1446 results about "Retrieval result" patented technology

The invention relates to a military training-oriented dynamic multi-modal knowledge graph retrieval method, which comprises the following steps of: acquiring a multi-source data set to generate a unified original event stream; based on the original event flow, preliminarily constructing a knowledge graph by adopting a Bayesian prior-based tetrad model, locally embedding and updating through a military training event incremental graph neural network, and dynamically re-estimating the confidence of nodes and edges through a Bayesian evidence propagation algorithm; a pre-training language model in the military field is adopted for analysis and semantic coding, a query semantic vector is generated, a retrieval module is constructed, and a retrieval result is output based on the knowledge graph; and constructing a military retrieval strategy optimization network based on a reinforcement learning framework, dynamically adjusting a retrieval strategy, writing back user feedback data, and optimizing node embedding and retrieval strategy parameters. According to the method, multi-source heterogeneous data can be integrated, a dynamic credible graph is constructed, retrieval accuracy, timeliness and self-adaption are improved, and military training decisions are supported.

The invention relates to the technical field of information retrieval, and provides a multi-dimensional heterogeneous knowledge recall enhanced hybrid retrieval method and system.The method comprises the steps that texts recalled through keyword retrieval, sparse vector retrieval and dense vector retrieval are screened through a reciprocal sorting fusion algorithm, and a text type candidate knowledge list is obtained; based on user query, generating and checking a query statement through a large language model, and retrieving an entity-relationship-attribute triple from the knowledge graph library; based on user query, generating enhanced knowledge through a knowledge graph enhanced retrieval method fusing keyword retrieval, vector retrieval and community retrieval; and carrying out format alignment and duplicate removal on the text type candidate knowledge list, the triple result and the enhanced knowledge to form a multi-modal candidate pool, and carrying out reordering score calculation and ordering on each piece of recall knowledge in the multi-modal candidate pool through a reordering model and a business rule to obtain a final retrieval result. And the coverage blind area of single retrieval on heterogeneous knowledge is solved.

The invention relates to the technical field of computers, and discloses a nuclear power safety report examination method and system, electronic equipment and a storage medium, and the method comprises the steps: receiving an examination request, carrying out the intention recognition and entity extraction based on a nuclear power field large language model, and disassembling an examination task into a structured subtask sequence based on an intention type and an entity attribute; performing multi-modal retrieval based on the structured sub-task sequence, and outputting a multi-source retrieval result; calling time-space association information of the evolutionary knowledge graph, and outputting knowledge support data in combination with the time sequence model; fusing a multi-source retrieval result and the knowledge support data, performing risk value calculation and uncertainty quantification through a deep learning model, and outputting a risk assessment result; and based on the risk assessment result, carrying out federal learning model parameter adjustment, carrying out uplink evidence storage on the review conclusion, the risk assessment result and the evidence chain, and outputting the review result. The method has the advantages of being efficient, accurate, dynamic, safe and closed-loop during examination.

The invention provides a full-text retrieval method and system fusing various types of documents, and relates to the technical field of information retrieval, and the method comprises the following steps: obtaining document representation through document content extraction and structure recognition, generating a cross-modal semantic vector by using word embedding and nonlinear transformation, constructing a hierarchical index and a cross-document association graph, and obtaining a full-text retrieval result; the basic correlation score is calculated after the query request is received, and the comprehensive score of the candidate content segments is calculated based on the association graph to determine the optimal retrieval result, so that unified representation and retrieval of heterogeneous documents are realized, the cross-document retrieval precision and relevance are improved, and the processing capability of a retrieval system on complex queries is enhanced.

The invention discloses a medical question and answer method based on knowledge graph and retrieval enhancement generation, and belongs to the technical field of natural language processing and knowledge management. The method includes the following steps that text blocks are obtained on the basis of medical document preprocessing and partitioning, and an abstract tree is constructed; entity relationship extraction is carried out based on the text blocks, and a knowledge graph is constructed; constructing a mapping relationship between the abstract tree and the knowledge graph, and obtaining a vector database; performing entity extraction and query association degree calculation based on user query, judging a retrieval type of the user query, and obtaining a corresponding retrieval result by utilizing the vector database according to the retrieval type; and inputting a retrieval result into the large language model to generate a result. Aiming at the defect of insufficient quality of intelligent questions and answers generated in the medical field by retrieval enhancement generation, the method disclosed by the invention has the advantages that depth and breadth can be considered by one-time retrieval by constructing an abstract tree and a knowledge graph, forming an abstract tree-knowledge graph double-layer index and dynamically typing according to the query association degree, so that the accuracy and efficiency of medical questions and answers are improved.

