Intelligent Q&A Method and System for Electric Power Audit Knowledge Based on Big Data AI

Abstract

This application belongs to the technical field of natural language models, and provides an intelligent power audit knowledge Q&A method and system based on big data AI. For different users' different Q&A habits, writing styles, Q&A purposes, etc., the satisfaction of users with the responses is identified through a feedback recognition model, and then a one-way three-level Q&A network is constructed through a pre-extraction model, an entity generation model, and a relationship generation model, which is beneficial to flexibly adjust the occupancy of computing resources by a single Q&A session and improve the accuracy and efficiency of Q&A.

Description

Technical Field

[0001] This application relates to the technical field of natural language models, and particularly to an intelligent question-answering method and system for power audit knowledge based on big data AI. Background Technique

[0002] The statements in this part merely provide background technical information related to this application and do not necessarily constitute prior art.

[0003] The original compilation of audit reports, question answering, and rectification supervision rely on manual labor, lacking a vertical training solution for specific audit fields. Based on the current situation, the company has developed and constructed a power audit knowledge base and knowledge graph that combines the "regulation library, question library, case library, and report library" using natural language models. By adopting advanced semantic analysis techniques and innovative cross-attribute knowledge fusion and expression techniques, the transformation of power audit knowledge from fragmentation to systematization and from implicit to explicit has been realized, constructing a knowledge-intensive and structured power audit intelligent knowledge graph, and providing a question-answering system based on natural language models.

[0004] In the actual application process of the knowledge graph, it is found that due to different question-and-answer writing habits, question-and-answer purposes, and the ability to conceive and summarize of different users, as well as the adaptability to large language models and the distrust of intelligent fusion answers of the knowledge graph, the question language is overly redundant or the semantic meaning of the question deviates from the actual question asked.

[0005] Typically, a large amount of text in a file is directly copied into the question-answering system, which contains a lot of invalid information, and then a concise question is put forward at the end. The question-answering system parses and semantically understands all the question data uniformly, resulting in a waste of computing resources, and also easily leading to a deviation in the semantic meaning of the question, reducing the accuracy and efficiency of question answering. Summary of the Invention

[0006] To solve the problems in the background technique, this application proposes an intelligent question-answering method and system for power audit knowledge based on big data AI.

[0007] This application provides an intelligent question-answering method for power audit knowledge based on big data AI, including the following steps:

[0008] Step 1: Establish a question-and-answer session and start receiving natural language question data;

[0009] Step 2: Use a pre-extraction model to intercept the core question sentence in the latest question data, extract entities and relationships according to the core question sentence, generate a pre-retrieval triple sequence, query the knowledge graph based on the pre-retrieval triple sequence, and output a reply;

[0010] Step 3: Use the feedback recognition model to perform satisfaction semantic recognition on the latest input reply feedback. If the recognized semantics is positive satisfaction, jump to Step 2; if it is recognized as negative satisfaction, jump to Step 4;

[0011] Step 4: Use the entity generation model to generate entity candidates from the latest question data, extract the question semantics based on the relationship generation model to generate relationship candidates, and use the entity candidates and relationship candidates to query the knowledge graph to obtain and output the reply triples;

[0012] Step 5: Use the feedback recognition model to perform satisfaction on the latest input reply feedback, associate the question data in the above text with the latest question data according to the recognition result, and loop to jump to Step 4 until the conversation ends.

[0013] Preferably, in the said Step 2, the specific method for the pre-extraction model to extract entities and relationships in the question data is:

[0014] Preprocess the question data, and extract the smallest natural language paragraph L located at the last paragraph and conforming to the question pattern;

[0015] Extract the entity naming nouns in L, convert the entity naming nouns into word vectors, use the word vectors to obtain semantic features, and obtain entities based on the semantic features by associating with the knowledge graph;

[0016] Retrieve the entity types to obtain a list of common relationships, extract the relationship feature words in L and perform semantic matching with the elements in the list of common relationships, and use the element with the highest matching value as the target relationship;

[0017] Query the knowledge graph based on the obtained entities and target relationships to obtain the target entities, and then obtain the reply triples;

[0018] The pre-extraction model is a composite model composed of a variant neural network combined with an LSTM and a Linear layer plus a RoBERTa-large semantic recognition model.

[0019] Preferably, the feedback recognition model is a composite model composed of a variant neural network combined with an LSTM and a Linear layer plus a RoBERTa-large semantic recognition model.

