at least one adjustment question from a preset target knowledge base; The target answer output module 550 is used to input the user question, the explanation information of each related professional term and the at least one target retrieval result into the trained large language model, and output the target answer information corresponding to the user question.

[0113] In an optional embodiment, when the target professional noun extraction module 510 is used to extract at least one target professional noun contained in the user question, the target professional noun extraction module 510 is used to: Input the user question into the trained professional term recognition model, and output the input professional term contained in the user question; Calculating the similarity between each candidate professional noun contained in the professional noun tree and the input professional noun; At least one candidate professional noun whose similarity to the input professional noun is greater than a first preset similarity threshold is determined as at least one target professional noun included in the user question.

[0114] In an optional embodiment, for each candidate professional noun, the target professional noun extraction module 510 is configured to determine the similarity between the candidate professional noun and the input professional noun by performing the following steps: Inputting the input professional noun into the trained vector processing model, and outputting a first noun vector corresponding to the input professional noun; For each candidate professional noun, input the candidate professional noun into the vector processing model, and output a second noun vector corresponding to the candidate professional noun; The similarity between the first noun vector and the second noun vector is calculated to determine the similarity between the candidate professional noun and the input professional noun.

[0115] In an optional embodiment, when the associated professional term determination module 520 is used to determine at least one associated professional term that has an associated relationship with the at least one target professional term based on the at least one target professional term and a pre-constructed professional term tree, the associated professional term determination module 520 is used to: Normalizing the at least one target professional term to determine at least one standard professional term after the normalization process; For each standard professional term, based on a plurality of professional term paths included in the professional term tree and a preset jump number, at least one associated professional term that has an associated relationship with the standard professional term is determined.

[0116] In an optional embodiment, when the associated professional term determination module 520 is used to determine, for each standard professional term, at least one associated professional term that has an associated relationship with the standard professional term based on multiple professional term paths included in the professional term tree and a preset jump number, the associated professional term determination module 520 is used to: For each standard professional noun, determining at least one target professional noun path containing the standard professional noun from a plurality of professional noun paths contained in the professional noun tree; For each target professional noun path, starting from the location of the corresponding standard professional noun, jumping on the target professional noun path according to the preset jump number, and determining the candidate professional noun corresponding to the jump end position as the associated professional noun.

[0117] In an optional embodiment, when the user question adjustment module 530 is used to adjust the user question based on the at least one associated professional term to obtain at least one adjusted question, the user question adjustment module 530 is used to: The target professional term contained in the user question is replaced by each determined associated professional term to obtain at least one adjusted question.

[0118] In an optional embodiment, when the knowledge base retrieval module 540 is used to retrieve at least one target retrieval result matching the user question and / or the at least one adjustment question from a preset target knowledge base based on the user question and the at least one adjustment question, the knowledge base retrieval module 540 is used to: Determining a user question encoding vector corresponding to the user question and an adjustment question encoding vector corresponding to each adjustment question; Calculating the similarity between the retrieval result encoding vector contained in the target knowledge base and the user question encoding vector and / or the adjustment question encoding vector in the target knowledge base; At least one retrieval result encoding vector having a similarity with the user question encoding vector and / or the adjustment question encoding vector greater than a second preset similarity threshold is determined as the at least one target retrieval result.

[0119] In an optional embodiment, as Figure 6 As shown, the knowledge question answering device 500 further includes a professional noun tree construction module 560, which is used to construct the professional noun tree through the following steps: Input the acquired multiple historical question and answer text information into a pre-trained professional term recognition model, and output multiple extracted professional terms; For multiple extracted professional nouns, based on the semantic relationship between the extracted professional nouns, multiple professional noun nodes and the superordinate and subordinate relationships between the professional noun nodes are determined; The professional noun tree is constructed based on a plurality of professional noun nodes, the superordinate and subordinate relationships between the professional noun nodes, and the explanation information of each professional noun.

[0120] In an optional embodiment, in the professional noun tree, the professional noun node corresponding to the superordinate professional noun is the parent node of the professional noun node corresponding to the subordinate professional noun; the professional noun nodes corresponding to professional nouns that are synonyms or appositives are located at the same level in the professional noun tree.