The invention provides a retrieval enhancement generation method and system for multi-modal semantic alignment in the technical field of computer vision and natural language processing. The method comprises the steps that S1, a large amount of text data, image data and 3D model data are acquired to construct a multi-modal vector knowledge base; s2, acquiring an input query image and a natural language question, and performing joint semantic coding on the query image and the natural language question to obtain a multi-modal query feature vector; s3, evaluating complexity, dynamically adjusting retrieval parameters based on the complexity, and retrieving Top-K candidate knowledge from the multi-modal vector knowledge base through an approximate nearest neighbor search method based on the retrieval parameters and the multi-modal query feature vectors; and S4, in combination with multi-dimensional similarity and token-level interaction alignment, screening each candidate knowledge, and outputting a retrieval result. The method has the advantages that the reasoning precision, the answer credibility and the response efficiency in a complex question and answer scene are greatly improved.

The invention relates to the field of artificial intelligence, in particular to a long text segmentation information retrieval result coherence enhancement method, system and equipment based on a knowledge graph, and the method comprises the following steps: receiving a long text input by a user; segmenting the long text into a plurality of segments according to a preset length, and recording the original position and sequence of each segment; performing named entity recognition on each fragment, and extracting a key entity; querying related information in a knowledge graph according to the key entity, and obtaining attributes of the entity and a relationship between the attributes; integrating the related information obtained by query into the corresponding text fragment to form a context enhanced fragment; receiving a user query request; calculating a correlation score based on the context enhanced fragment and the user query, and retrieving a related text fragment by using a hybrid retrieval strategy; and selecting the related text fragment with the highest score to generate a final answer, and outputting the final answer to the user. Therefore, the information integrity and the context continuity in the retrieval process are ensured.

The invention relates to a knowledge graph enhanced multi-modal file retrieval method, and belongs to the field of artificial intelligence and multi-modal information retrieval. The invention aims to solve the problems of multi-modal information semantic segmentation, weak semantic reasoning ability and low semantic matching precision in the existing multi-modal archive resource retrieval process. Through four stages of archive multi-modal data preprocessing and feature extraction, knowledge graph construction and enhancement, semantic retrieval request analysis and intention modeling, and multi-modal semantic matching and sorting, a semantic relationship is enhanced by utilizing a knowledge graph, so that the semantic relevancy and context consistency of a retrieval result are remarkably improved; a user is allowed to input and inquire in various forms such as texts, images and voices, and semantic extension retrieval is supported. According to the method, more efficient and accurate archive resource retrieval can be realized, and the user retrieval experience and archive knowledge utilization are improved.

The invention discloses a product data question and answer method based on a large model and mixed retrieval and related equipment. The method comprises the following steps: constructing a double-track knowledge base based on multi-source heterogeneous product data; performing multi-dimensional analysis on the current query statement in combination with the large model to obtain a multi-dimensional analysis result; the multi-dimensional analysis result comprises a query vector, a structured query instruction and a keyword; inputting the query vector into a vector database for retrieval to obtain a vector retrieval result; inputting the structured query instruction into a graph database for retrieval to obtain an entity relationship retrieval result; inputting the keyword into the text knowledge fragment for keyword matching to obtain a keyword retrieval result; and combining the vector retrieval result, the entity relationship retrieval result and the keyword retrieval result, and sorting and screening the combined mixed retrieval result through a large language model to obtain target answer data. The method can significantly improve the accuracy and reliability of the retrieval result, and can be widely applied to the technical field of artificial intelligence.

The invention relates to the technical field of large language models, and discloses a zero-illusion large language model output method and system, and the method comprises the steps: sequentially carrying out the deterministic lexical analysis and standardized mapping of a received original query, and generating a standard term set; locking to-be-reasoned items through hierarchical matching to form a deterministic retrieval result set; executing five rounds of progressive verification processing on the result set to obtain an integrated result set with an LLM verification mark; and outputting a layered zero illusion result through multiple verification and auditing of structure compliance, semantic consistency and knowledge base double recheck. According to the method, probabilistic reasoning is replaced by full-link deterministic operation, so that each code can be traced back to a credible knowledge source, illusion is thoroughly eliminated, absolute reliability and complete traceability of an output result in a professional field are realized by a verifiable deterministic path, and a foundation is laid for zero-illusion application.

