An intelligent question-answering method, apparatus, device, and medium

Abstract

This application discloses an intelligent question-answering method, apparatus, device, and medium, relating to the field of intelligent question-answering technology. Applied to a question-answering system, the method includes: acquiring user questions containing multimodal data and their corresponding candidate answer sets; determining initial candidate answers based on the semantic similarity between the user questions and the candidate answer sets; acquiring historical dialogue information during user interaction and performing contextual analysis using a sliding window to generate intent prediction results for the user questions; generating new question sequences semantically similar to the user questions using a semantic generation model based on the intent prediction results, and constructing an expanded question-answering library based on the new question sequences; acquiring user feedback data, and generating optimized question-answering results based on the user feedback data and data in the expanded question-answering library using a closed-loop optimization model, and pushing these results to the user. The technical solution of this application can significantly improve the accuracy of question-answering matching and optimize the user query experience.

Description

Technical Field

[0001] This invention relates to the field of intelligent question-answering technology, and in particular to an intelligent question-answering method, apparatus, device, and medium. Background Technology

[0002] With the development of artificial intelligence technology, intelligent question-answering systems have been widely used in many fields, such as customer service, education, and healthcare. However, existing intelligent question-answering systems often face problems of comprehension bias and inaccurate response when dealing with complex semantics and diverse questions. In particular, when user input is ambiguous, ambiguous, or has multi-layered semantics, the system struggles to accurately predict user intent and generate reasonable answers.

[0003] Currently, most intelligent question-answering systems rely on rule-based matching methods or simple keyword search. However, these methods are prone to insufficient semantic matching and misinterpretation of intent when dealing with semantically rich and context-dependent dialogues, impacting the system's response accuracy and user experience. Therefore, improving the semantic understanding and answer generation of intelligent question-answering systems through more precise semantic similarity matching and intent prediction mechanisms, thereby optimizing overall system performance and user experience, has become a crucial issue in intelligent question-answering technology. Summary of the Invention

[0004] In view of this, the purpose of this invention is to provide an intelligent question-answering method, apparatus, device, and medium that can significantly improve the accuracy of question-answering matching and optimize the user query experience. The specific solution is as follows:

[0005] Firstly, this application discloses an intelligent question-answering method, applied to a question-answering system, comprising:

[0006] Obtain user questions containing multimodal data and a set of candidate answers corresponding to the user questions;

[0007] Determine a first semantic similarity between the user's question and the candidate answer set, and determine initial candidate answers from the candidate answer set based on the first semantic similarity;

[0008] Based on the initial candidate answers, historical dialogue information during the interaction with the user is obtained, and the historical dialogue information is analyzed in context using a sliding window to generate an intent prediction result for the user's question.

[0009] Based on the intent prediction results, a new question sequence similar to the user's question is generated using a semantic generation model, and an extended question-answering library is constructed based on the new question sequence.

[0010] Obtain user feedback data, and based on the user feedback data and the data in the extended question-and-answer database, use a closed-loop optimization model to generate optimized question-and-answer results for the initial candidate answers, and push the question-and-answer results to the user; the user feedback data is the rating data of the user's generated answer feedback for different questions.

[0011] Optionally, the multimodal data includes any one or a combination of text data, image data, voice data, facial expression data, and human posture data; wherein, after obtaining the user question containing multimodal data and the candidate answer set corresponding to the user question, the method further includes:

[0012] The format conversion and feature extraction are performed on the user question and the candidate answer set respectively to generate their respective cross-modal semantic embeddings.

[0013] Optionally, the user query is format-converted and features are extracted to generate a corresponding cross-modal semantic embedding, including:

[0014] The text data is transformed into word vectors to obtain a first feature vector, and the first feature vector is input into a pre-trained word embedding model to obtain a first feature representation of the text data.

[0015] A second feature vector of the image data is extracted using a preset convolutional neural network, and a second feature representation of the image data is determined based on the second feature vector;

[0016] The speech data is processed using a preset acoustic feature extraction method to obtain a third feature vector, and the third feature representation of the speech data is determined based on the third feature vector.

[0017] The fourth feature vector of the facial expression data is extracted using a preset facial expression analysis tool, and the fourth feature representation of the facial expression data is determined based on the fourth feature vector;

[0018] The fifth feature vector of the human posture data is extracted using a preset human posture estimation model, and the fifth feature representation of the human posture data is determined based on the fifth feature vector.

[0019] The first feature representation, the second feature representation, the third feature representation, the fourth feature representation, and the fifth feature representation are aligned, and the aligned first feature representation, second feature representation, third feature representation, fourth feature representation, and fifth feature representation are mapped to a shared semantic space using a preset multimodal embedding model to obtain a unified cross-modal semantic embedding.

[0020] Optionally, obtaining the candidate answer set corresponding to the user's question includes:

[0021] Obtain the first candidate answer corresponding to the user's question from the original question and answer database, and construct a candidate answer set based on the first candidate answer;

[0022] Accordingly, after constructing the extended question-answering library based on the new question sequence, the method further includes:

[0023] Obtain a second candidate answer corresponding to the new question sequence from the extended question-answer library, and add the second candidate answer to the candidate answer set.

[0024] Optionally, after constructing the extended question-answering library based on the new question sequence, the method further includes:

[0025] Check the second semantic similarity between the new question sequence generated in the expanded question-answer database and the existing question sequence in the original question-answer database, and review the answers to the new question sequence;

[0026] When the second semantic similarity is greater than or equal to the preset similarity threshold, and / or the review result of the question answer does not meet the preset answer standard, a target question answer pair is determined, and quality control is performed on the target question answer pair.

[0027] Optionally, the step of generating optimized question-and-answer results based on the user feedback data and the data in the expanded question-and-answer database using a closed-loop optimization model includes:

[0028] Based on the user feedback data and the data in the extended question and answer database, a quality score is determined for each question-answer pair in the extended question and answer database;

[0029] The optimization parameters of the expanded question-and-answer database are determined using the quality score, and the optimization parameters and the user feedback data are input into a closed-loop optimization model to generate question-and-answer results optimized from the initial candidate answers.

[0030] The formula for calculating the quality score is as follows: The formula for calculating the optimization parameters is as follows: ; Let be the quality score of the j-th question-answer pair, and n be the total number of question-answer pairs in the expanded question-answer database. Rate the user's feedback on the i-th question. For the user's i-th question and the j-th problem in the new problem sequence Semantic similarity between them; is the adjustment coefficient, and m is the number of questions in the question sequence under the current user intent category.

