125 results about "Knowledge sources" patented technology

The invention discloses a multi-modal fused intelligent question answering and knowledge retrieval method and system, and the method comprises the steps: building a multi-modal data index model oriented to a heterogeneous knowledge source, carrying out the feature mapping of text, image, table, chart, audio and video contents through a unified semantic embedding space, and generating a cross-modal index set; after a query request is received, performing semantic matching and structure matching on the cross-modal index set by using a multi-channel retriever to obtain candidate evidence fragments; and based on an evidence granularity decomposition strategy, performing minimum evidence unit division on text statements, table units, chart data points and multimedia frame contents in the candidate evidence fragments, and establishing a semantic consistency graph among the units. According to the method, high-credibility traceable generation of question and answer results is realized through multi-modal fusion and space-time consistency constraint, and the retrieval precision and interpretation transparency in a complex knowledge scene are remarkably improved.

The invention relates to an enhanced LLM-RAG multi-hop question and answer method based on logic tree reasoning, and belongs to the technical field of new-generation information, and the method comprises the following steps: inputting a multi-hop question and answer question into a computer system; the computer system calls a pre-training large language model LLM, the multi-hop question-answer question is decomposed into a hierarchical logic tree in a recursive mode, and each node of the logic tree comprises a sub-question and a corresponding hypothesis answer; performing image retrieval from a structured knowledge source Wikidata and performing text retrieval from an unstructured knowledge source Wikipedia on the basis of each node sub-question and the hypothesis answer to obtain corresponding evidence; traversing the logic tree, verifying the consistency between the hypothetical answer of each node and the evidence through LLM, if the contradiction exists, reconstructing the corresponding sub-tree, and dynamically correcting the reasoning path; and integrating the verified logic tree node information, and outputting an accurate answer to the multi-hop question and answer question.

The invention provides an agent collaborative question-answering method and system based on context awareness and authority control, and belongs to the technical field of artificial intelligence. The method comprises the following steps: acquiring user query, and triggering a user context manager to generate a session state user context object containing a permission label and preference; the central scheduler performs task decomposition based on the query intention and the context, dynamically selects an intelligent agent through a multi-dimensional scoring model and binds a knowledge source; the authority-aware agent execution engine executes triple authority verification and intermediate result desensitization processing in the whole process, and supports agent fault tolerance and audit log recording. The method realizes unification of knowledge security and access flexibility, supports multi-scene adaptive question and answer, and has good expandability and fault-tolerant capability.

The invention relates to the technical field of artificial intelligence, and discloses a professional knowledge question and answer intelligent customer service system based on a self-optimization mechanism. The system comprises a user intention analysis module, a knowledge processing module, a self-optimization learning module and an interactive presentation module. A semantic understanding unit of the user intention analysis module generates a user intention signal; after the knowledge processing module receives the signal, a knowledge retrieval unit outputs related knowledge fragments and confidence, an answer generation unit forms candidate answers, and a quality evaluation unit determines an optimal answer according to the confidence; in the self-optimization learning module, a feedback analysis unit adjusts answer generation parameters according to user interaction data, a strategy adjustment unit optimizes a retrieval strategy in combination with an optimal answer and a historical dialogue, and a knowledge updating unit updates a knowledge base depending on an external knowledge source; and a multi-round dialogue management unit of the interactive presentation module adjusts a dialogue process, and a visual presentation unit outputs a natural language text and collects user feedback to a feedback analysis unit.

The invention provides a private deployment-oriented multi-modal AI knowledge base construction system. The private deployment-oriented multi-modal AI knowledge base construction system comprises a knowledge storage module, an intelligent document loading module, a document partitioning engine module, a data enhancement engine module, a multi-language semantic vector alignment module and a private deployment module, the knowledge storage module comprises a knowledge authority management sub-module and a knowledge source management sub-module, the knowledge authority management sub-module is used for managing and storing knowledge from different sources, and the knowledge source management sub-module is used for managing electronic documents and multimedia documents; according to the method, intelligent identification, partitioning, vectorization and source file storage can be carried out on different types of electronic files, knowledge graph construction is carried out for specific fields, semantic relevance between texts and topics and key entities is fully considered, the method has wider applicability, higher robustness and controllability, the data leakage risk is effectively reduced, and the method is suitable for popularization and application. The method is suitable for enterprise sensitive data protection and personal user elastic computing power requirements.

