30 results about "Explicit knowledge" patented technology

The invention relates to the technical field of intelligent water conservancy, and discloses and provides a hydraulic structure large language model fine tuning and graph retrieval enhanced generation method based on knowledge distillation, and the method specifically comprises the steps: obtaining a hydraulic structure instruction fine tuning data set in combination with a knowledge distillation method; performing supervised training on the large language model by using a low-rank decomposition technology to obtain a hydraulic structure large language model; identifying a question intention of a user by using a hydraulic structure large language model, and converting a question statement into a knowledge graph query language; and enhancing the professional and credibility of water conservancy generated by the answer of the large model of the hydraulic structure by using a knowledge graph retrieval result. According to the method, the hydraulic structure knowledge is embedded into the large language model, the large language model customized for the hydraulic structure field is obtained, and the explicit knowledge stored in the water conservancy knowledge graph and the implicit knowledge of the large language model are effectively combined, so that the answer is more accurate and traceable, and an effective technical means is provided for intelligent water conservancy construction.

The invention provides an industrial chain data analysis method and system based on a knowledge graph and intelligent risk control, and relates to an intelligent risk control technology, and the method comprises the steps: constructing an industrial chain dominant knowledge graph which comprises a main body node and a dominant association edge, recognizing implicit association, forming an enhanced knowledge graph, dividing levels according to a centrality metric value and a control path number, and obtaining an enhanced knowledge graph; and a differential risk propagation coefficient is set, an abnormal change mode is identified in combination with time sequence correlation analysis, and finally a comprehensive risk assessment result is generated. The risk association hidden in the industrial chain can be effectively found, and the risk identification accuracy and the early warning timeliness are improved.

The invention relates to a medical diagnosis method and system based on multi-modal retrieval enhancement and guide guidance. The method comprises the steps that text information including reports and / or electronic health records and medical image information are obtained; encoding the medical image information and the text information by using an image encoder and a text encoder respectively to obtain visual features and text features; respectively utilizing a guide branch decoder and a label branch decoder, taking diagnosis guide features and disease type labels of samples in the training stage as queries of a Transform structure, taking splicing features obtained by splicing text features and visual features as keys and values, and decoding to obtain first prediction probability distribution and second prediction probability distribution of disease types, so as to obtain first prediction probability distribution and second prediction probability distribution of the disease types; and a final disease prediction result is obtained. According to the method, disease specificity knowledge is dynamically retrieved based on a multi-source medical knowledge base, redundancy and noise are removed through a large language model, a standardized and structured diagnosis guide is generated, and explicit guidance of the knowledge is achieved.

The invention relates to the technical field of artificial intelligence, in particular to an intelligent customer service dialogue generation system and method based on a knowledge graph, and the system comprises a language model, a knowledge graph module, a query module, a matching module, a reasoning module, a strategy generator, a reinforcement learning unit and a language generation module. The module comprises a knowledge subgraph isomorphic mapping unit, a graph entropy gradient reasoning unit and a Markov decision optimization unit, and the three units form a progressive processing link. The knowledge sub-graph isomorphic mapping unit maps the retrieval sub-graph to an isomorphic sub-graph in a knowledge graph topological space; the map entropy gradient reasoning unit constructs an entropy gradient field to realize reasoning path optimization from a high-entropy region to a low-entropy region; the Markov decision optimization unit models an inference process into a Markov decision process, a reinforcement learning method optimizes an inference strategy, and effective inference from dominant knowledge to implicit knowledge is realized in combination with a knowledge graph and a deep reinforcement learning technology.

The invention provides a circuit design optimization method based on knowledge rule induction and implicit knowledge mining. The method comprises the following steps: 1) classifying and sorting design rules; 2) extraction and modeling of dominant design rules; 3) mining and structured expression of the implicit design rule; 4) constructing and optimizing a knowledge document; 5) establishing and applying a design rule base; and 6) reasoning and verifying the key links in the electrical design process. According to the method, explicit knowledge such as electrical design specifications and periodical papers is systematically summarized, and implicit knowledge such as design experience and skills is extracted in combination with a data mining and knowledge reasoning technology, so that systematic management of key knowledge information in the electrical design process is realized, the intelligent level of an electrical design software platform can be effectively improved, and the design efficiency is improved. The knowledge sharing and multiplexing capability in the design process is enhanced, so that the cable wiring scheme is optimized, and the reliability of the design result in the aspects of safety, economy, scheme consistency and the like is improved.