[0031] Optionally, after generating the optimized question-and-answer result based on the initial candidate answer using the closed-loop optimization model, the method further includes:

[0032] Based on the question and answer results, a corresponding voice broadcast is generated, and the voice-driven face model is used to drive the virtual interactive agent to perform lip-sync and facial movements based on the voice broadcast.

[0033] Based on the intent prediction results, the virtual interaction agent is driven by a pose synthesis model to generate a coherent sequence of body movements.

[0034] Secondly, this application discloses an intelligent question-answering device, applied to a question-answering system, comprising:

[0035] The data acquisition module is used to acquire user questions containing multimodal data and a set of candidate answers corresponding to the user questions;

[0036] A semantic matching module is used to determine a first semantic similarity between the user's question and the candidate answer set, and to determine an initial candidate answer from the candidate answers based on the first semantic similarity.

[0037] The intent prediction module is used to obtain historical dialogue information during the interaction with the user based on the initial candidate answer, and to perform contextual analysis on the historical dialogue information using a sliding window to generate intent prediction results for the user's question.

[0038] An extended question-answering library construction module is used to generate a new question sequence that is semantically similar to the user's question based on the intent prediction result, using a semantic generation model, and to construct an extended question-answering library based on the new question sequence;

[0039] The question-and-answer result generation module is used to acquire user feedback data, and based on the user feedback data and the data in the extended question-and-answer database, use a closed-loop optimization model to generate optimized question-and-answer results for the initial candidate answers, and push the question-and-answer results to the user; the user feedback data is the rating data of the user's generated answers to different questions.

[0040] Thirdly, this application discloses an electronic device, which includes a processor and a memory; wherein the memory is used to store a computer program, which is loaded and executed by the processor to implement the intelligent question-answering method as described above.

[0041] Fourthly, this application discloses a computer-readable storage medium for storing a computer program; wherein the computer program, when executed by a processor, implements the intelligent question-answering method as described above.

[0042] This application provides an intelligent question-answering method applied to a question-answering system, comprising: acquiring user questions containing multimodal data and a set of candidate answers corresponding to the user questions; determining a first semantic similarity between the user questions and the set of candidate answers, and determining initial candidate answers from the set of candidate answers based on the first semantic similarity; acquiring historical dialogue information during user interaction based on the initial candidate answers, and performing contextual analysis on the historical dialogue information using a sliding window to generate an intent prediction result for the user questions; generating a new question sequence semantically similar to the user questions using a semantic generation model based on the intent prediction results, and constructing an extended question-answering library based on the new question sequence; acquiring user feedback data, and generating an optimized question-answering result for the initial candidate answers using a closed-loop optimization model based on the user feedback data and data in the extended question-answering library, and pushing the question-answering result to the user; wherein the user feedback data is rating data of user feedback on generated answers to different questions.

[0043] The beneficial technical effects of this application are as follows: For user questions based on multimodal data, firstly, initial candidate answers are selected by calculating the semantic similarity between the question and the candidate answers, thereby improving the accuracy of question-and-answer matching; secondly, a sliding window prediction mechanism is introduced to generate intent prediction results for user questions by predicting user needs, and the knowledge base of the question-and-answer system is enriched by semantic similarity question generation technology to expand the question-and-answer database, significantly enhancing the diversity and comprehensiveness of the knowledge base and improving the system's ability to cover user questions; finally, based on user feedback and the expanded question-and-answer database, more accurate answers that meet user needs are provided through closed-loop optimization.

[0044] In addition, the intelligent question-answering device, equipment and storage medium provided in this application correspond to the above-mentioned intelligent question-answering method and have the same effect. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0046] Figure 1 This is a flowchart of an intelligent question-answering method disclosed in this application;

[0047] Figure 2 This is a schematic diagram of the structure of a question-and-answer system disclosed in this application;

[0048] Figure 3 This is a schematic diagram of the structure of an intelligent question-answering device disclosed in this application;

[0049] Figure 4 This is a structural diagram of an electronic device disclosed in this application. Detailed Implementation

[0050] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0051] Currently, intelligent question-answering systems often face problems of misunderstanding and inaccurate response when dealing with complex semantics and diverse questions. In particular, when user input is ambiguous, ambiguous, or has multi-layered semantics, the system struggles to accurately predict user intent and generate reasonable answers.

[0052] To this end, this application provides an intelligent question-answering solution that can accurately and effectively predict user intent, significantly improve the accuracy of question-answering matching, and enhance the level of interactive intelligence and user experience.

[0053] This invention discloses an intelligent question-answering method, see [link to relevant documentation]. Figure 1 As shown, applied to a question-answering system, the method includes:

[0054] Step S11: Obtain user questions containing multimodal data and a set of candidate answers corresponding to the user questions.

[0055] To meet the needs of real-time interaction, the intelligent system employs a streaming output mechanism and feedback parameter input to process user requests through a process of simultaneous generation and feedback. In this embodiment, the system first acquires the multimodal data input by the user when asking a question. Upon receiving the user request, the system immediately performs initial intent judgment and semantic parsing without waiting for all processing to complete. At this point, the system provides a set of candidate answers corresponding to the user's question.

[0056] In one specific implementation, user questions using multimodal data can include any one or a combination of text data, image data, voice data, facial expression data, and human posture data input by the user when asking the question. By integrating information from multiple modalities such as text, images, voice, facial expressions, and posture, synchronous processing of multimodal data can be achieved, improving the accuracy of question-and-answer matching.

[0057] In another specific implementation, the candidate answer set is a collection of at least one candidate answer generated in response to a user's question. The candidate answer set comes from two sources: one is the first candidate answer, initially selected from the original question-and-answer database of the intelligent system, corresponding to the user's question; the other is the second candidate answer, obtained from an expanded question-and-answer database based on a streaming output mechanism, corresponding to the generated new question sequence.