[0121] The knowledge question and answer device provided in the embodiment of the present application obtains a user question input by a user and extracts at least one target professional term contained in the user question; based on the at least one target professional term and a pre-constructed professional term tree, determines at least one associated professional term that has an association relationship with the at least one target professional term and explanation information for each professional term; adjusts the user question based on the at least one associated professional term to obtain at least one adjusted question; based on the user question and the at least one adjusted question, retrieves at least one target retrieval result that matches the user question and / or at least one adjusted question from a pre-set target knowledge base; inputs the user question, the explanation information for each professional term and the at least one target retrieval result into a trained large language model, and outputs target answer information corresponding to the user question. In this way, after confirming that the user question input by the user has been received, the target professional terms in the user question can be extracted, and combined with the pre-built professional term tree, the related terms and explanations associated with the target professional terms can be obtained, and then the user question can be expanded. After matching with the target knowledge base, a target retrieval result with a higher matching degree and a wider matching range can be obtained, and then the user question, the explanation for the target professional term and the target retrieval result can be input into the large language model to output the target answer, which helps to improve the accuracy of the output target answer and the user experience.

[0122] See also Figure 7 , Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. Figure 7 As shown in FIG, the electronic device 700 includes a processor 710 , a memory 720 and a bus 730 .

[0123] The memory 720 stores machine-readable instructions executable by the processor 710. When the electronic device 700 is running, the processor 710 communicates with the memory 720 via the bus 730. When the machine-readable instructions are executed by the processor 710, the above-mentioned Figure 1 The steps of the knowledge question answering method in the illustrated method embodiment and the specific implementation thereof can be found in the method embodiment and will not be described in detail here.

[0124] The embodiment of the present application also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the computer program can execute the above-mentioned Figure 1 The steps of the knowledge question answering method in the illustrated method embodiment and the specific implementation thereof can be found in the method embodiment and will not be described in detail here.

[0125] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0126] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. There may be other division methods in actual implementation. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed may be through some communication interface, indirect coupling or communication connection of devices or units, which may be electrical, mechanical or other forms.

[0127] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0128] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0129] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium that is executable by a processor. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes various media that can store program code, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0130] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present application, which are used to illustrate the technical solutions of the present application, rather than to limit them. The scope of protection of the present application is not limited thereto. Although the present application has been described in detail with reference to the above-mentioned embodiments, those skilled in the art should understand that any person skilled in the art can modify or easily conceive of changes to the technical solutions described in the above-mentioned embodiments within the technical scope disclosed in the present application, or perform equivalent replacements for some of the technical features thereof. These modifications, changes, or replacements do not deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A knowledge question answering method, characterized in that: Applied to a knowledge question answering system, the knowledge question answering method includes: Obtaining a user question input by a user, and extracting at least one target professional term contained in the user question; Based on the at least one target professional noun and a pre-constructed professional noun tree, determining at least one associated professional noun associated with the at least one target professional noun and explanation information for each associated professional noun; Adjusting the user question based on the at least one associated professional term to obtain at least one adjusted question; Based on the user question and the at least one adjustment question, at least one target retrieval result matching the user question and / or the at least one adjustment question is retrieved from a preset target knowledge base; The user question, the explanation information of each of the related professional terms and the at least one target retrieval result are input into a trained large language model, and target answer information corresponding to the user question is output.

2. The knowledge question answering method according to claim 1, characterized in that: The extracting of at least one target professional term contained in the user question includes: Input the user question into the trained professional term recognition model, and output the input professional term contained in the user question; Calculating the similarity between each candidate professional noun contained in the professional noun tree and the input professional noun; At least one candidate professional noun whose similarity to the input professional noun is greater than a first preset similarity threshold is determined as at least one target professional noun included in the user question.

3. The knowledge question answering method according to claim 2, characterized in that: For each candidate professional noun, the similarity between the candidate professional noun and the input professional noun is determined by the following steps: Inputting the input professional noun into the trained vector processing model, and outputting a first noun vector corresponding to the input professional noun; For each candidate professional noun, input the candidate professional noun into the vector processing model, and output a second noun vector corresponding to the candidate professional noun; The similarity between the first noun vector and the second noun vector is calculated to determine the similarity between the candidate professional noun and the input professional noun.