The application discloses a kind of self-adapting intelligent teaching content recommendation system, it is related to intelligent teaching technical field, including data acquisition analysis module: collection student in the learning process multiple types of data, including student's learning behavior data, physiological signal, environmental data and learning achievement cognitive feedback data, after acquisition, multiple types of data are preprocessed with noise reduction, standardization and analysis feature extraction, generate multimodal data;The application is analyzed by constructing knowledge point, error type, three-dimensional knowledge graph of thinking path and time sequence convolution network TCN, not only can identify dominant knowledge defects, but also can diagnose implicit knowledge leaks, so that students can accurately know their own knowledge system leaks, targeted to check and fill the gap and strengthen training;By extracting learning interest information from multimodal data, an explicit and implicit multidimensional interest model is established, and the teaching content is filtered in combination with the knowledge gap diagnosis result.

This invention discloses a system for converting tacit knowledge into explicit knowledge based on knowledge graphs and large models. It includes a knowledge system ontology modeling and management module for constructing and maintaining the knowledge system ontology model in large model applications; a data access and knowledge graph construction module for mapping and loading multi-source heterogeneous data onto the knowledge system ontology; an ontology virtual business environment engine module; a tacit knowledge mining and explicit knowledge generation module that, in a virtual environment, uses graph analysis and large model reasoning to mine tacit knowledge and generate explicit knowledge; and a business strategy planning and workflow orchestration module that transforms explicit knowledge into executable business strategy plans and workflow orchestrations. Through the knowledge system ontology modeling and management module, a systematic knowledge framework is constructed, achieving systematic accumulation of enterprise tacit knowledge, effectively preventing knowledge loss, and improving knowledge reusability and scalability.

The application provides a method and system for dynamic modeling and closed-loop iteration of individual knowledge and cognitive patterns of scientific researchers, relating to the fields of computer technology, artificial intelligence and information processing technology. The method includes: obtaining explicit knowledge data and interaction behavior data of scientific researchers; performing enhancement and structuralization processing; constructing objective knowledge base information based on explicit knowledge data; constructing historical behavior graph based on interaction behavior data and performing metacognition analysis to extract cognitive pattern information; constructing a structured context assembler dominated by cognitive patterns; in response to a scientific task request, calling the assembler to generate a task-level scientific portrait structured context, and inputting it together with the task request into a large language model to generate personalized output; collecting multi-modal feedback data and updating the cognitive pattern to realize continuous iteration. Through deep twin modeling and closed-loop optimization of objective knowledge and subjective cognition, the application significantly improves the personalization and intelligence level of scientific research assistance.

The present application relates to the technical field of multi-modal dialogue, and especially relates to a knowledge extraction and mining method for a smart home multi-modal dialogue system, which comprises the following steps: S1. smart home multi-modal dialogue system input data acquisition; S2. multi-modal explicit knowledge graph construction; S3. multi-modal explicit knowledge representation learning; S4. multi-modal explicit knowledge reasoning and relationship completion; and S5. multi-modal implicit knowledge mining and representation learning. The present application extracts and aligns knowledge from massive unstructured texts, pictures and videos, constructs a multi-modal knowledge graph, performs representation learning on the graph based on semantic meta paths, extracts rich subgraph information through graph convolution operation to complete knowledge reasoning, and further completes the missing relationships. In addition, based on a large-scale pre-training model, a pre-training task for mining implicit knowledge is designed to guide the model to learn implicit knowledge such as emotion, which can improve the logical reasoning ability of the dialogue system.

The invention discloses a task scheduling method and system for a distributed optical fiber sound wave sensing system, and relates to the technical field of distributed optical fiber sound wave sensing, and the method comprises the steps: collecting a distributed optical fiber sound wave sensing signal; establishing a target function for full waveform inversion of the sensing signal through forward modeling, and improving an adjoint state source equation according to an axial component residual error in the target function; obtaining an inverted distributed optical fiber sound wave sensing signal by using the improved adjoint state source equation; determining a mapping relation between the inverted distributed optical fiber sound wave sensing signal and the corresponding physical quantity; performing explicit knowledge migration of the main task and the subtasks according to the mapping relation to realize multi-task scheduling; according to the method, the problem of low signal-to-noise ratio of sensing signals can be effectively solved, and the multi-task scheduling quality of the distributed optical fiber acoustic wave sensing system is improved.