[0058] It should be noted that the extended question-answering library is a knowledge base built based on user questions, used to store a series of different questions semantically similar to the user's question. The question is generated based on the inferred user intent. For a given user question, the extended question-answering library can identify a new question sequence containing at least one corresponding question. These questions may have different syntactic structures, but should semantically remain consistent with the user's original question. For example, assuming the user's question is "How to program?", the new question sequence could include a series of questions such as "How to learn programming?" and "Where should programming begin?". Once the extended question-answering library has been built and a new question sequence corresponding to the user's question exists within it, the system can match one or more candidate answers for each generated question in the new question sequence to obtain a second candidate answer, and add it to the original candidate answer set constructed from the first candidate answer, forming the current candidate answer set.

[0059] Step S12: Determine the first semantic similarity between the user's question and the candidate answer set, and determine the initial candidate answer from the candidate answer set based on the first semantic similarity.

[0060] In this embodiment of the application, the semantic similarity between the user's question and the candidate answer set is calculated, and the most relevant answer is selected from the candidate answer set based on the semantic similarity as the output of this round of question answering.

[0061] Specifically, by using cosine similarity The formula for calculating the similarity between a user's question and the set of candidate answers is: Where Q is an embedding vector in M ​​representing a user's question; , For the candidate answer set, each The semantic embedding vector representing the candidate answer. d is the vector dimension; Let represent the dot product of the question vector and the candidate answer vector, and q represent the semantic embedding vector of the user question Q obtained after cross-modal embedding. ; and These represent the magnitudes of the question vector and the candidate answer vector, respectively. This similarity value reflects the semantic similarity between the user's question and each candidate answer.

[0062] In this embodiment of the application, the semantic similarity between all candidate answers in the candidate answer set and the user's question is calculated. Select the candidate answer with the highest similarity from them. ,Right now As the initial candidate answer determined in this round of question and answer.

[0063] Step S13: Based on the initial candidate answers, obtain historical dialogue information during the interaction with the user, and use a sliding window to perform contextual analysis on the historical dialogue information to generate an intent prediction result for the user's question.

[0064] To better understand user intent, in this embodiment, the system employs a three-round conversational sliding window mechanism, combining historical contextual information from the user's interaction with the system to construct a semantically enhanced representation. Specifically, in each round of user questioning, the system automatically captures the content of the user's most recent three rounds of question-and-answer interactions with the system (including user questions and system responses), forming a sliding window context sequence.

[0065] Among them, context This is historical dialogue information during user-system interactions. The sliding window, size w, is used to analyze the user's context at the current moment. Each context unit... Corresponding to an embedding vector .

[0066] When using a sliding window to process the context, assuming the context window at the current time t is... This window contains the user's conversation history over the past w time steps. The window size w=3, indicating that the most recent three turns of conversation are used as the current context. For each sliding window... Calculate its semantic representation This involves merging the vector representations of all context units within the current window to form a new semantic embedding representation: This expression represents a window. The average vector representation of the window is used to generate a global semantic embedding of the window using contextual information within the window.

[0067] Furthermore, combining the semantic vector q of the user's question with the current sliding window Average semantic vector representation Calculate the prediction result of user intent. Among them, For sliding windows The average value of each context vector is used to enhance the contextual information for question semantics.

[0068] In this embodiment of the application, an intent prediction model is used when generating intent prediction results for user queries. This model combines question and contextual information to predict user intent. The input to the intent prediction model is the semantic representation of the question and the semantic representation of the context window; the output is the predicted user intent label I. This prediction process can be described by the following formula: ;in, This is an intent prediction model that calculates the user's intent based on the semantic vectors of the question and the context. The intent prediction result I is a classification result, representing the most likely intent category determined by the system based on the user's question and the context. It plays a "guiding and controlling role" in the subsequent semantic generation model, instructing the generation model to generate a sequence of questions semantically related to that intent.

[0069] Step S14: Based on the intent prediction result, generate a new question sequence that is semantically similar to the user's question using a semantic generation model, and build an extended question-answering library based on the new question sequence.

[0070] In this embodiment of the application, the category of the user's intent can be inferred based on the intent prediction result I obtained from the aforementioned steps. For example, suppose a user asks "How to program?", the intent prediction result is... (e.g., "programming tutorial"), thus enabling the determination of the need to generate a series of variant questions with similar semantics for this intent.

[0071] Specifically, the intent prediction result I is used as input to the process of generating a new question sequence, and the input also includes the user's original question text. Based on this, a semantic generation model is used. Generate a set of semantically similar question sequences The mathematical model of this generation process can be represented as: ;in, This is the original question entered by the user when asking the question. That is, the intention prediction result obtained from the intention prediction model. It is a semantic generation model.

[0072] Furthermore, a new sequence of questions is generated, containing multiple questions. After generation, it will be stored in the extended question and answer database. In other words, the extended question-and-answer database is built upon a sequence of new questions generated in response to user inquiries. It's important to note that the extended database not only stores these question sequences but also their corresponding standard answers. That is, the standard answer for each generated question, ensuring that the new question can be effectively matched with the data in the original question-and-answer database.

[0073] This step can be completed using the following formula: ;in, This indicates an expansion of the question-and-answer database. This represents the generated new question and its corresponding standard answer. In this way, newly generated question sequences and their answer pairs can be inserted into an expanded question-and-answer database, further enriching the database's content.

[0074] Step S15: Obtain user feedback data, and based on the user feedback data and the data in the extended question-and-answer database, use a closed-loop optimization model to generate optimized question-and-answer results for the initial candidate answers, and push the question-and-answer results to the user; the user feedback data is the rating data of the user's generated answer feedback for different questions.

[0075] In this embodiment of the application, the extended question-and-answer database and user feedback data are analyzed. User feedback data... The system generates rating data for users to provide feedback on answers to different questions. Ask a question for the i-th user. This is a rating system for user feedback on questions asked by the i-th user. User feedback data provides a quality assessment of the questions and answers generated during user interactions, while the extended question-and-answer database provides relevant semantic matching and answer information. The data in the extended question-and-answer database consists of question-answer pairs composed of a new question sequence and its corresponding standard answer. ,in, For the new problem generated, The standard answer to the new question.

[0076] It's important to note that analyzing the expanded question-and-answer database and user feedback data yields a quality score for each question-and-answer pair in the database. This score can then be used to determine the optimization parameters for the expanded database. These optimization parameters are adjustment factors generated based on user feedback data and low-scoring questions in the database. Furthermore, by analyzing the user feedback data and the optimization parameters of the expanded database using a closed-loop optimization model, various parts of the question-and-answer system are comprehensively adjusted to generate optimized question-and-answer results, improving the overall question-and-answer matching effect. Ultimately, the optimized question-and-answer system can better match user needs and generate more accurate answers.