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 provides a large model retrieval enhancement generation method fused with a multi-source knowledge base, and relates to the technical field of large model retrieval enhancement generation. The method comprises the steps that firstly, user query is analyzed, multiple implied independent retrieval intentions are recognized and separated, and a knowledge source identifier is distributed to each retrieval intention subtask based on a mapping rule; according to the method, related evidence fragments are acquired in parallel from a heterogeneous knowledge source library in a targeted manner, multi-level relation analysis is further performed on the original evidence fragments by introducing a cross validation engine, so that a structured evidence conflict graph is constructed, and then the evidence conflict graph is intelligently processed by applying a game theory excitation algorithm. The method comprises the following steps: obtaining an evidence set with internal consistency, forming a guide signal through structured analysis, and finally, inputting original user query, the synthesized evidence set and the guide signal into a large language model in combination with a pre-constructed cue word template to generate a final answer under constraint, so as to realize retrieval enhancement generation fused with multi-source knowledge base retrieval.

The invention belongs to the technical field of knowledge Token generation and traceability, and discloses a knowledge Token generation and traceability method based on structured metadata, which is characterized in that a current knowledge scene is quickly matched and a core field is dynamically adjusted through few-sample learning in combination with various technical models and a preset industry field model library; a mixed semantic segmentation mode is adopted, knowledge fragments are completely segmented according to business logic, then a refined element extraction strategy is matched, three types of key elements including entities, relationships and clauses are accurately extracted, and errors are reduced through multi-dimensional verification; originally dispersed PDF, audio and video and other unstructured knowledge can be converted into standardized assets with anchoring information. Meanwhile, a multi-dimensional verification process is designed, comprehensive verification is carried out from a data source, access permission and logic consistency to a health state, it is ensured that a knowledge source can be checked, authenticity can be distinguished, and the information distortion risk during large model calling is reduced.

The invention discloses an automatic generation method for health science popularization articles, and belongs to the crossing field of computer technology and health science popularization. The method comprises the following steps that S1, health authority information is collected through a multi-source crawler, and an original material pool is obtained through Sension-BERT vectorization duplicate removal and BioBERT medical entity labeling; s2, classifying nine types of health themes by using a RoBERTa fine tuning model; s3, multiple AI model supplementary materials are made into a structured package; s4, predetermining optimal selection questions in combination with AI three-dimensional scoring and editing; s5, calling the multi-dimensional knowledge base to generate a professional knowledge packet; s6, generating an outline and performing three-dimensional five-score system auditing of knowledge point fullness and the like; s7, the LLM expands and writes the text and marks knowledge sources; s8, performing double-stage auditing; s9, anthropomorphic draft moistening; s10, intelligently illustrating the picture; s11, when the matching degree is smaller than a preset threshold value, automatically updating the knowledge base; and S12, integrating and outputting. According to the method, the health science popularization article is generated, the manual participation time is shortened to be within 30 minutes, the medical error rate is reduced to be below 5%, the daily output of a 10-person team is improved by 3-5 times, the annual human cost is reduced by 60% or above, and the large-scale and high-quality science popularization requirements are met.

The invention relates to the field of deep learning, and particularly discloses an intelligent management method and system for equipment operation and maintenance based on artificial intelligence, and the method comprises the steps: carrying out the parallel execution of sparse and dense retrieval to guarantee the comprehensiveness of recall, carrying out the interaction analysis of deep semantic and structured entity features of user query through a self-adaptive weight network, and carrying out the calculation of the deep semantic and structured entity features. An optimal and dynamic fusion weight is generated in real time for each query. According to the weight, the dependency degree of current query on keyword matching and the dependency degree of current query on semantic understanding can be accurately judged, and intelligent dynamic weighting and rearrangement are carried out on two retrieval results according to the dependency degree and the dependency degree, so that accurate candidate document screening with one query and one strategy is realized. The dynamic fusion mechanism fundamentally solves the problem that the static weight cannot adapt to query diversity, and ensures that the knowledge source provided for the downstream generation model is always highly related to the current query intention, so that the accuracy and reliability of the finally generated answer are remarkably improved.