The invention discloses a long document intelligent retrieval method and system based on hierarchical analysis and multi-modal fusion, and the method comprises the following steps: carrying out hierarchical analysis on an input long document, and obtaining a physical layout structure, a logic structure and a division and association relationship of multi-modal elements of the long document; respectively extracting text features of the text elements and visual features of the multi-modal elements; performing deep fusion on the text features and the visual features through a cross-modal fusion model to generate a unified multi-modal semantic representation vector; constructing a hybrid index; receiving user query, and performing preliminary retrieval by using the hybrid index to obtain a candidate result set; reordering the candidate result set; and outputting the retrieval result after reordering. Through visual-logic collaborative analysis, gating enhanced bilinear attention fusion and graph attention-based context reordering, a collaborative gain effect is generated in a long document intelligent retrieval task, and the retrieval precision and the user experience are remarkably improved.

The invention discloses a business-adaptive RAG knowledge base rapid question and answer implementation method, relates to the technical field of natural language processing and knowledge base retrieval, and realizes deep fusion and accurate context extraction of fragmented and structured business knowledge by constructing a business enhanced knowledge base and executing multi-stage dynamic retrieval. When an original business document is subjected to intelligent structured analysis and segmentation, a link business entity is identified, rich business attribute metadata is marked, in the process, an unstructured text is converted into an enhanced knowledge fragment carrying clear business semantics, a semantic vector is generated through an embedded model, and on the basis, the semantic vector is subjected to semantic segmentation; according to the method, rapid preliminary screening based on keywords and metadata, fine arrangement based on semantic vectors and association expansion based on a business knowledge graph are sequentially executed, the progressive retrieval strategy ensures that retrieval results are highly related to business contexts, and the accuracy of generated answers is improved.

The invention discloses a data scarcity industrial vertical field-oriented knowledge-guided retrieval enhancement generation method, belongs to the technical field of natural language processing and industrial intelligence crossing, and can improve the retrieval accuracy and generation reliability of a large language model in an industrial scene. According to the method, a'sparse vector + dense vector 'mixed knowledge base is constructed, and general knowledge and industrial field texts are fused; after the model generates an initial text, judging whether external retrieval is needed or not through multi-dimensional evaluation token confidence; extracting attention weights for the low-confidence tokens, and screening key tokens to generate a retrieval query; dynamically adjusting the weight of a retriever based on a BGE-M3 model, and optimizing a retrieval result through reordering; the retrieval knowledge is converted into a context with an index, and a prompt template is constructed to generate a correction value iteration calibration text; and finally, optimizing the text format, and generating a structured response meeting industrial requirements. The method solves the problems of lack of professional knowledge of large language models in the industrial field, shallow retrieval and generation fusion, lack of knowledge calibration mechanisms and the like.

The invention belongs to the technical field of artificial intelligence task planning, and provides a task planning PDDL file automatic generation method based on natural language input, and the method comprises the steps: (10) constructing task environment description words: analyzing environment elements and participants, and designing a diversified scene framework; (20) building a knowledge enhancement fine tuning model: fusing a vector RAG knowledge base retrieval result and pre-training model parameters by the model, and inhibiting logic illusion; (30) PDDL file generation and knowledge constraint: driving the fine tuning model to output a correct PDDL file in combination with scene semantics and knowledge base rules; (40) multi-stage dynamic evaluation: executing a feasibility index verification instruction through a knowledge base rule matching degree; and (50) model closed-loop iterative optimization: updating RAG knowledge base content and model parameters based on execution feedback, and constructing a closed-loop iterative optimization model. The method has the beneficial effects that the instruction logic deviation rate is reduced through RAG knowledge base constraint, the complex scene instruction generation accuracy is improved, and the dynamic decision response time is shortened.

The invention relates to the technical field of document processing, and discloses a document knowledge retrieval method and system, electronic equipment and a storage medium. Based on a language model, synchronously generating abstract contents and a semantic association query set for each text fragment, and aggregating the abstract contents of all text fragments corresponding to the same document to form a fragment abstract set; constructing a multi-path retrieval pool based on the text fragment, the abstract content and the semantic association query set, and respectively constructing a plurality of independent retrieval data sources; the method comprises the following steps: constructing a dual-granularity abstract index system, receiving a query request of a user, executing query routing processing based on dual-granularity abstract index and a hierarchical retrieval process based on a metadata structure, returning retrieval result data, performing fusion calculation and duplicate removal processing on the retrieval result data recalled by multiple paths, and obtaining a retrieval result. And meanwhile, the document abstract short sentence group is injected into a search suggestion function of a search engine. According to the method, the retrieval accuracy can be improved, and the user experience is improved.