[0077] This application provides an intelligent question-answering method applied to a question-answering system, comprising: acquiring user questions containing multimodal data and a set of candidate answers corresponding to the user questions; determining a first semantic similarity between the user questions and the set of candidate answers, and determining initial candidate answers from the set of candidate answers based on the first semantic similarity; acquiring historical dialogue information during user interaction based on the initial candidate answers, and performing contextual analysis on the historical dialogue information using a sliding window to generate an intent prediction result for the user questions; generating a new question sequence semantically similar to the user questions using a semantic generation model based on the intent prediction results, and constructing an extended question-answering library based on the new question sequence; acquiring user feedback data, and generating an optimized question-answering result for the initial candidate answers using a closed-loop optimization model based on the user feedback data and data in the extended question-answering library, and pushing the question-answering result to the user; wherein the user feedback data is rating data of user feedback on generated answers to different questions.

[0078] The beneficial technical effects of this application are as follows: For user questions based on multimodal data, firstly, initial candidate answers are selected by calculating the semantic similarity between the question and the candidate answers, thereby improving the accuracy of question-and-answer matching; secondly, a sliding window prediction mechanism is introduced to generate intent prediction results for user questions by predicting user needs, and the knowledge base of the question-and-answer system is enriched by semantic similarity question generation technology to expand the question-and-answer database, significantly enhancing the diversity and comprehensiveness of the knowledge base and improving the system's ability to cover user questions; finally, based on user feedback and the expanded question-and-answer database, more accurate answers that meet user needs are provided through closed-loop optimization.

[0079] Based on the foregoing embodiments, in a feasible implementation, in order to achieve effective fusion of multimodal data, after obtaining the user question and candidate answer set, this application embodiment performs format conversion and feature extraction on the user question and candidate answer set respectively to generate their respective cross-modal semantic embeddings. These are two parallel processes processed separately but with the same structure. Specifically, performing format conversion and feature extraction on the user question to generate the corresponding cross-modal semantic embedding includes the following steps:

[0080] Step 1: Perform word vector transformation on the text data to obtain a first feature vector, and input the first feature vector into a pre-trained word embedding model to obtain a first feature representation of the text data.

[0081] In this embodiment, the input text data is denoted as T. When performing format conversion and feature extraction on the text data T, word vector conversion is first performed on each text unit to obtain the text's feature vector. Where d represents the dimension of the word vector. Furthermore, the feature representation of the text data is obtained through the trained word embedding model. Each element in the represents a feature representation corresponding to a single text unit.

[0082] Step 2: Extract the second feature vector of the image data using a preset convolutional neural network, and determine the second feature representation of the image data based on the second feature vector.

[0083] In this embodiment, the input image data is denoted as I. When performing format conversion and feature extraction on image data I, a convolutional neural network (CNN) is first used to extract the feature vector of each image. Further, feature representations of the image data are obtained. Each element in the array represents a feature representation of a single image.

[0084] Step 3: Process the speech data using a preset acoustic feature extraction method to obtain a third feature vector, and determine the third feature representation of the speech data based on the third feature vector.

[0085] In this embodiment, the input speech data is denoted as S. When performing format conversion and feature extraction on the speech data S, acoustic feature extraction methods, such as MFCC (Mel-frequency cepstral coefficients), are used to process each speech signal to obtain a speech feature vector. Further, feature representations of the speech data are obtained. Each element in the denominator represents a feature representation of a single speech signal.

[0086] Step 4: Use a preset facial expression analysis tool to extract the fourth feature vector of the expression data, and determine the fourth feature representation of the expression data based on the fourth feature vector.

[0087] In this embodiment, the input facial expression data is denoted as E. When performing format conversion and feature extraction on the facial expression data E, the feature vector of each facial expression frame is first extracted using a facial expression analysis tool, such as FACS (Facial Action Coding System). Further, feature representations of facial expression data are obtained. Each element in the array represents a feature representation corresponding to a facial expression frame.

[0088] Step 5: Extract the fifth feature vector of the human posture data using a preset human posture estimation model, and determine the fifth feature representation of the human posture data based on the fifth feature vector.

[0089] In this embodiment, the input human pose data is denoted as P. When performing format conversion and feature extraction on the human pose data P, a human pose estimation model, such as OpenPose (an open-source human pose estimation model), is first used to extract the feature vector of each human pose input, thus obtaining the feature representation of the pose data. Further, feature representations of human posture data are obtained. Each element in the graph represents a feature representation corresponding to a pose skeleton keypoint map.

[0090] Step 6: Align the first feature representation, the second feature representation, the third feature representation, the fourth feature representation, and the fifth feature representation, and use a preset multimodal embedding model to map the aligned first feature representation, second feature representation, third feature representation, fourth feature representation, and fifth feature representation to a shared semantic space to obtain a unified cross-modal semantic embedding.

[0091] In this embodiment, after obtaining the feature representations corresponding to each of the multimodal data, the corresponding feature representations are aligned, and then a unified cross-modal semantic embedding is generated through a cross-modal learning method. ;in This represents the embedding vector for each modality of data.

[0092] Specifically, when mapping aligned feature representations to the same vector space using cross-modal learning methods, a multimodal embedding model is used to map T, I, S, E, and P, calculating the representation of each modality's data in the shared semantic space, and obtaining a unified cross-modal embedding matrix M. The multimodal embedding model uses a mapping function... Completed It can map data from different modalities to a shared semantic space. .

[0093] It is worth noting that the candidate answer set also needs to undergo feature extraction based on its corresponding multimodal data and be processed using the same embedding function. This yields a semantic vector representation. This ensures that the user's question and the answers in the candidate answer set reside in the same semantic space.

[0094] Based on the foregoing embodiments, this embodiment will specifically elaborate on steps S14 and S15 in the above embodiments. In order to ensure the quality of the extended question-and-answer database and guarantee that users can obtain high-quality and accurate answers, after constructing the extended question-and-answer database based on the new question sequence, the following steps may also be included:

[0095] Step 1: Check the second semantic similarity between the new question sequence generated in the expanded question-answer database and the existing question sequence in the original question-answer database, and review the answers to the new question sequence;

[0096] Step 2: When the second semantic similarity is greater than or equal to the preset similarity threshold, and / or the review result of the question answer does not meet the preset answer standard, determine the target question answer pair, and perform quality control on the target question answer pair.