The invention discloses an air and space big data intelligent data management system and method based on four-database collaboration and knowledge generation. A core layer adopts a temporary storage database, a controlled database, a product database and a knowledge database to form a four-database architecture; and the service layer adopts a data governance engine, a productization engine, a knowledge generation engine and an intelligent enabling engine. The system combines a data governance engine and a productization engine to carry out collaborative linkage to carry out data governance. Depositing knowledge assets into a knowledge base by relying on a knowledge generation engine; performing intelligent enabling application of each link of the data governance process by means of an intelligent enabling engine; and an intelligent enabling application effect is fed back, and knowledge assets in the knowledge base are optimized. According to the intelligent closed-loop data governance system, data governance implicit knowledge is converted into dominant knowledge and reused, the self-adaptability and the intelligent level of the data governance process are improved through knowledge reverse enabling, and then an intelligent closed-loop data governance system capable of self-learning, self-optimizing and continuously adapting to service requirements is formed.

The invention relates to the field of computer technology and artificial intelligence technology, and discloses an authorized purchase-oriented risk management and control system and method, and the system comprises a knowledge base construction module which is used for collecting multi-source heterogeneous data related to authorized purchase, and constructing a knowledge base; the intelligent template module is configured to generate a service quantity list and a technical parameter template based on the knowledge base, and perform real-time semantic risk identification and early warning when a user inputs a purchase demand text; and the risk assessment and price limit module is configured to receive a to-be-approved authorized purchase demand, analyze the authorized purchase demand based on the knowledge base and the semantic risk identification result, and output a comprehensive risk index and a dynamic maximum price limit. Various scattered and heterogeneous data generated in a purchase whole chain are processed and precipitated into a structured dynamic knowledge base, and implicit knowledge originally stored in personal experience or scattered documents is converted into enterprise-level, computable and inheritable dominant knowledge assets.

The invention discloses a gray measurement expert knowledge generalization model construction method, which comprises the following steps of: constructing a dual-task deep learning model sharing an encoder by synchronously acquiring quantitative scores and visual focus data of experts, so that the dual-task deep learning model learns and predicts expert scores and regions of interest at the same time; the method comprises the following steps of: establishing a deep model, migrating implicit knowledge of the deep model to an interpretable shallow model by utilizing knowledge distillation, extracting dominant and readable discrimination rules, and finally establishing a man-machine collaborative closed loop: submitting highly uncertain samples of the rules and the model to an expert for auditing and correcting, and utilizing feedback iteration to finely adjust the model to realize continuous optimization of performance and interpretability. According to the method, recessive and intuitive knowledge and dominant knowledge of experts are effectively fused, the generalization performance, robustness and interpretability of the model are remarkably improved, and a solid technical support is provided for high-precision and high-reliability automatic gray scale measurement and calculation.

The invention discloses a multi-task Bayesian optimization self-adaptive knowledge migration method, which comprises the following steps of: capturing a non-linear relationship among data sets by utilizing a kernel automatic coding mechanism so as to measure the similarity among tasks; the priority of auxiliary task selection is dynamically adjusted through a heuristic rule, selective cooperation between tasks is ensured, and interference is reduced. Bayesian optimization modeling and black box function optimization are used to reduce the number of times of task real evaluation. According to the method, an explicit knowledge migration mechanism is introduced, the nonlinear mapping matrix is used for converting the optimal solution of the auxiliary task, efficient knowledge migration and collaborative optimization among the multiple related tasks are achieved, and the optimization efficiency and the resource utilization rate are remarkably improved.

The invention discloses a multi-agent collaborative chemical knowledge explicit distillation method and system based on a cognitive apprentice mechanism, and relates to the technical field of artificial intelligence and machine learning, and the method comprises the steps: employing an explicit knowledge distillation framework to process collected chemical problems, and generating a knowledge list; the knowledge list is evaluated by applying a two-dimensional arbitration evaluation mechanism, whether an evaluation result is qualified or not is judged through a decision committee, and if yes, the knowledge list is transmitted to a learner decision maker; if not, triggering a relearning teacher intelligent agent, generating a knowledge point supplement list, and transmitting the knowledge point supplement list to a learner decision maker; verifying whether the receiving result meets a preset knowledge quality standard or not through a learner decision maker, and if yes, storing the receiving result into a preset knowledge base; and if not, triggering the explicit knowledge distillation framework to carry out the next round of explicit distillation process. According to the method, an explicit knowledge distillation framework is adopted to simulate cognitive apprentice logic layering to process chemical problems, and knowledge interpretability and subject preciseness are greatly improved.