A system and method for automated domain advertising utilizing artificial intelligence is provided. The system includes a processor and memory in communication with the processor. The memory includes a user interface module, a domain advertising (DA) module, an artificial intelligence (AI) module, a posting module, and a domain parking module. The system receives the domain name from a user. The AI module generates the domain advertising suggestion based on the domain name. The domain advertising suggestion is provided to the user and may be posted to social media or a domain parking website. The AI module includes a large language model (LLM) and a retrieval-augmented generation (RAG) module. The RAG module includes a knowledge base to retrieve a knowledge-based response from an external knowledge source, an expert module to retrieve an expert response from a specialized domain model, and a ranker to rank the knowledge-based response and the expert response.

The invention discloses a field event element extraction method and system based on prompt learning and knowledge completion, and the method comprises the steps: converting an extraction task into a constrained generation problem through structured prompt, achieving the stable extraction under a small number of labeled samples through the existing knowledge prior of a large language model, reducing the dependence on large-scale labeled data, and improving the extraction efficiency. And the cross-domain adaptive capacity is improved. A vectorization knowledge retrieval mechanism is introduced, related evidences are obtained from an external knowledge source, elements which are not clearly expressed in a text are complemented, the integrity of information is enhanced, and the accuracy and credibility of a result are improved through knowledge verification. And through a self-adaptive fusion mechanism of prompt and knowledge, the model can flexibly balance text context and external knowledge, and the robustness of complex context and fuzzy expression is improved. And finally, standardized and structured element data are output, a convenient interface is provided for downstream event atlas construction and analysis tasks, and the automation degree of domain knowledge processing and the system integration efficiency are improved.

The invention discloses an aspect-level sentiment analysis optimization method based on multivariate external knowledge fusion, which relates to the technical field of sentiment analysis optimization, and comprises the following steps: constructing a multivariate external knowledge source comprising a Chinese sentiment dictionary, a domain knowledge graph and a user comment prior mode library; the sentiment module is used for providing vocabulary-level sentiment polarity, entity attribute relations and high-frequency evaluation semantic modes; semantic coding is performed on the input text and the specified aspect words to generate context semantic representation, and global semantic features and local position features are extracted in combination with aspect word position information; based on a semantic coding result, converting the multivariate external knowledge sources into structured knowledge representations, and dynamically adjusting contribution weights of various types of knowledge through a gating fusion mechanism to generate fused knowledge representations; performing dependency syntactic analysis on the input text, constructing an original syntactic structure, and calculating the correlation strength of each grammatical component and aspect words in combination with context semantic representation; and pruning the original syntactic structure according to the correlation intensity.

According to the abnormal information monitoring and processing method based on the large model and the intelligent agent, system operation data and log information are collected in real time, abnormity is detected based on threshold rules, automatic discovery of abnormal events is achieved, and manual intervention is not needed; after the exception is detected, the exception event object is constructed and the context semantic analysis is performed, so that the exception category, the trigger module and the operation environment parameters can be accurately extracted, and the accurate positioning of the complex exception is realized. Cue words are generated based on the semantic enhancement model, and a scheme retrieval request is constructed, so that the retrieval process better conforms to semantic logic, and the accuracy and correlation of scheme matching are improved. Through semantic similarity calculation and an external knowledge source extension mechanism, the system can efficiently retrieve candidate solutions inside and outside a knowledge base, and determines an optimal solution based on credibility scores, so that the intelligence and reliability of decision processing are remarkably improved.

The invention provides an industry knowledge base dynamic construction method and system based on multi-source heterogeneous data, and relates to the technical field of data processing.The method comprises the steps that target industry multi-source heterogeneous data, user historical behavior data, a natural language question text and a processing text are collected to generate a query data stream; mapping the user question into a query semantic vector, converting the user interaction behavior into a behavior feature vector, and generating a joint query vector; industry knowledge units are automatically extracted from multi-source heterogeneous data and external knowledge sources, and a dynamic knowledge topological graph is constructed; establishing a dynamic knowledge routing table, taking the joint query vector as a routing request, matching a target knowledge node, recording a query path and caching an association rule; and counting and querying path node transition probability to identify a hotspot path, mining a deep association rule, and reversely adjusting a knowledge topological graph structure, so that user intentions can be accurately matched, an industry knowledge base can be dynamically optimized, and knowledge retrieval and association efficiency can be improved.