[0097] In this embodiment of the application, to ensure the data quality in the expanded question-answer database, a quality control module is set up in the intelligent system to verify the question-answer pairs corresponding to the generated new question sequences. This step is performed in the following way:

[0098] Examine new question sequences using semantic similarity metrics, such as cosine similarity or similarity calculations using deep learning models. With existing question sequences in the original question-answering database Second semantic similarity between This is to ensure that there is not too much duplication in the generated issues.

[0099] For the generated answer The review process aims to ensure the high accuracy and semantic consistency of the answers in the expanded question-and-answer database. Understandably, although the standard answer is automatically generated by the system or automatically matched from the original question-and-answer database, potential issues such as semantic drift and insufficient ambiguity coverage still exist. Therefore, a human-assisted review mechanism is needed to verify its accuracy. The review process may include a dual strategy of confidence screening using natural language processing models and manual review to ensure that the answers ultimately included in the database do not mislead users or reduce the reliability of the question-and-answer system.

[0100] If the similarity is too high or the answer is problematic, the system will roll back and regenerate or adjust the question and answer pair. This process can be mathematically represented as: ;in, It is a set similarity threshold. If the similarity exceeds this threshold, the new problem is considered too similar to the existing problem and needs to be adjusted.

[0101] Based on the content of the foregoing embodiments, it is known that analyzing user feedback data... Matching questions and answers in the extended question-and-answer database Ultimately, these parameters are used to determine the optimization parameters in the input closed-loop optimization model. The specific process for determining the optimization parameters of the extended question-and-answer database is explained below:

[0102] Specifically, the step of generating optimized question-and-answer results based on the user feedback data and the data in the expanded question-and-answer database using a closed-loop optimization model includes the following steps:

[0103] Step 1: Determine the quality score for each question-answer pair in the extended question-answer database based on the user feedback data and the data in the extended question-answer database.

[0104] First, we need to rate the user's feedback on the question asked by the i-th user. Perform weighted processing, and then combine it with the corresponding user's question. With the answer Analyze the results to obtain a quality score for each question-answer pair. : ;in, Let be the quality score of the j-th question-answer pair, and n be the total number of question-answer pairs in the expanded question-answer database. Rate the user's feedback on the i-th question. For the user's i-th question and the j-th problem in the new problem sequence Semantic similarity between them;

[0105] Step 2: Use the quality score to determine the optimization parameters of the extended question-and-answer database, and input the optimization parameters and the user feedback data into the closed-loop optimization model to generate the optimized question-and-answer results based on the initial candidate answers.

[0106] In this embodiment of the application, based on the obtained quality score The system can further evaluate the quality of questions in the expanded question-and-answer database, filtering out question-and-answer pairs that do not meet the standards, i.e., those with low quality scores, and then deleting or replacing them, providing a basis for subsequent optimization steps. Through this analysis, the system can identify questions that do not match user feedback or are of low quality, and then optimize or replace these questions.

[0107] For example, if the quality score of a question-and-answer pair is lower than a certain predetermined threshold... Then the question-and-answer pair will be marked as an item to be updated: .

[0108] In this embodiment of the application, based on the obtained quality score The system generates optimized parameters for an extended question-and-answer database. This is used to adjust and optimize the matching and generation strategies of the question-answering system. Specifically, it generates optimization parameters. The process can be represented by the following formula: ;in, The adjustment coefficient is m, which represents the number of questions in the question sequence under the current user intent category. For the user's i-th question and the j-th problem in the new problem sequence Semantic similarity between them.

[0109] As can be seen, user feedback data provides a quality assessment of the questions and answers generated during user interactions, while the extended question-and-answer database provides relevant semantic matching and answer information. Determining the optimization parameters of the extended question-and-answer database reveals which question-and-answer pairs are of suitable quality and which require optimization. These optimization parameters will then be used to adjust the questions and answers in the database to improve the overall performance of the question-and-answer system.

[0110] Furthermore, in generating optimization parameters Afterwards, the system needs to update and expand the question-answer database based on the optimization results to ensure that the question-answer pairs in the database better meet user needs and improve the accuracy of semantic matching and intent prediction. This is based on the obtained quality score. It can filter out low-quality question-answer pairs, and for the low-quality question-answer pairs that are replaced, optimization parameters can be generated. and user feedback data This allows for the generation of new question-and-answer pairs, or the improvement of the quality of existing pairs by adjusting them. The updated question-and-answer pairs will be reintroduced into the expanded question-and-answer library and continue to be used by the question-and-answer system.

[0111] In one feasible implementation, by adjusting optimization parameters, the system can also optimize the data structure in the extended question-and-answer database to improve retrieval efficiency and matching accuracy. For example, optimizing the database index structure or adjusting the storage format can improve query speed. Ultimately, the updated extended question-and-answer database can contain question-and-answer pairs that better meet user needs, improving the accuracy of semantic matching and intent prediction.

[0112] Furthermore, in this embodiment, user feedback data and optimization parameters of the extended knowledge base are input into a closed-loop optimization model to jointly optimize the semantic matching process, intent prediction process, and final answer generation process in the process of generating question-and-answer results, so as to improve the overall performance of the system.

[0113] The core of the closed-loop optimization model lies in achieving dynamic performance improvement and feedback adjustment. The input of the model can be represented as: The process of combining semantic matching, intent prediction, and final answer generation includes the following steps:

[0114] First, in input data Next, the semantic matching process compares the semantic similarity between the user input and existing questions to perform question matching. The optimized semantic matching algorithm incorporates the semantic similarity score from user feedback data to update the matching strategy and enhance matching accuracy. This update is performed using the following formula: ;in, Semantic similarity represents the degree of matching between the user's question and the generated question; The rating is based on user feedback, indicating user satisfaction with the issue. and The weighting coefficients are obtained through optimized calculations and are used to balance the influence of the original similarity and the user feedback score.