The invention discloses a gas operation expert experience precipitation management method, belongs to the technical field of knowledge management and artificial intelligence, and aims to solve the technical problems that traditional expert knowledge is difficult to solidify, slow to update and low in utilization rate. The method comprises the following steps: analyzing recessive and dominant knowledge of front-line employees and field experts into structured knowledge units through a multi-modal acquisition and natural language processing technology, constructing a fuel gas field knowledge graph, carrying out networked storage and association on the knowledge units, and carrying out intelligent application service. A knowledge graph and a first-line business scene are deeply fused, a structured user behavior sequence is accurately captured in the process, and a dynamic optimization model is established based on the user behavior sequence. Static expert experience can be converted into a dynamic knowledge system capable of interacting with services and continuously learning, the reuse value and decision support efficiency of knowledge assets are remarkably improved, and dependence on core experts is reduced.

The application belongs to the technical field of brain-computer interface, and relates to a classroom teaching quality evaluation method, system and terminal based on a multi-person brain-computer interface. The evaluation method comprises the following steps: in a pre-class stage, first, second and third evaluation parameters are calculated to obtain first and second grade thresholds; in a class stage, whether the first, second and third evaluation parameters of each student are greater than or equal to the corresponding first grade threshold is judged, if yes, A-level teaching quality is output, if not, whether the first, second and third evaluation parameters are greater than or equal to the corresponding second grade threshold is judged, if yes, B-level teaching quality is output, otherwise, C-level teaching quality is output; and the teaching quality grades corresponding to the three evaluation parameters are weighted and summed to obtain a final teaching quality grade. The application can effectively measure the explicit knowledge level and implicit cognitive level of students, reduce the interruption of traditional methods to teaching activities, avoid environmental noise and interference, enhance the accuracy of classroom teaching quality evaluation, and have stronger cross-scene generalization.

The application relates to the technical field of dialogue processing, and discloses a dialogue method, system, device and medium with enhanced knowledge base retrieval, wherein the method comprises the following steps: in response to a dialogue context input by a user, performing attribute selection and entity selection on entities in an external knowledge base to obtain a target entity set; configuring corresponding implicit knowledge labels for entities in the target entity set to obtain an implicit knowledge enhanced entity set; wherein the implicit knowledge labels represent labels for distinguishing similar entities in the target entity set; configuring corresponding display knowledge texts for entities in the implicit knowledge enhanced entity set to obtain an explicit knowledge enhanced entity set; wherein the display knowledge texts represent descriptive texts of the entities in the implicit knowledge enhanced entity set; and generating a target response based on the explicit knowledge enhanced entity set and the dialogue context. The technical scheme provided by the application can improve the quality and accuracy of generated replies.

The invention provides a special structured query language generation method and device, equipment and a storage medium. The method comprises the steps that a conventional structured query language is converted into a first special structured query language based on a constructed grammar conversion rule; collecting a second special structured query language written manually in a historical business scene, wherein the first special structured query language, the second special structured query language and corresponding text questions, dominant knowledge and implicit knowledge form a special data set; performing parameter fine adjustment on to-be-optimized parameters of the general large language model based on the special data set and a low-rank adaptive parameter adjustment technology to obtain a special structured query language generation model; and obtaining a natural language query demand and real-time metadata of the database, and inputting the natural language query demand and the real-time metadata into the special structured query language generation model to obtain a target structured query language. The accuracy of the target structured query language is improved, and the calculation cost of the target structured query language is reduced.

This method for generating entity-aware sports video subtitles by combining explicit and implicit knowledge enhancement belongs to the field of video analysis and understanding. Existing traditional video subtitle generation methods have difficulty directly generating text descriptions with player identities and specific scene categories based on the visual content of a video. This invention uses the list of players participating in the game as explicit knowledge to help the model acquire pre-match knowledge. It proposes using learnable vectors to adaptively capture scene-related video features under the guidance of knowledge. Based on the attention mechanism, the relationship between players and video content is modeled to generate entity-related video features. A decoder based on the principle of "scene first, entity later" is proposed to enhance the accuracy and coherence of generated subtitles. The spatiotemporal dynamic information of the video is modeled to capture the temporal and spatial relationships between video frames. The effectiveness of this invention is verified on the basketball datasets VC_NBA_2022 and NSVA and the football dataset Goal.