The invention discloses an AI knowledge base retrieval method and system based on context awareness and dynamic fusion, belongs to the technical field of artificial intelligence, and aims to solve the technical problems of insufficient context understanding, weak model adaptability and low retrieval precision in a traditional retrieval method. Comprising the steps of collecting and preprocessing multi-dimensional context data; obtaining a context analysis result comprising a query semantic vector, a query intention type, a scene demand weight vector and a user preference model; constructing a plurality of basic retrieval models and registering metadata for each basic retrieval model; on the basis of context analysis results, the fusion weight of each basic retrieval model is calculated through a multi-factor decision algorithm, and the initial retrieval results are fused and sorted; and carrying out iterative optimization on a model algorithm involved in context data analysis and parameters of each basic model based on interaction feedback data of a user on a retrieval result.

The invention provides a data retrieval method, system and device, and relates to the technical field of big data processing, the method comprises the following steps: obtaining a search text of a user; performing data permission assembly according to the search text to obtain a user permission set of the user; according to the user permission set, generating a retrieval filtering condition of the search text; according to the retrieval filtering condition, performing multi-source data search in combination with the search text to obtain an initial retrieval result corresponding to the search text; and performing unauthorized rejection operation on the initial retrieval result to obtain a final retrieval result of the search text. According to the method, the range is defined through permission assembly, the condition focusing target is filtered, the efficiency is improved through pre-matching, and the precision is guaranteed through post-removing, so that the retrieval data volume is compressed to improve the efficiency, and the matching pertinence is enhanced to improve the precision.

The invention discloses an automatic SCADE test case generation method based on a large language model, which comprises the following steps of: analyzing various SCADE related documents to obtain an analysis result; the analysis result is converted into JSON data; performing key information extraction on the JSON data, and performing block processing on the extracted key information to obtain text blocks; vectorizing the text blocks, storing the vectorized text blocks into a vector database, converting a natural language query requirement of a user into a query vector, and performing approximate nearest neighbor search with the query vector in the vector database by utilizing RAG retrieval to obtain a preliminary retrieval result; based on metadata in the preliminary retrieval result, filtering, timeliness sorting and deduplication fusion processing are carried out, and a retrieved context information set is obtained; adopting a dynamic cue word template to construct an enhanced cue word based on the retrieved context information set and the natural language query of the user; and calling a large language model API, inputting an enhanced cue word, and generating a structured test case.

The invention relates to a multi-path RAG result optimization method and device and electronic equipment. The method comprises the steps that a query instruction of a user is received, information related to the query instruction is retrieved from multi-source data through a standardized interface, and a retrieval result is obtained; performing hybrid coding and duplicate removal processing on the retrieval result to form a candidate information pool; based on the candidate information pool, calling a heterogeneous large language model to generate multiple paths of candidate results in parallel; and calling an independent evaluation large language model to perform multi-dimensional quality evaluation on the candidate result of each path, and based on a quality evaluation result, determining an optimal result from the candidate results of all paths and returning the optimal result to the user. The technical problems that the multi-source RAG result screening of a traditional RAG system is single, and a dynamic and multi-dimensional optimization mechanism is lacked are solved.

The invention relates to a query optimization method and device based on agent planning, electronic equipment and a medium. The method comprises the following steps: pre-retrieving a target tree structure index based on a user query request to obtain a pre-retrieval result set; generating a retrieval-reasoning double-stage directed acyclic graph task chain comprising a task type, a retrieval / reasoning target and a dependency relationship; executing each task according to a topological sequence, and obtaining a target text fragment through query rewriting and expansion; and performing multi-hop logical reasoning in combination with the task chain and the target text fragment, and outputting a final answer. Therefore, dynamic task planning of environment perception is carried out by constructing a retrieval-reasoning dual-stage directed acyclic graph task chain and combining a knowledge base tree structure index; the problems that when an existing retrieval enhancement generation system processes a multi-hop question and answer task, logic association understanding ability among retrieval fragments is insufficient, semantic matching precision is low, and a reasoning path is prone to deviation are solved, and the retrieval recall rate and the answer accuracy under a multi-hop question and answer scene are improved.