[0020] Preferably, in the said Step 3, the specific method for the feedback recognition model to perform satisfaction semantic recognition on the reply feedback is:

[0021] Perform word segmentation on the question data and reply feedback to obtain a question word segmentation list, a feedback word segmentation list, and a feedback feature word list. Convert each word segmentation list into word vectors, use the word vectors to obtain semantic features, match the semantic features of the feedback feature words with a preset satisfaction evaluation feature word library to obtain the highest matching satisfaction evaluation, evaluate the correlation between the semantic features of the question word segmentation and the semantic features of the feedback word segmentation, and judge the satisfaction based on the satisfaction evaluation and the correlation evaluation.

[0022] Preferably, in step 5, when the reply feedback satisfaction semantic recognition of the feedback recognition model for the latest input is negative satisfaction, perform correlation recognition on the semantic features of the question data in the above text and the semantic features of the latest question, add the word segmentation of the associated previous question to the word segmentation list of the latest question and perform deduplication processing.

[0023] Preferably, in step 4, the entity generation model is a GRU-CRF model, and the entity extraction method is as follows:

[0024] Perform word segmentation on the natural language question to obtain a word sequence, perform part-of-speech tagging on each word segmentation to obtain a part-of-speech sequence;

[0025] Convert the part-of-speech sequence into word vectors, and input the word vectors into the tagging model to obtain the optimal tagging sequence;

[0026] Link the optimal tagging solution to the knowledge graph to obtain entity candidates.

[0027] Preferably, in step 4, the relationship generation model is a variant neural network of LSTM with an attention mechanism, and the relationship extraction method is as follows:

[0028] Preprocess the natural language question, exclude entity proper nouns or use placeholders to replace them;

[0029] Input the preprocessed question into the LSTM decoder to obtain a relationship list and relationship scores, and select the one with the highest score as the candidate relationship.

[0030] Preferably, input the original question and the preprocessed question into the LSTM decoder to obtain relationship lists L1 and L2 respectively, sort L1 and L2 based on the relationship scores, and select the element with the highest ranking in L1 and matching the elements in L2 as the candidate relationship.

[0031] This application also provides a power audit knowledge intelligent question-answering system based on big data AI, including: a session module, a pre-extraction model, a feedback recognition model, an entity generation model, and a relationship generation model. The functions of each module are as follows:

[0032] Session module: Establish and end a session, receive user natural language input during the session, and output a reply;

[0033] Pre-extraction model: intercepts the last paragraph of the question in the question data, performs lightweight entity and relationship extraction, and quickly generates answers;

[0034] Feedback recognition model: Identify the user's satisfaction with the reply, switch the extraction model, and form a unidirectional three-level recognition network;

[0035] Entity generation model and relationship generation model: Process the latest question data as a whole to extract entities and relationships, and query the knowledge graph to obtain answers.

[0036] Compared with the prior art, the beneficial effects of this application are:

[0037] This application targets different users' different question-and-answer habits, writing styles, and question-and-answer purposes, and uses a feedback recognition model to identify users' satisfaction with responses. It then constructs a unidirectional three-level question-and-answer network through a pre-extraction model, an entity generation model, and a relationship generation model, which is conducive to flexibly adjusting the computing resources occupied by a single question-and-answer session and improving the accuracy and efficiency of question-and-answering. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The drawings in the specification, which constitute a part of the present application, are used to provide further understanding of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute improper limitations on the present application.

[0039] Figure 1 This is a schematic diagram of the system composition of this application.

[0040] Figure 2 Schematic diagram of the method flow of this application. DETAILED DESCRIPTION

[0041] The present application is further described below in conjunction with the accompanying drawings and embodiments.

[0042] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present disclosure. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0043] In the present disclosure, terms such as "upper", "lower", "left", "right", "front", "back", "vertical", "horizontal", "side", "bottom" and the like indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings. They are relational words determined only for the convenience of describing the structural relationships of the various parts or elements of the present disclosure, and do not specifically refer to any part or element in the present disclosure and should not be understood as limitations on the present disclosure.