[0115] Secondly, the intent prediction process is optimized. This process analyzes the user's input text to predict their query intent. During the closed-loop optimization process, user feedback data helps improve the accuracy of the model in identifying user intent. Specifically, the system optimizes the intent prediction model using the following adjustment formula: ;in, Indicates user intent The predicted probability; For the optimized semantic similarity, m represents the similarity in the current user intent category. The number of question sequences generated. The optimized intent prediction model can better identify and match users' actual needs, thereby improving the system's accuracy in predicting intent.

[0116] Finally, the answer generation process is optimized. This process is responsible for generating specific answers from the intent prediction results. During closed-loop optimization, the generation module needs to be adjusted based on the optimized intent prediction results and semantic matching to generate more accurate and appropriate answers. This adjustment is achieved using the following formula: ;in The generated answer; It's a generative model that predicts results based on user intent. Semantic similarity and optimization parameters Generate answers. Through closed-loop optimization, the generative model can provide more accurate answers that meet user needs based on user feedback and optimized parameters.

[0117] After completing the joint optimization process described above, the closed-loop optimization model outputs optimized question-answering results. This result not only combines user feedback data and optimization parameters but also improves the overall system performance through optimizations in semantic matching, intent prediction, and answer generation. Ultimately, the optimized question-answering system can better match user needs and generate more accurate answers. The specific output is as follows:

[0118] ;in, This represents the optimized set of generated answers; The probability of the intended prediction; This is the minimum acceptance probability threshold; answers below this threshold will be discarded. This output allows the system to push optimized question-and-answer results to users, improving the user experience.

[0119] Based on the above embodiments, in one feasible implementation, to enhance the expressiveness and immersiveness of the system interaction, the system introduces an audio-driven dynamic facial expression and gesture synthesis mechanism after generating question-and-answer results. This mechanism can drive a virtual interactive agent, such as a digital human, to perform facial expressions and body movements in real time based on the generated speech or answer content. Specifically, it may also include the following steps:

[0120] Based on the question and answer results, a corresponding voice broadcast is generated, and the voice-driven face model is used to drive the virtual interactive agent to perform lip-sync and facial movements based on the voice broadcast.

[0121] Based on the intent prediction results, the virtual interaction agent is driven by a pose synthesis model to generate a coherent sequence of body movements.

[0122] In one specific implementation, the voice-driven face model can be a Wav2Lip model, which precisely drives lip-sync and facial movements based on speech; the pose synthesis model can be an E-NeRF model, which generates natural and coherent sequences of body movements based on semantic intent and emotional tags. This achieves multimodal consistency of expression, movement, and speech, enhancing the realism and naturalness of human-computer interaction.

[0123] It is worth noting that this invention employs a streaming output mechanism and feedback parameter input to meet real-time interaction requirements. The system generates and provides feedback on user requests simultaneously through an in-stream output mechanism. This mechanism optimizes the output strategy of the system's generation module to achieve a smoother, lower-latency response experience. Specifically, upon receiving a user request, the system immediately performs initial intent judgment and semantic parsing without waiting for full processing; the generation module adopts a step-by-step generation strategy, outputting partial results in real-time during content generation; the output content can be synchronously transmitted to the voice broadcast module and the digital human rendering module, enabling coordinated updates of voice, facial expressions, and gestures, achieving a natural interactive effect of "speaking and moving simultaneously" and "answering and displaying simultaneously." Therefore, this mechanism significantly reduces system response latency and is suitable for applications with high real-time requirements, such as virtual human dialogue and voice question-and-answer. Furthermore, after the streaming output is completed, the system inputs user feedback data and optimization parameters into a closed-loop optimization model.

[0124] Based on the foregoing embodiments, this embodiment provides, exemplarily, modules included in a question-answering system. The question-answering system may include: a data acquisition module, a semantic matching module, an intent prediction module, an answer generation module, a question-answering quality assessment and optimization module, a knowledge base update module, and a user interaction and visualization module.

[0125] The data acquisition module 10 is used to acquire data related to user questions and system responses from multiple data sources. Specifically, this module collects user-input questions in real time from user interactions with the system. The answer returned by the system and user feedback data on the answers. This includes evaluation scores, suggested improvements, and additional contextual information. The collected data will provide a foundation for subsequent semantic analysis and question-answering quality assessment.

[0126] The semantic matching module 20 is mainly used to match user input questions. And questions in the extended question-and-answer database Semantic similarity matching is performed. By calculating the similarity between questions, natural language processing (NLP) techniques and deep learning models are used to evaluate the relevance between user questions and system answers. The system uses pre-trained word embeddings or semantic vector models to obtain high-quality semantic matching results. The calculated semantic similarity results serve as the basis for subsequent intent prediction and answer generation.

[0127] The intent prediction module 30 receives similarity data provided by the semantic matching module and predicts the user's true needs by analyzing the user's intent. The system uses deep learning algorithms to train the intent prediction model, identifying the core intent of the user's question based on a large amount of historical data and user feedback, and outputting possible answer categories. The accuracy of intent prediction directly affects the system's answer generation performance; therefore, this module employs a dynamic adjustment mechanism to ensure the model is continuously optimized to meet different types of user needs.

[0128] The answer generation module 40, based on the user needs output by the intent prediction module, combines existing answers from the knowledge base. The system combines system-generated content with Natural Language Generation (NLG) technology to generate personalized answers that meet user needs. The system considers multiple dimensions, including question type, user preferences, and contextual information, to generate the best answer, ensuring its relevance and readability. The generated answers undergo quality evaluation, and the generation process is further optimized using feedback data.

[0129] The question-and-answer quality assessment and optimization module 50 is responsible for evaluating the quality of the answers generated by the system. This is done by analyzing user feedback data on the answers. and quality score The system continuously optimizes the question-answer pairs in its question-answer database. Specifically, the system calculates a quality score for each question-answer pair and uses the score results to update the knowledge entries in the question-answer database, ensuring that the system consistently provides high-quality answers. Simultaneously, the system adjusts and optimizes parameters based on user feedback, continuously improving the accuracy of semantic matching and answer generation.

[0130] The knowledge base update module 60 is used to dynamically adjust and expand the question-and-answer database based on quality assessment results and optimization parameters. Specifically, the system will adjust the database based on optimization parameters. This module updates the questions and answers in the question-and-answer database, removes low-quality entries, and adds new relevant content. Through a closed-loop feedback mechanism, the knowledge base can continuously adapt to new user needs and question types, improving the overall flexibility and accuracy of the question-and-answer system.