The invention belongs to the technical field of brain-computer interfaces, and relates to a classroom teaching quality evaluation method and system based on a multi-human brain-computer interface, and a terminal. The evaluation method comprises the following steps: respectively calculating I-level grade thresholds and II-level grade thresholds of first, second and third evaluation parameters in a pre-class stage; in the in-class stage, respectively calculating and judging whether first, second and third evaluation parameters of each student individual are greater than or equal to corresponding I-level grade thresholds, if so, outputting A-level teaching quality, if not, judging whether the first, second and third evaluation parameters are greater than or equal to corresponding II-level grade thresholds, if so, outputting B-level teaching quality, and otherwise, outputting C-level teaching quality; and carrying out weighted summation on the teaching quality grades correspondingly output by the three evaluation parameters to obtain a final teaching quality grade. According to the method, the dominant knowledge level and the implicit cognition level of students can be effectively measured, interruption of teaching activities caused by a traditional method is reduced, environmental noise and interference are avoided, the classroom teaching quality evaluation accuracy is enhanced, and higher cross-scene generalization is achieved.

The disclosure provides a metaphor video understanding enhancement method based on a metaphor knowledge graph, belonging to the field of multimedia technology. The method comprises: identifying input information to obtain a plurality of source words, each source word being used to represent an object that appears explicitly in the input information; for any source word in the plurality of source words, determining a node in which the source word is located from a metaphor knowledge graph; taking the node in which the source word is located as a starting node, performing multi-hop query in the metaphor knowledge graph according to a preset number of times to obtain a plurality of first target words; processing the input information and the plurality of first target words corresponding to the plurality of source words through a large model to obtain a metaphor interpretation text, the metaphor interpretation text being used to interpret metaphorical semantics in the input information. The above method fuses explicit knowledge retrieval with a large model, which not only improves the accuracy of metaphor interpretation, but also improves the processing efficiency.

The invention discloses a text-to-image model capability verification method and system based on semantic element analysis, terminal equipment and a medium, and relates to the technical field of artificial intelligence model capability verification, the method comprises the following steps: obtaining domain knowledge, and generating a structured test prompt containing dominant knowledge and implicit knowledge based on the domain knowledge; performing semantic element analysis on the prompt to generate a semantic element table, calculating a coverage rate and complementing missing elements to cover all explicit and implicit knowledge; inputting a prompt into the to-be-evaluated text to the image model, generating a target image and generating process metadata; performing verification and reverse verification by using a pre-training verification model based on the semantic element table to obtain a semantic element verification result; and result aggregation is visualized, and a model capability evaluation report is generated. The method can comprehensively verify the reduction capability of the model to explicit and implicit knowledge, ensures verification traceability, accurately locates model knowledge blind spots, and improves the evaluation reliability.

This invention discloses a multi-agent collaborative knowledge reasoning system based on a large language model, relating to the fields of intelligent manufacturing and artificial intelligence. The invention constructs a multi-agent diagnostic system centered on explicit knowledge representation and collaborative reasoning. Through knowledge acquisition and compilation, the natural language output of the large language model is transformed into rules, facts, and ontology fragments that can be directly consumed by the inference engine. Combined with metadata containing source and timestamps, continuous incremental updates of cross-domain knowledge are achieved, overcoming the limitations of traditional static knowledge bases in covering dynamic faults. Secondly, using a blackboard and agenda mechanism as a collaborative carrier, intermediate assertions of agents such as vibration, circuits, and logs are structured for publication and subscription. A conflict resolution and consistency verification module makes unified decisions based on specificity, time freshness, and source credibility, thereby avoiding the accumulation of delays and uncertainties caused by long-chain natural language dialogues in highly real-time scenarios.