[0044] Example 1

[0045] As Figures 1 to 2 shown, this application provides an intelligent Q&A method for power audit knowledge based on big data AI, including the following steps:

[0046] Step 1: Establish a Q&A session and start receiving natural language question data;

[0047] Step 2: Use a pre-extraction model to intercept the core question sentence in the latest question data, extract entities and relationships according to the core question sentence, generate a pre-retrieval triple sequence, query the knowledge graph based on the pre-retrieval triple sequence, and output a reply;

[0048] Step 3: Use a feedback recognition model to perform satisfaction semantic recognition on the latest input reply feedback. If the recognized semantics is positive satisfaction, jump to Step 2; if it is recognized as negative satisfaction, jump to Step 4;

[0049] Step 4: Use an entity generation model to generate entity candidates from the latest question data, extract the semantic meaning of the question sentence based on the relationship generation model to generate relationship candidates, use the entity candidates and relationship candidates to query the knowledge graph, obtain the reply triple and output it;

[0050] Step 5: Use a feedback recognition model to perform satisfaction on the latest input reply feedback, and associate the question data in the above text with the latest question data according to the recognition result, and loop to jump to Step 4 until the session ends.

[0051] The session is a chat window, which is actively established by the user or automatically established after inputting initial Q&A data, and the session ends by the user's active operation.

[0052] This application uses a feedback recognition model to identify the user's reply feedback. Based on the pre-extraction model, entity generation model, and relationship generation model, the semantic parsing operation of the question sentence is divided into a one-way three-level parsing network. First, use the pre-extraction model to perform lightweight entity and relationship extraction on the core question sentence in the question data. If it is recognized that the user is satisfied with the reply, the pre-extraction model is always used for entity and relationship extraction. When it is recognized that the user is not satisfied with the reply, the entity generation model and relationship generation model are used to perform entity and relationship extraction on the latest question as a whole, and the operation is maintained until the feedback recognition model recognizes that the user is not satisfied with a certain round of reply, then the context is associated with the latest question data and the entity generation model and relationship generation model are used to perform entity and relationship extraction, and it is maintained until the session ends. The improvement of the "complexity" of the question data is one-way irreversible in each session, which is beneficial to flexibly adjust the occupation of computing resources in a single session and improve the accuracy and efficiency of Q&A.

[0053] Specifically, in step 2, the pre-extraction model is a composite model composed of a variant neural network combining LSTM and Linear layers and a RoBERTa-large semantic recognition model. The specific method for extracting entities and relationships in the question data is as follows:

[0054] First, preprocess the question data, extract the smallest natural language paragraph L at the last paragraph and conforming to the question sentence pattern, train the neural network to recognize and extract the ability of typical question sentence formats such as "why", "what", "how", etc., intercept the smallest natural language paragraph at the last paragraph and conforming to the question sentence pattern in the latest question data, and consider it as the core question sentence.

[0055] Extract the entity naming words in the question sentence L to generate a segmented word sequence S-L = (s 1 , s 2 ,..., s n ), where s n is a word. Perform word embedding and vector transformation on S-L to obtain an embedded vector E-L = (e 1 , e 2 ,..., e m ), where m is the maximum value of the element number. Input the vector E-L into the language model RoBERTa-large to extract semantic features, input the semantic features into the LSTM and Linear layers for BIO sequence annotation, and then obtain the entity and the position of the entity in the question sentence.

[0056] Obtain the type of the entity according to the preset entity type table, retrieve the entity type to obtain the common relationship list, extract the relationship feature words in L and perform semantic matching with the elements in the common relationship list, and take the element with the highest matching value as the target relationship. Query the knowledge graph based on the obtained entity and target relationship to obtain the target entity, and then obtain the reply triple.

[0057] As an optional technical solution, due to the complexity of Chinese, to avoid the situation where the relationship feature words are fuzzy or missing or the common relationship list is not entered in special cases, use the above method of extracting entities to extract relationships as an alternative or parallel technical solution, as follows:

[0058] Convert the question sentence into a vector, extract the semantic features and then input them into the neural network model for BIO sequence annotation, and then obtain the relationship and its position in the question sentence.

[0059] The LSTM model is a long short-term memory network, which is a special type of recurrent neural network (RNN). It is specifically designed to address the vanishing gradient or exploding gradient problems encountered by traditional RNNs when processing long sequence data. Through its unique "gate" mechanism, LSTM can effectively learn long-term dependency information and performs excellently in various tasks involving sequence data, such as natural language processing, speech recognition, and time series prediction, etc.; The Linear layer is a fully connected layer in a neural network; RoBERTa-large is a large pre-trained language model based on the Transformer architecture and belongs to the large-scale model in the RoBERTa (Robustly Optimized BERT approach) model series, which is widely used in semantic feature extraction and entity extraction.