[0131] The user interaction and visualization module 70 provides users with an intuitive interface for entering questions and viewing system-generated answers. Users can provide feedback on the quality and accuracy of the answers through this interface, and the system will further optimize its question-and-answer database based on this feedback. Furthermore, this module supports multi-scenario simulations and interactive operations, helping users understand the system's operation and question-and-answer logic, thus improving the user experience.

[0132] As can be seen, modular design enables efficient processing of user input, accurate prediction of user intent, and generation of high-quality personalized answers. Through continuous optimization and updates to the expanded question-and-answer database, the system can learn and adjust itself in multiple rounds of interaction, improving overall performance. In particular, through the closed-loop feedback mechanism of the question-and-answer quality assessment and optimization module, the system can continuously improve its knowledge base based on user feedback, ensuring that each interaction provides the most relevant and accurate answers. The system's intelligent and adaptive characteristics enable its application in multiple fields, including customer service, technical support, and intelligent assistants, providing users with more efficient and accurate services.

[0133] Accordingly, this application also discloses an intelligent question-answering device, see [link to relevant documentation]. Figure 3 As shown, the device, applied to a question-and-answer system, includes:

[0134] Data acquisition module 11 is used to acquire user questions containing multimodal data and a set of candidate answers corresponding to the user questions;

[0135] The semantic matching module 12 is used to determine a first semantic similarity between the user's question and the candidate answer set, and to determine an initial candidate answer from the candidate answers based on the first semantic similarity.

[0136] The intent prediction module 13 is used to obtain historical dialogue information during the interaction with the user based on the initial candidate answer, and to perform contextual analysis on the historical dialogue information using a sliding window to generate intent prediction results for the user's question.

[0137] The extended question-answering library construction module 14 is used to generate a new question sequence that is semantically similar to the user's question based on the intent prediction result using a semantic generation model, and to construct an extended question-answering library based on the new question sequence;

[0138] The question-and-answer result generation module 15 is used to acquire user feedback data, and based on the user feedback data and the data in the extended question-and-answer database, use a closed-loop optimization model to generate question-and-answer results optimized from the initial candidate answers, and push the question-and-answer results to the user; the user feedback data is the rating data of the user's generated answers to different questions.

[0139] For more detailed information on the working process of each of the above modules, please refer to the relevant content disclosed in the foregoing embodiments, which will not be repeated here.

[0140] Therefore, the above-described solution in this embodiment, when applied to a question-and-answer system, includes: acquiring user questions containing multimodal data and a set of candidate answers corresponding to the user questions; determining a first semantic similarity between the user questions and the set of candidate answers, and determining initial candidate answers from the set of candidate answers based on the first semantic similarity; acquiring historical dialogue information during user interaction based on the initial candidate answers, and performing contextual analysis on the historical dialogue information using a sliding window to generate an intent prediction result for the user questions; generating a new question sequence semantically similar to the user questions using a semantic generation model based on the intent prediction result, and constructing an extended question-and-answer library based on the new question sequence; acquiring user feedback data, and generating optimized question-and-answer results for the initial candidate answers using a closed-loop optimization model based on the user feedback data and the data in the extended question-and-answer library, and pushing the question-and-answer results to the user; the user feedback data is the rating data of the user's generated answer feedback for different questions.

[0141] The beneficial technical effects of this application are as follows: For user questions based on multimodal data, firstly, initial candidate answers are selected by calculating the semantic similarity between the question and the candidate answers, thereby improving the accuracy of question-and-answer matching; secondly, a sliding window prediction mechanism is introduced to generate intent prediction results for user questions by predicting user needs, and the knowledge base of the question-and-answer system is enriched by semantic similarity question generation technology to expand the question-and-answer database, significantly enhancing the diversity and comprehensiveness of the knowledge base and improving the system's ability to cover user questions; finally, based on user feedback and the expanded question-and-answer database, more accurate answers that meet user needs are provided through closed-loop optimization.

[0142] Furthermore, embodiments of this application also disclose an electronic device, Figure 4 This is a structural diagram of an electronic device according to an exemplary embodiment. The content of the diagram should not be construed as limiting the scope of this application.

[0143] Figure 4 This is a schematic diagram of an electronic device provided in an embodiment of this application. Specifically, the electronic device may include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the intelligent question-answering method disclosed in any of the foregoing embodiments. Alternatively, the electronic device in this embodiment may specifically be a computer.

[0144] In this embodiment, the power supply 23 is used to provide operating voltage for various hardware devices on the electronic device; the communication interface 24 can create a data transmission channel between the electronic device and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.

[0145] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk, or optical disk, etc. The resources stored on it can include an operating system 221, computer programs 222, and data 223, etc. The data 223 can include various types of data. The storage method can be temporary storage or permanent storage.

[0146] The operating system 221 is used to manage and control the various hardware devices on the electronic device and the computer program 222, which may be Windows Server, Netware, Unix, Linux, etc. In addition to including a computer program capable of performing the intelligent question-answering method executed by the electronic device as disclosed in any of the foregoing embodiments, the computer program 222 may further include a computer program capable of performing other specific tasks.

[0147] Furthermore, this application also discloses a computer-readable storage medium, which includes random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, magnetic disks, optical disks, or any other form of storage medium known in the art. The computer program, when executed by a processor, implements the aforementioned intelligent question-answering method. Specific steps of this method can be found in the corresponding content disclosed in the foregoing embodiments, and will not be repeated here.

[0148] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.