Embodiments described herein provide systems and methods for retrieval augmented generation. A neural network based language model may be provided a question as a user input. Based on the user input, semantically diverse queries may be generated for retrieval from diverse data sources. For example, a structured data source (e.g., database or knowledge base) and unstructured data (e.g., text articles) may be used to retrieve information relevant to the user input. The retrieve information may be ranked so that the most relevant information is used by the language model in generating an answer to the question in the user input. A non-retrieval based answer generated by the language model may be utilized in some embodiments in generating the final answer.

The invention discloses a typhoid theory teaching method based on a knowledge graph, and belongs to the technical field of typhoid theory teaching. The method comprises the steps of multi-version data acquisition and standardization, ancient Chinese context semantic unit construction, term semantic disambiguation based on comparative learning, semantic conflict detection based on consistency constraint, multi-source information fusion and map optimization and teaching path generation. Through ancient Chinese context semantic modeling and comparative learning, precise disambiguation of traditional Chinese medicine ancient book terms is realized, and knowledge errors caused by polysemy of one word are reduced; through an automatic detection mechanism of logic consistency constraint, the quality and reliability of the knowledge graph are improved; a version traceability and belief propagation optimization strategy is adopted, so that the finally constructed knowledge graph not only can integrate multi-version essence, but also can clearly trace knowledge sources and evaluate knowledge credibility; the teaching path chain generated based on the atlas can dynamically simulate the whole process of traditional Chinese medicine syndrome differentiation treatment, and the clinical thinking training effect is remarkably improved.

The invention relates to the technical field of digital data processing, in particular to an automatic session quality detection method applied to a customer service scene. The detection method comprises the following steps: model training: selecting an LLaMA pre-training large language model as a core engine, and guiding training by means of cue words with session key element extraction and semantic consistency comparison functions; judging the format of the target session data to be subjected to quality detection, and converting the voice format into a text format; performing key element extraction and comparative analysis on the session data in the text format by means of the trained model; consistency comparison is carried out on the robot reply and a corresponding standard reply in a knowledge base; verifying whether business data contained in the robot reply is consistent with real data of a back-end business system; and generating a detailed report of the quality inspection result. Defects are accurately recognized, understood and expressed by means of the large model reasoning ability, and robot reply and a semantic verification closed loop of a knowledge source are achieved.

The invention discloses a digital operation management method and device based on industrial Internet and a medium, and relates to the technical field of operation management, and the method comprises the steps: collecting multi-source heterogeneous data, extracting the causal relationship between the internal process parameters and performance indicators of each enterprise through a causal discovery algorithm, and obtaining the data of the multi-source heterogeneous data; coding the causal relationship into a local knowledge vector by using a deep learning model; sorting the candidate knowledge vectors in a descending order according to the expected efficiency gain value, and collecting real efficiency data after the target enterprise actually applies the candidate knowledge vectors; and based on the adjusted efficiency weight, calculating the contribution degree of each enterprise in the global knowledge value network through a Shapley value algorithm, generating an incentive distribution scheme, and triggering the execution of the smart contract. According to the method, the contribution degree of each enterprise in the global knowledge value network is calculated through the Shapley value algorithm, and the marginal value contribution of each knowledge source in cross-enterprise collaboration can be accurately quantified based on the dynamic efficiency weight of the knowledge node.

The invention discloses an advertisement material intelligent optimization system and method based on knowledge extraction and knowledge fusion, and the method comprises the following steps: S1, obtaining and preprocessing a to-be-optimized advertisement material, and collecting an external knowledge source; s2, combined semantic analysis is executed, and cross-modal semantic alignment is carried out; s3, extracting entity units and corresponding semantic attributes, performing type labeling on the entity units, and establishing a semantic relation structure; s4, performing semantic fusion with an external knowledge source, calculating similarity between semantic vectors, performing entity mapping, performing normalization, and correcting conflict boundaries and attribute value divergence; s5, extracting a target expression fragment, and performing optimization according to a preset optimization rule set; and S6, verifying and detecting the optimized advertisement material, and outputting an optimization result. According to the method, deep understanding and cross-modal fusion optimization of advertisement materials can be realized, and the accuracy, attraction and propagation effect of advertisement semantic expression are improved.