[0060] Specifically, the feedback recognition model is a composite model composed of a variant neural network combined by LSTM and the Linear layer plus the RoBERTa-large semantic recognition model. In step 3, the specific method for performing satisfaction semantic recognition on the reply feedback is as follows:

[0061] Perform word segmentation on the question data and the reply feedback to obtain a question word segmentation list, a feedback word segmentation list, and a feedback feature word list. Convert each word segmentation list into word vectors, use the word vectors to obtain semantic features, match the semantic features of the feedback feature words with a preset satisfaction evaluation feature word library to obtain the satisfaction evaluation with the highest matching value. At the same time, evaluate the correlation between the semantic features of the question word segmentation and the semantic features of the feedback word segmentation, and judge the satisfaction based on the satisfaction evaluation and the correlation evaluation.

[0062] The method for obtaining semantic features is the same as that in the above entity extraction method and will not be elaborated here. The key points in the satisfaction recognition method are two aspects. One is to train the neural network for the recognition accuracy of feedback feature words. The feedback feature words typically include negative words, such as phrases like "wrong", "incorrect", "not", etc. However, in the actual intelligent question-answering system, after the user fails to obtain the desired answer, they may not directly express feedback but modify the writing style of the question or add more conditions to obtain the accurate answer. Therefore, recognizing the repeatability (the above-mentioned correlation) between the user's two questions can also obtain the satisfaction with the previous question. When training the neural network model, the matching degree of semantic features can be quantified through preset conditions, and then the quantified satisfaction evaluation and correlation evaluation can be obtained.

[0063] Specifically, in step 4, the entity generation model is a GRU-CRF model, and the method for extracting entities is as follows:

[0064] Perform word segmentation on natural language questions to obtain a word sequence, perform part-of-speech tagging on each word segment to obtain a part-of-speech sequence, convert the part-of-speech sequence into a word vector, input the word vector into the tagging model to obtain the optimal tagging sequence, and link the optimal tagging solution to the knowledge graph to obtain entity candidates.

[0065] The GRU-CRF model is a recurrent neural network that integrates conditional random fields. It combines the advantages of the gated recurrent unit (GRU) and the conditional random field (CRF). Compared with LTSM, its model complexity is lower, the training speed is faster, and the ability to connect context is stronger. More importantly, the use of different models to extract entities and relationships is conducive to improving recognition accuracy and preventing model confusion.

[0066] After obtaining a question, the GRU layer performs word segmentation on the question. Preferably, part-of-speech tagging (noun, verb, adjective) can be performed to generate word vectors, which are then transferred to the CRF layer transfer probability matrix to generate a possible set of sequence labels and their corresponding probabilities. The optimal label sequence is selected as the entity labeling result. In the process, the BIO (head, middle, outer) mode is used for sequence labeling.

[0067] For example, the score of each annotation result is as follows:

[0068]

[0069] Among them, X indicates a question, X=(x 1 ,x 2 ,... ,x n ), x is the word in the question, y=(y 1 ,y 2 ,...,y m ) represents the labeling result, P is the state feature matrix of the conditional random field, A is the state transfer matrix, Ai,j indicates the score of transferring from the i-th label to the j-th label, for all possible labeling result sets Y X , use the softmax regression function to get the probability of each annotation:

[0070]

[0071] The training goal of the model is to maximize the log-conditional probability of the correct label sequence:

[0072]

[0073] In practice, the Viterbi algorithm is used to solve the optimal labeling sequence.

[0074] Specifically, in step 4, the relationship generation model is an LSTM variant neural network with an attention mechanism, and the relationship extraction method is specifically as follows:

[0075] Preprocess the natural language question, excluding entity nouns or using placeholders to replace them;

[0076] Input the preprocessed question into the LSTM decoder, obtain the relationship list and relationship scores, and select the one with the highest score as the candidate relationship.

[0077] The relationship generation model consists of a two-layer LSTM encoder and an LSTM decoder with an attention mechanism. Given a question, after being encoded by the encoder (including word segmentation, vector processing, etc.), the decoder can directly obtain the relationship set and relationship corresponding probabilities (scores) corresponding to the question. In this process, the semantic contribution of entity words is not significant, and they are replaced with placeholders. The attention mechanism performs weight-based screening on the context, enabling the neural network to more accurately find information related to the current output and improving the accuracy of the output.