The invention discloses a model-based data retrieval method and electronic equipment, and relates to the technical field of large model data retrieval, the method comprises the following steps: receiving a retrieval request of a user, the retrieval request carrying request data; according to reward estimation values corresponding to a plurality of strategies pre-stored in the target language model, a target retrieval strategy and initial retrieval parameters corresponding to the target retrieval strategy are determined, and the reward estimation values are used for representing the efficiency of the strategies; retrieving the request data through the target language model according to the target retrieval strategy and the initial retrieval parameters to obtain an initial retrieval result; outputting the initial retrieval result, and determining feedback reward data of the initial retrieval result in response to a feedback operation of the user on the initial retrieval result; determining a target retrieval parameter according to the feedback reward data and the initial retrieval parameter; according to the target retrieval strategy and the target retrieval parameters, the request data are retrieved through the target language model, and a target retrieval result is obtained.

The invention provides a product scheme PPT automatic generation method, system and device and a medium, and belongs to the technical field of artificial intelligence. The method comprises the steps that product demand information is collected, an NLP tool is called for entity recognition and relation extraction, and a demand feature vector is generated; based on a MySQL database, retrieval is performed according to the demand feature vector, a structured record set is returned, and an associated document ID is obtained; based on a Milvus vector database, retrieval is carried out according to the demand feature vectors, and an unstructured document fragment set is returned; filtering the unstructured document fragment set according to the document ID, retrieving the unstructured document fragment set by adopting a mixed strategy of a BM25 algorithm and semantic retrieval, comprehensively sequencing retrieval results through a weighted scoring mechanism, and generating a final content fragment set; and calling a PPT template according to user requirements, and filling the final content fragment set into the PPT template to generate a PPT draft file.

The invention discloses an interactive question and answer task processing method based on an AI large model, and relates to the technical field of AI questions and answers, and the method comprises the following steps: performing correlation screening and function label labeling on high-confidence sub-queries in a retrieval result set, performing weight reduction on low-confidence sub-queries, constructing a cross-source consistency constraint vector, and generating an input data set; based on the input data set, a multi-source evidence consistency verification channel is constructed, entity-by-entity alignment and conflict detection are executed, the credibility interval of answers is calculated, meanwhile, voice emotions are recognized, and an answer data set is generated; and converting the answer data set into multi-modal feedback, monitoring user behaviors in real time, calculating behavior response strength indexes, dynamically adjusting a feedback form, and generating an interactive question and answer data set. The semantic consistency constraint of the multi-modal evidence is realized, and the robustness of answer credibility evaluation in the natural language processing task is improved.

The invention relates to a table information retrieval method and device, computer equipment and a storage medium. The method comprises the following steps: determining a table picture and a table title of a searchable table; based on the cue word template, according to the position information of the table link of the searchable table, the table title, the table abstract generation instruction and the table abstract specification, generating a table cue word; generating a table summary text according to the table prompt word, determining summary fragments, performing vectorization processing on the summary fragments, and determining summary text vectors; storing the table abstract text, a text identifier corresponding to the table abstract text, a link sequence number of a table link, a table picture address of the searchable table and an abstract text vector into a vector database; and obtaining an information retrieval request sent by a user, performing information retrieval on the vector database, and determining an information retrieval result corresponding to the information retrieval request. According to the scheme, the accuracy and integrity of information retrieval can be improved.

The invention discloses a meteorological operation and maintenance intelligent question-answering method based on multi-mode GraphRAG. The method comprises the following steps: S1, a meteorological operation and maintenance intelligent question-answering system divides input data into unstructured data and structured data and processes the unstructured data and the structured data respectively; s2, constructing a knowledge base system composed of a vector database, a relational database and a graph database; s3, analyzing the user query text and dividing question types; s4, selecting a corresponding retrieval strategy according to the question type and starting a processing flow, and finally processing a retrieval result through a fusion and rearrangement module to obtain a knowledge evidence set; and S5, the meteorological operation and maintenance intelligent question-answering system organizes the knowledge evidence set and the user questions into structured cues and submits the structured cues to the large language model for processing to generate structured answers. According to the method, optimization is carried out according to characteristics of the meteorological operation and maintenance field, professional terms and field specific knowledge can be effectively processed, and high-quality field professional services are provided.