The embodiment of the invention provides an automatic driving cloud simulation platform test method and device, and the method comprises the steps: obtaining multi-source data of an automatic driving cloud simulation platform, carrying out the classification and structural extraction of the multi-source data, obtaining corresponding knowledge fragments, and storing the knowledge fragments into a knowledge source database, according to the types of the knowledge fragments, vectorizing the knowledge fragments by adopting different language models to obtain a document vector set, a code vector set and a scene parameter vector set; receiving and vectorizing a test demand described by a natural language, performing vector retrieval in the document vector set and the scene parameter vector set according to a retrieval enhancement generation algorithm, and inputting a retrieval result into the artificial intelligence model to understand the scene demand of the test, so as to obtain test case logic output by the model, the test code template is retrieved in the code vector set according to the test case logic, the input parameters are injected into the test code template to obtain the test code for testing, and the efficiency and accuracy of automatic driving cloud simulation platform testing can be improved.

The invention provides a knowledge graph and deep learning fused research course automatic generation method and system. The method comprises the steps of receiving event stream data from a plurality of knowledge sources; performing source identification and theme clustering processing on the event stream, and generating a standardized update demand unit; writing the update demand unit into a temporary change buffer area outside the atlas; inputting the knowledge fragments in the change buffer area into a cross-source semantic alignment model, and executing entity mapping and relation prediction operation; generating a multi-scale updating plan based on the influence hierarchy of the candidate nodes and the edge set, and executing a graph updating operation; based on the updated knowledge graph, determining a learning target and ability dimension of the research course, a required core knowledge point sequence and an association relationship; inputting the core knowledge point sequence into a course structure generation model, and outputting a preliminary study course outline; and generating a final study course by using the course outline. According to the method, the research course can be automatically generated, and the timeliness and accuracy of the course are ensured.

The present disclosure generally relates to evaluating and enhancing LLM responses. In some implementations, a system includes multiple language models with different specialized roles that work together to improve response reliability and transparency. A responder model can generate initial responses to user queries, providing diverse perspectives on the same input. An evaluator model can assess and combines responses from the responder models into an accurate and reliable output. A reporter model can generate summaries and alerts about response quality and confidence levels, providing transparency to users about the decision-making process. An artificial intelligence (AI) engine can manage the flow of information between the different models, orchestrating their interactions and ensuring proper sequencing of operations. A retrieval system can provide additional context from external knowledge sources, allowing the system to generate accurate and well-informed responses.

The invention provides an industrial product parameter extraction method based on domain knowledge learning, and the method comprises the steps: 1, collecting a plurality of technical documents of a preset industrial product field as knowledge sources, and constructing a structured domain knowledge model through a man-machine cooperation mode; 2, based on the domain knowledge model, generating a training data set with entity type labels, and training one or more deep learning extraction models by using the data set to enable the extraction models to learn parameter knowledge of the industrial product field; 3, receiving a new to-be-processed digital technology document, and processing the document by adopting the trained deep learning extraction model so as to identify and extract structured data corresponding to parameter entities in the domain knowledge model; and 4, storing the extracted structured data into a database, and selectively utilizing the data to carry out iterative updating and enhancement on the domain knowledge model.

The invention relates to the technical field of computers, and discloses a question and answer method and device, electronic equipment, a storage medium and a program product. The method comprises the following steps: acquiring query content, which is proposed by a requester in a current dialogue and aims at a target object; analyzing the query content based on a preset analysis rule, and determining a query intention of the requester for the target object; based on the query intention, determining a target knowledge source corresponding to the query intention from the plurality of candidate knowledge sources; and retrieving the target knowledge source based on the query content to obtain target reply content corresponding to the query content. By implementing the technical scheme of the invention, the appropriate target knowledge source can be dynamically selected from the plurality of candidate knowledge sources for retrieval according to the query intention of the user, and the targeted reply is generated, so that the accuracy and efficiency of question answering and the user experience are effectively improved.