[0078] As an alternative technical solution, considering the redundant negative contribution of entity words and the possible semantic skew after removing entity words in the reverse direction, input the original question and the preprocessed question into the LSTM decoder, respectively obtain the relationship lists L1 and L2, sort L1 and L2 based on the relationship scores, and select the element with the highest ranking in L1 and matching the elements in L2 as the candidate relationship. Matching the elements in L2 means that the corresponding relationship exists in the L2 list.

[0079] Preferably, in step 5, when the feedback recognition model recognizes the semantic recognition of the satisfaction of the latest input reply as negative satisfaction, perform an association recognition on the semantic features of the previous question data and the semantic features of the latest question, add the word segmentation of the associated previous question to the word segmentation list of the latest question and perform deduplication processing.

[0080] Train the feedback recognition model to recognize the semantic feature correlation degree. When the answer is inaccurate, add the word segmentation data of the previous text to the latest question. In the continuous extraction of entities and relationships, retain the recognized semantic features to facilitate the association matching with the semantic features of the latest question.

[0081] Embodiment 2

[0082] Based on Embodiment 1, the present application also provides an intelligent power audit knowledge Q&A system based on big data AI, including: a session module, a pre-extraction model, a feedback recognition model, an entity generation model, and a relationship generation model. The functions of each module are as follows:

[0083] Session module: Establish and end a session, receive user natural language input during the session, and output a reply;

[0084] Pre-extraction model: Intercept the last paragraph of the question in the question data, perform lightweight entity and relationship extraction, and quickly generate a reply;

[0085] Feedback recognition model: recognize the user's satisfaction with the reply, switch the extraction model, and form a one-way three-level recognition network;

[0086] Entity generation model, relationship generation model: overall process the latest question data to extract entities and relationships, and query the knowledge graph to obtain answers.

[0087] The above are only the preferred embodiments of the present application and are not used to limit the present application. For those skilled in the art, the present application can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

[0088] Although the specific implementation manners of the present application have been described above in conjunction with the accompanying drawings, they do not limit the protection scope of the present application. Those skilled in the art should understand that on the basis of the technical solutions of the present application, various modifications or deformations that can be made by those skilled in the art without creative efforts are still within the protection scope of the present application.

Claims

1. The intelligent question-answering method of power audit knowledge based on big data AI is characterized by: The steps include: Step 1: Establish a question-answering session and start receiving natural language question data; Step 2: Use the pre-extraction model to extract core questions from the latest question data, extract entities and relationships based on the core questions, generate pre-retrieval triple sequences, query the knowledge graph based on the pre-retrieval triple sequences, and output the answers; Step 3: Use the feedback recognition model to perform satisfaction semantic recognition on the latest input reply feedback. If the recognition semantics is positive satisfaction, jump to step 2; if the recognition semantics is negative satisfaction, jump to step 4; Step 4: Use the entity generation model to generate entity candidates from the latest question data, extract the question semantics based on the relationship generation model, generate relationship candidates, use the entity candidates and relationship candidates to query the knowledge graph, obtain the answer triples and output them; Step 5: Use the feedback recognition model to evaluate the satisfaction of the latest input response feedback, associate the question data in the previous text with the latest question data based on the recognition result, and loop back to step 4 until the session ends; In step 2, the specific method of extracting entities and relationships in the question data by the pre-extraction model is: The question data is preprocessed to extract the smallest natural language paragraph L that is at the end and conforms to the question pattern, extract the entity naming words in L, convert the entity naming words into word vectors, use the word vectors to obtain semantic features, associate the knowledge graph based on the semantic features to obtain entities, retrieve the entity type to obtain a list of common relations, extract the relationship feature words in L and perform semantic matching with the elements in the list of common relations, take the element with the highest matching value as the target relation, query the knowledge graph based on the obtained entities and target relations to obtain the target entities, and then obtain the answer triples.

2. The intelligent question-answering method for power audit knowledge based on big data AI according to claim 1 is characterized by: The pre-extraction model is a composite model consisting of a variant neural network combining LSTM and Linear layers and a RoBERTa-large semantic recognition model.

3. The intelligent question-answering method for power audit knowledge based on big data AI according to claim 1 is characterized by: The feedback recognition model is a composite model consisting of a variant neural network combining LSTM and Linear layers and a RoBERTa-large semantic recognition model.