[0149] The steps of the intelligent question-answering method or algorithm described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0150] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0151] The present invention provides a detailed description of an intelligent question-answering method, apparatus, device, and medium. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. An intelligent question-answering method, characterized in that, Applications in question-answering systems include: Obtain user questions containing multimodal data and a set of candidate answers corresponding to the user questions; A first semantic similarity is determined between the user's question and the candidate answer set, and an initial candidate answer is determined from the candidate answer set based on the first semantic similarity; the initial candidate answer is the candidate answer that has the highest similarity to the user's question. Based on the initial candidate answers, historical dialogue information during the interaction with the user is obtained, and the historical dialogue information is analyzed in context using a sliding window to generate an intent prediction result for the user's question. Based on the intent prediction results, a new question sequence similar to the user's question is generated using a semantic generation model, and an extended question-answering library is built based on the new question sequence. Acquire user feedback data, and based on the user feedback data and the data in the extended question-and-answer database, use a closed-loop optimization model to generate optimized question-and-answer results for the initial candidate answers, and push the question-and-answer results to the user; the user feedback data is the rating data of the user's generated answer feedback for different questions; The step of generating optimized question-and-answer results based on the user feedback data and the data in the expanded question-and-answer database using a closed-loop optimization model includes: Based on the user feedback data and the data in the extended question and answer database, a quality score is determined for each question-answer pair in the extended question and answer database; The optimization parameters of the expanded question-and-answer database are determined using the quality score, and the optimization parameters and the user feedback data are input into a closed-loop optimization model to generate question-and-answer results optimized from the initial candidate answers. The formula for calculating the quality score is as follows: The formula for calculating the optimization parameters is as follows: ; Let be the quality score of the j-th question-answer pair, and n be the total number of question-answer pairs in the expanded question-answer database. Rate the user's feedback on the i-th question. For the user's i-th question and the j-th problem in the new problem sequence Semantic similarity between them; is the adjustment coefficient, and m is the number of questions in the question sequence under the current user intent category.

2. The intelligent question-answering method according to claim 1, characterized in that, The multimodal data includes any one or a combination of text data, image data, voice data, facial expression data, and human posture data; wherein, after obtaining the user question containing multimodal data and the candidate answer set corresponding to the user question, the process further includes: The format conversion and feature extraction are performed on the user question and the candidate answer set respectively to generate their respective cross-modal semantic embeddings.

3. The intelligent question-answering method according to claim 2, characterized in that, The user query undergoes format conversion and feature extraction to generate a corresponding cross-modal semantic embedding, including: The text data is transformed into word vectors to obtain a first feature vector, and the first feature vector is input into a pre-trained word embedding model to obtain a first feature representation of the text data. A second feature vector of the image data is extracted using a preset convolutional neural network, and a second feature representation of the image data is determined based on the second feature vector; The speech data is processed using a preset acoustic feature extraction method to obtain a third feature vector, and the third feature representation of the speech data is determined based on the third feature vector. The fourth feature vector of the facial expression data is extracted using a preset facial expression analysis tool, and the fourth feature representation of the facial expression data is determined based on the fourth feature vector; The fifth feature vector of the human posture data is extracted using a preset human posture estimation model, and the fifth feature representation of the human posture data is determined based on the fifth feature vector. The first feature representation, the second feature representation, the third feature representation, the fourth feature representation, and the fifth feature representation are aligned, and the aligned first feature representation, second feature representation, third feature representation, fourth feature representation, and fifth feature representation are mapped to a shared semantic space using a preset multimodal embedding model to obtain a unified cross-modal semantic embedding.

4. The intelligent question-answering method according to claim 1, characterized in that, Obtaining the candidate answer set corresponding to the user's question includes: Obtain the first candidate answer corresponding to the user's question from the original question and answer database, and construct a candidate answer set based on the first candidate answer; Accordingly, after constructing the extended question-answering library based on the new question sequence, the method further includes: Obtain a second candidate answer corresponding to the new question sequence from the extended question-answer library, and add the second candidate answer to the candidate answer set.

5. The intelligent question-answering method according to claim 4, characterized in that, After constructing the extended question-answering library based on the new question sequence, the method further includes: Check the second semantic similarity between the new question sequence generated in the expanded question-answer database and the existing question sequence in the original question-answer database, and review the answers to the new question sequence; When the second semantic similarity is greater than or equal to the preset similarity threshold, and / or the review result of the question answer does not meet the preset answer standard, a target question answer pair is determined, and quality control is performed on the target question answer pair.

6. The intelligent question-answering method according to any one of claims 1 to 5, characterized in that, After generating the optimized question-and-answer result based on the initial candidate answer using a closed-loop optimization model, the method further includes: Based on the question and answer results, a corresponding voice broadcast is generated, and the voice-driven face model is used to drive the virtual interactive agent to perform lip-sync and facial movements based on the voice broadcast. Based on the intent prediction results, the virtual interaction agent is driven by a pose synthesis model to generate a coherent sequence of body movements.

7. An intelligent question-and-answer device, characterized in that, Applications in question-answering systems include: The data acquisition module is used to acquire user questions containing multimodal data and a set of candidate answers corresponding to the user questions; A semantic matching module is used to determine a first semantic similarity between the user's question and the set of candidate answers, and to determine an initial candidate answer from the candidate answers based on the first semantic similarity; the initial candidate answer is the candidate answer that has the highest similarity to the user's question. The intent prediction module is used to obtain historical dialogue information during the interaction with the user based on the initial candidate answer, and to perform contextual analysis on the historical dialogue information using a sliding window to generate intent prediction results for the user's question. An extended question-answering library construction module is used to generate a new question sequence that is semantically similar to the user's question based on the intent prediction result, using a semantic generation model, and to construct an extended question-answering library based on the new question sequence; The question-and-answer result generation module is used to acquire user feedback data, and based on the user feedback data and the data in the extended question-and-answer database, use a closed-loop optimization model to generate question-and-answer results optimized from the initial candidate answers, and push the question-and-answer results to the user; the user feedback data is the rating data of the user's generated answers to different questions; The question-and-answer result generation module is specifically used for: Based on the user feedback data and the data in the extended question and answer database, a quality score is determined for each question-answer pair in the extended question and answer database; The optimization parameters of the expanded question-and-answer database are determined using the quality score, and the optimization parameters and the user feedback data are input into a closed-loop optimization model to generate question-and-answer results optimized from the initial candidate answers. The formula for calculating the quality score is as follows: The formula for calculating the optimization parameters is as follows: ; Let be the quality score of the j-th question-answer pair, and n be the total number of question-answer pairs in the expanded question-answer database. Rate the user's feedback on the i-th question. For the user's i-th question and the j-th problem in the new problem sequence Semantic similarity between them; is the adjustment coefficient, and m is the number of questions in the question sequence under the current user intent category.

8. An electronic device, characterized in that, The electronic device includes a processor and a memory; wherein the memory is used to store a computer program, which is loaded and executed by the processor to implement the intelligent question-answering method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, Used to store computer programs; wherein the computer programs, when executed by a processor, implement the intelligent question-answering method as described in any one of claims 1 to 6.

Images