The application relates to a knowledge-driven multi-agent cooperative reactor scheme demonstration system and a design method, medium and equipment thereof, the system comprising: a knowledge internalization unit for converting unstructured and semi-structured documents into a dynamic knowledge source which can be queried and utilized by a model in real time; a knowledge structuring unit for extracting core entities, relationships and attributes from the knowledge source and constructing a professional knowledge graph of a reactor design field; and a knowledge application unit for constructing intelligent agents and a collaborative working mechanism of multi-professional intelligent agents based on the knowledge source and the knowledge graph, and constructing an intelligent question and answer interface. Through automatic knowledge management and intelligent agent collaborative work, the application significantly reduces the time and effort of manual intervention and improves the efficiency of reactor scheme demonstration.

The present application relates to the technical field of artificial intelligence, and more particularly to an AIGC content generation method and system based on a brand knowledge base. The method comprises: obtaining a content generation request carrying a brand identifier and a domain identifier; extracting brand knowledge units from multiple brand knowledge bases according to the brand identifier; identifying the knowledge coverage density of each knowledge base on the domain identifier through domain coverage range annotation; marking the knowledge sources with a coverage density lower than a preset benchmark as sparse knowledge sources, and marking the knowledge sources not lower than the benchmark as dense knowledge sources; taking the knowledge fragment with the nearest neighbor semantics to the domain identifier in the dense knowledge source as an anchor point, performing cross-knowledge migration completion on the sparse knowledge sources, and generating the completed sparse knowledge units by remapping the style of the anchor point brand tone constraint according to the audience portrait of the sparse knowledge sources; and outputting a content generation strategy bound with the brand identifier and the domain identifier.

The application discloses a kind of data knowledge extraction method, system, equipment and storage medium based on large language model, method includes: the original data obtained is preprocessed;Feature is extracted from original data, select corresponding task feature, remove the feature of high correlation by calculating the correlation coefficient matrix between features;Select the large language model of task, capture general language and knowledge, fine-tune the large language model on the labeled data set of specific field or task;Based on large language model inference and prompt word engineering, knowledge source information is extracted using a secondary block iteration method, and the context understanding ability of the large language model after fine-tuning is used to integrate the information extracted from different data sources;Define evaluation criteria, analyze the error cases of the large language model, identify the improvement direction, and adjust the model parameters and training data according to the evaluation results and error analysis.The application has the advantages of improving the efficiency and accuracy of the knowledge extraction method.

This application provides a testing method and apparatus for an autonomous driving cloud simulation platform. The method includes: acquiring multi-source data from the autonomous driving cloud simulation platform; classifying and structurally extracting the multi-source data to obtain corresponding knowledge fragments and storing them in a knowledge source database; vectorizing the knowledge fragments according to their types using different language models to obtain document vector sets, code vector sets, and scene parameter vector sets; receiving test requirements described in natural language and quantizing them; performing vector retrieval in the document vector set and scene parameter vector set using a retrieval enhancement generation algorithm; inputting the retrieval results into an artificial intelligence model to understand the test scenario requirements; obtaining test case logic output by the model; retrieving test code templates from the code vector set based on the test case logic; and injecting input parameters into the test code templates to obtain test code for testing. This application can improve the efficiency and accuracy of testing autonomous driving cloud simulation platforms.

The invention discloses a large model reasoning stability enhancement method and system in a text scene, and relates to the technical field of knowledge enhancement deep learning, and the method comprises the steps: building a structured text knowledge graph based on a text knowledge base, and generating a dynamic text knowledge embedding matrix through a graph attention network; inserting a text knowledge gating cross attention module into a decoder selection layer of the pre-trained large model, taking the text knowledge gating cross attention module as an external knowledge source, obtaining a knowledge enhanced hidden state after gating fusion, and constructing a transformation model; semantic equivalent perturbation is carried out on an input text to obtain a perturbation sample, the perturbation sample is input into the transformation model in parallel to obtain extraction probability distribution, and divergence and gradient direction consistency loss between two distributions are calculated; and combining cross entropy loss and gradient direction consistency loss to train and transform the model, and updating parameters to convergence to obtain a final large model. According to the method, the fact consistency of output can be improved, common optimization of knowledge guidance and stability constraint is realized, and the result is accurate and reliable.