4. The intelligent question-answering method for power audit knowledge based on big data AI according to claim 3 is characterized by: In step 3, the specific method of the feedback recognition model to perform satisfaction semantic recognition on the reply feedback is: Perform word segmentation on question data and answer feedback to obtain a question word segmentation list, a feedback word segmentation list, and a feedback feature word list. Convert each word segmentation list into a word vector, use the word vector to obtain semantic features, match the preset satisfaction evaluation feature word library based on the semantic features of the feedback feature words, obtain the satisfaction evaluation with the highest matching value, evaluate the correlation between the semantic features of the question word segmentation and the semantic features of the feedback word segmentation, and judge the satisfaction based on the satisfaction evaluation and the correlation evaluation.

5. The intelligent question-answering method for power audit knowledge based on big data AI according to claim 4 is characterized by: In step 5, when the feedback recognition model semantically recognizes the latest input reply feedback satisfaction as negative satisfaction, the semantic features of the question data in the previous text are associated with the semantic features of the latest question, and the word segmentation of the associated previous question is added to the word segmentation list of the latest question and deduplication is performed.

6. The intelligent question-answering method for power audit knowledge based on big data AI according to claim 1 is characterized by: In step 4, the entity generation model is a GRU-CRF model, and the entity extraction method is specifically as follows: Perform word segmentation on natural language questions to obtain word sequences, and perform part-of-speech tagging on each word to obtain a part-of-speech sequence; Convert the part-of-speech sequence into word vectors, and input the word vectors into the annotation model to obtain the optimal annotation sequence; Link the labeled optimal solution to the knowledge graph to obtain entity candidates.

7. The intelligent question-answering method for power audit knowledge based on big data AI according to claim 6 is characterized by: In step 4, the relationship generation model is an LSTM variant neural network with an attention mechanism, and the relationship extraction method is specifically as follows: Preprocess natural language questions to exclude entity names or replace them with placeholders; The preprocessed question is input into the LSTM decoder to obtain the relationship list and relationship score, and the highest score is selected as the candidate relationship.

8. The intelligent question-answering method for power audit knowledge based on big data AI according to claim 7 is characterized by: The original question and the preprocessed question are input into the LSTM decoder to obtain the relationship lists L1 and L2 respectively. L1 and L2 are sorted based on the relationship scores, and the element with the highest ranking in L1 and matching the element in L2 is selected as the candidate relationship.

9. The power audit knowledge intelligent question-answering system based on big data AI is characterized by: include: Conversation module, pre-extraction model, feedback recognition model, entity generation model, relationship generation model, the functions of each module are as follows: Conversation module: establishes and ends conversations, receives user natural language input in conversations, and outputs responses; Pre-extraction model: intercepts the core questions in the question data, performs lightweight entity and relationship extraction, and quickly generates answers; Feedback recognition model: Identify the user's satisfaction with the reply, switch the extraction model, and form a unidirectional three-level recognition network; Entity generation model and relationship generation model: Process the latest question data as a whole to extract entities and relationships, and query the knowledge graph to obtain answers; The method of using the pre-extraction model to extract lightweight entities and relations is as follows: pre-process the question data, extract the smallest natural language paragraph L that is at the end and conforms to the question pattern, consider it as the core question, extract the entity naming words in L, convert the entity naming words into word vectors, use the word vectors to obtain semantic features, associate the knowledge graph based on the semantic features to obtain entities, retrieve the entity type to obtain a list of common relations, extract the relation feature words in L and perform semantic matching with the elements in the list of common relations, use the element with the highest matching value as the target relation, query the knowledge graph based on the obtained entities and target relations to obtain the target entity, and then obtain the answer triple; The feedback recognition model identifies the user's satisfaction with the answer, and the method of forming a unidirectional three-level recognition network is as follows: first, the pre-extraction model is used to perform lightweight entity and relationship extraction on the core questions in the question data. If it is recognized that the user is satisfied with the answer, the pre-extraction model is used to extract entities and relationships. When it is recognized that the user is dissatisfied with the answer, the entity generation model and the relationship generation model are used to extract entities and relationships from the latest question as a whole. The operation is maintained until the feedback recognition model recognizes that the user is dissatisfied with a certain round of answers. The context is then associated with the latest question data, and the entity generation model and the relationship generation model are used to extract entities and relationships. This is maintained until the end of the session. The increase in the complexity of the question data is unidirectional and irreversible in each session.

Images