4. The knowledge question answering method according to claim 1, wherein: The determining, based on the at least one target professional noun and the pre-constructed professional noun tree, at least one associated professional noun that has an associated relationship with the at least one target professional noun includes: Normalizing the at least one target professional term to determine at least one standard professional term after the normalization process; For each standard professional term, based on a plurality of professional term paths included in the professional term tree and a preset jump number, at least one associated professional term that has an associated relationship with the standard professional term is determined.

5. The knowledge question answering method according to claim 4, characterized in that: The method of determining, for each standard professional term, at least one associated professional term associated with the standard professional term based on multiple professional term paths and a preset jump number included in the professional term tree, includes: For each standard professional noun, determining at least one target professional noun path containing the standard professional noun from a plurality of professional noun paths contained in the professional noun tree; For each target professional noun path, starting from the location of the corresponding standard professional noun, jumping on the target professional noun path according to the preset jump number, and determining the candidate professional noun corresponding to the jump end position as the associated professional noun.

6. The knowledge question answering method according to claim 1, characterized in that: The step of adjusting the user question based on the at least one associated professional term to obtain at least one adjusted question includes: The target professional term contained in the user question is replaced by each determined associated professional term to obtain at least one adjusted question.

7. The knowledge question answering method according to claim 1, characterized in that: The step of retrieving at least one target retrieval result matching the user question and / or the at least one adjustment question from a preset target knowledge base based on the user question and the at least one adjustment question includes: Determining a user question encoding vector corresponding to the user question and an adjustment question encoding vector corresponding to each adjustment question; Calculating the similarity between the retrieval result encoding vector contained in the target knowledge base and the user question encoding vector and / or the adjustment question encoding vector in the target knowledge base; At least one retrieval result encoding vector having a similarity with the user question encoding vector and / or the adjustment question encoding vector greater than a second preset similarity threshold is determined as the at least one target retrieval result.

8. The knowledge question answering method according to claim 1, characterized in that: The professional term tree is constructed by the following steps: Input the acquired multiple historical question and answer text information into a pre-trained professional term recognition model, and output multiple extracted professional terms; For multiple extracted professional nouns, based on the semantic relationship between the extracted professional nouns, multiple professional noun nodes and the superordinate and subordinate relationships between the professional noun nodes are determined; The professional noun tree is constructed based on a plurality of professional noun nodes, the superordinate and subordinate relationships between the professional noun nodes, and the explanation information of each professional noun.

9. The knowledge question answering method according to claim 8, characterized in that: In the professional noun tree, the professional noun node corresponding to the superordinate professional noun is the parent node of the professional noun node corresponding to the subordinate professional noun; the professional noun nodes corresponding to professional nouns that are synonyms or appositives are located at the same level in the professional noun tree.

10. A knowledge question-answering device, characterized in that: Applied to a knowledge question answering system, the knowledge question answering method device includes: A target professional term extraction module is used to obtain a user question input by a user and extract at least one target professional term contained in the user question; An associated professional term determination module is configured to determine, based on the at least one target professional term and a pre-constructed professional term tree, at least one associated professional term associated with the at least one target professional term and explanation information for each associated professional term; A user question adjustment module, configured to adjust the user question based on the at least one associated professional term to obtain at least one adjusted question; a knowledge base retrieval module, configured to retrieve, based on the user question and the at least one adjustment question, at least one target retrieval result that matches the user question and / or the at least one adjustment question from a preset target knowledge base; The target answer output module is used to input the user question, the explanation information of each related professional term and the at least one target retrieval result into the trained large language model, and output the target answer information corresponding to the user question.

11. An electronic device, characterized in that: include: A processor, a storage medium and a bus, wherein the storage medium stores machine-readable instructions executable by the processor. When the electronic device is running, the processor and the storage medium communicate through the bus, and the processor executes the machine-readable instructions to perform the steps of the knowledge question and answer method as described in any one of claims 1 to 9.

12. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which, when executed by a processor, executes the steps of the knowledge question-answering method according to any one of claims 1 to 9.

Images