The application discloses a kind of implicit knowledge explicitization method for tobacco enterprise production scheduling, comprising the following steps: step S1: extraction and pre-processing scheduling knowledge;Step S2: build implicit and explicit scheduling knowledge classification word library;Step S3: key word extraction is carried out to implicit knowledge and explicit knowledge, the similarity of key word of the two is compared using n-gram matching algorithm, if the similarity m of the two is greater than or equal to the set threshold n, then according to the explicit knowledge, construct scheduling knowledge framework, combine scheduling knowledge framework to carry out knowledge integration to implicit knowledge key word set, complete implicit knowledge explicitization;Step S4: integrate implicit and explicit scheduling knowledge;Based on explicit knowledge, directly adopt OWL language and with the aid of Protege ontology development tool, top-down scheduling ontology is built;Through ontology mapping and ontology fusion method, the semantic integration between scheduling IS semantic model and scheduling ontology is realized, and the integration of implicit and explicit scheduling knowledge is completed.

The invention discloses a student knowledge assessment method and system based on chain reasoning and cognitive enhancement generation, and the method comprises the following steps: obtaining an original question set and an answer record text set of a student, and processing the original question set and the answer record text set to obtain a standard question set and a standard answer set; explicit knowledge point extraction is carried out on the standard question set, implicit knowledge point extraction is carried out on the original question set based on chain reasoning prompt, and then a knowledge point set is formed; performing generative adversarial network modeling on the non-mastered knowledge points based on the standard answer set to generate an enhanced sample set; extracting historical error frequency distribution of similar knowledge points as an auxiliary feature, and further verifying whether a generated error sample is compliant or not; inputting the mastered knowledge point features and the compliant enhanced sample set into a neurocognitive diagnosis model to obtain a mastering probability; and judging the mastering probability, and generating a knowledge state evaluation result. And in combination with a NeuralCD model, inputting student and practice codes and cognitive parameters, and outputting a knowledge mastering state of the student.

The invention provides an enterprise digital transformation knowledge graph construction and informatization collaborative management system. The system comprises a data acquisition module used for acquiring a multi-source data set corresponding to an enterprise; the knowledge graph construction module is used for performing knowledge extraction and fusion on the multi-source data set and constructing a target knowledge graph corresponding to the enterprise; the collaborative evaluation and analysis module is used for evaluating the maturity, the collaboration degree and the transformation gradient of enterprise digitization based on the target knowledge graph to obtain an evaluation result; the information management module is used for generating a task plan according to the evaluation result; and the visualization module is used for visually displaying specified data, and the specified data comprises the target knowledge graph and the evaluation result. Therefore, dominance of implicit knowledge is facilitated, and cognition on enterprises is enhanced; and information sharing and cross-department collaboration are facilitated, and the collaborative management efficiency of enterprises is improved.

The invention relates to the technical field of traditional Chinese medicine information, and discloses a traditional Chinese medicine ancient medical case implicit knowledge dominance and structuring processing method and device, a medium and a product. The method comprises the steps of obtaining a standardized medical case corpus according to original medical case data; determining a triple according to the standardized medical case corpus; the triple is used for mining implicit knowledge; obtaining a dominant knowledge set according to the triple; the dominant knowledge set is used for representing existing knowledge and implicit knowledge in a traditional Chinese medicine case; according to the dominant knowledge set, obtaining a traditional Chinese medicine ancient medical case knowledge graph; aiming at the traditional Chinese medicine ancient medical case knowledge graph, obtaining a fusion knowledge base; and realizing medical case knowledge service and visual application according to the fused knowledge base.

The application provides an analysis framework supporting natural language structure calculation, comprising three functional modules of a grid, a data table and a finite state automaton, and designs a series of API to cooperate with a Lua script language to complete writing of a language structure calculation script, including a general control script and a finite state automaton script; the language structure calculation script is executed through an executable program, so that language structure calculation is realized. The application takes symbolic calculation as a guide, receives input and outputs results, and has better controllability, expandability and interpretability; in the calculation process, various parameter calculation models can be dispatched, so that the capacity of the parameter models can be fully brought into play; finally, the framework can also introduce explicit knowledge, so that deep and fine language structure calculation is possible under the guidance of the explicit knowledge.

The data processing method comprises the following steps: acquiring a question; the question comprises a first entity; determining a second entity from first retrieval content related to the question according to the question; a first relationship exists between the second entity and the first entity; the first entity, the second entity and the first relationship are used for forming a first relationship team; and according to the question and the first relation team, performing reply determination of the question through a language model. According to the method and the device, the structured knowledge graph is formed through explicit knowledge tracking and knowledge triple exploration for the retrieval content, the LLM reasoning precision can be improved based on the structured knowledge graph, the problem that the underlying logic structure is difficult to perceive in the prior art is solved, and the reply generation precision is improved.