The invention relates to the technical field of intelligent retrieval, and discloses an intelligent dynamic K value retrieval optimization system and method, and the system comprises a vector retrieval and storage module, a dynamic K value generation module and an adversarial training optimization module. The system introduces a problem multi-dimensional feature library, wherein multi-dimension of the problem comprises a domain label, a complexity level and a candidate result cardinal number of the problem; the system adjusts the top-k value during retrieval through the dynamic K value generation module, and optimizes the performance of the generator in combination with the adversarial training optimization module. According to the intelligent dynamic K value retrieval optimization system and method, multiple technologies of vector database storage, embedded vector caching, GAN model training, dynamic temperature adjustment, multi-index fusion optimization, incremental updating and hyper-parameter automatic tuning are integrated. The method is characterized in that a k value selection problem is converted into a learnable generation task, a generator can predict the number of optimal retrieval results for a specific problem through adversarial training, and the limitation that the k value is fixed in a traditional retrieval system is broken through.

The invention relates to the crossing field of an artificial intelligence technology and an automotive electronic technology, and discloses a voice interaction question answering method and system based on a vehicle-mounted large model knowledge base, and the method comprises the following steps: receiving a user query text, and executing intention recognition and structured analysis; based on an analysis result, constructing and executing a mixed retrieval request including structured filtering, full-text keyword retrieval and semantic vector retrieval; fusing and reordering the plurality of retrieval result sets obtained respectively to obtain a high-quality final retrieval result set; and finally, constructing a context based on the final retrieval result set and a user query text, and calling a large language model to generate and output a final answer. By introducing a hybrid retrieval and fusion reordering mechanism, comprehensive context information is provided for a large language model, the accuracy and reliability of questions and answers are improved, and the ability of a system to understand and process complex fuzzy queries is enhanced.

The invention discloses a question answering method and device for credit review knowledge based on a large model and a medium, and the method comprises the steps: obtaining a to-be-processed credit review question, extracting a core keyword from the question through a natural language processing algorithm, and converting the question into a deep semantic vector through a semantic vector model; inputting the questions into an intention classification model, generating filtering labels, inputting the questions into a distributed full-text search engine, screening a target knowledge base according to the filtering labels, and performing retrieval in the target knowledge base based on the core keywords and the deep semantic vectors to obtain a preliminary retrieval result set; calculating the score of each preliminary retrieval result in the result set according to the core keyword, the deep semantic vector and the weight factor, and screening out final reference content in a descending order; and on the basis of credit and loan business compliance requirements, risk control rules and question and answer output specifications, the credit and loan prompt words are constructed, and the final reference content is input into the large model to obtain structured credit and loan review answers, so that the question and answer efficiency and accuracy are improved.

A computer system for evaluating information technology (IT) documentation including machine learning models, vector databases, processors, and memories to generate responses for queries associated with one or more of: IT problems, IT solutions, or IT devices. A computer-implemented method involving receiving input data including one or more user queries; generating at least one prompt by interpolating the user queries into a template prompt; processing, via an embedding model, the prompt to generate a retrieval vector corresponding to the prompt; retrieving, by querying a vector database using the retrieval vector as an input parameter, one or more retrieval results; processing, using a trained language model, the prompt, the retrieval results and an assistant prompt including a set of assistant instructions for restricting the output of the trained language model; and displaying, via a graphical user interface, one or more responses corresponding to the user queries from the trained language model.

The invention discloses a two-stage multi-dimensional fusion retrieval enhancement generation method and system, and belongs to the field of large model language processing. Comprising the following steps: dual-stage mixed retrieval: performing BM25 retrieval and vector retrieval in parallel and performing dynamic weighted fusion to realize file-level rough selection and content-level fine selection so as to give consideration to literal matching and semantic understanding; performing intelligent content integrity judgment: performing three-dimensional evaluation on a retrieval result by utilizing a lightweight language model, and triggering supplementary retrieval when information is insufficient; and multi-source information fusion generation: fusing multi-source information by adopting a model based on an attention mechanism, and driving a large language model to generate accurate and traceable answers. The method is also matched with a continuous training optimization method of a small model and an efficient system management architecture based on an asynchronous queue. According to the method, the accuracy and the integrity of the answers are remarkably improved, the retrieval efficiency and the expandability of the system are improved, and the calculation cost is effectively reduced.