408 results about "Keyword extraction" patented technology

The invention belongs to the field of question and answer generation, and provides an interactive retrieval enhancement question and answer generation method and system based on a knowledge graph, and the method comprises the steps: carrying out the document partitioning based on an original document set, generating a global block set, carrying out the entity extraction of each text block in the global block set, and obtaining an entity set; performing relation extraction on entity subsets in each text block in the entity set to obtain a global relation set; generating a plurality of sub-knowledge maps based on the global block set, the entity set and the global relationship set, and performing entity fusion and relationship fusion on the sub-knowledge maps to obtain a knowledge map; performing keyword extraction and semantic embedding on the original problem to obtain a dense vector, performing semantic embedding based on the knowledge graph to obtain an embedded vector, and generating a candidate entity set according to the dense vector and the embedded vector; and based on the candidate entity set, utilizing a large language model calling tool to carry out extended search to generate a candidate information set, and utilizing a large language model to obtain an answer to the original question based on the candidate information set.

The invention relates to the technical field of computer vision and human-computer interaction, and discloses an AI digital human intelligent interaction method, system and equipment based on multiple modalities, and the method comprises the steps: pre-awakening a digital human when a human face is detected, and further thoroughly awakening the digital human based on recognized preset voice information or preset gesture information; voice and video information of a user in the interaction process is obtained, a keyword extraction result, a gesture recognition result and an emotional state tag are generated, a pre-constructed knowledge base is utilized to retrieve related information, a big language generation model module is combined to generate an answer text, and the answer text is input into a preset voice synthesis model to generate emotional voice output. And based on the current emotional state label of the user, driving the digital human animation to be output in an emotional manner. According to the method and the system, the digital human for understanding the emotion of the user, generating personalized answers, providing voices with rich emotions and displaying natural expressions and actions can be created, better interaction with the user can be realized, and more humanized and effective services can be provided.

The invention discloses a science and technology public text intelligent classification and service method and device based on deep learning. The method comprises the steps that multi-source science and technology public text data are acquired and preprocessed; extracting keywords by adopting a keyword extraction algorithm, and splicing the keywords with the public text title to form enhanced text features; constructing a multi-dimensional public text classification system and performing data annotation; performing feature extraction and fine adjustment by adopting a BERT pre-training model to obtain a classification model; automatically classifying the newly-added public texts and visually presenting the newly-added public texts; and generating a personalized recommendation result based on the user portrait and the public text feature index. The invention further relates to a technical scheme of multi-objective quality diversity optimization, heterogeneous resource allocation and fusion of the LPLC2 neural network and the BERT. The technical problems that a traditional method is limited in complex semantic understanding ability, single in classification dimension and lack of an integrated solution are solved, and the accuracy of science and technology public text classification and the intelligent level of service are improved.

The invention relates to the technical field of red tide anomaly detection, in particular to a red tide anomaly detection method and system based on an improved multi-mode Transform. The method comprises the following steps: acquiring a remote sensing image and text data; respectively carrying out data preprocessing according to the obtained remote sensing image and text data; performing visual positioning and text selection based on the preprocessed data; performing cross-modal feature learning on the basis of a hierarchical Transform of a multi-modal capsule mechanism; guiding an attention mechanism based on a semantic path to carry out image-semantic feature alignment optimization; and carrying out multi-modal knowledge distillation on the optimized features. According to the method, an image and text preprocessing module, a visual positioning module, a keyword extraction module and other modules are combined, multi-angle accurate perception of a complex red tide scene is achieved, and the bottleneck that a red tide area is difficult to accurately recognize under the condition that data are single and information dimensions are limited in a traditional method is broken through.

The invention discloses a college policy question and answer large model combined with a retrieval enhancement generation technology and a construction method. The method comprises the following steps: inputting a user question large language model and question-answer text vector data about college policies; multi-level indexes are adopted to retrieve question-answer text vector data, and policy content positioning related to the user question is achieved according to a retrieval result; semantic sorting is conducted on the retrieval results according to the scores, secondary evaluation of the policy content is achieved, and the policy content with the high semantic relevancy is preferentially provided; strategy content reasoning is carried out in a large and small model combination mode, including problem rewriting and keyword extraction by using the small model and strategy content generation by using the large model, so that computing resources are optimized and time delay is reduced; and generating answers to the user questions according to content reasoning, and outputting accurate college policy-related answers. The invention aims to improve the accuracy and efficiency of college policy information retrieval.

The invention relates to the technical field of information retrieval, and provides a privacy information retrieval system and method based on a block chain and function secret sharing, and the method comprises the steps: carrying out the encryption and keyword extraction of a data document; generating a Bloom filter index table between the keyword and the document based on the keyword trap door; uploading the encrypted file and the Bloom filter index table to a DSSP end, and transmitting the hash value of the encrypted file and the Bloom filter index table to a block chain for on-chain storage; storing the encrypted file in a distributed storage node under the chain according to the divided share; and the block chain calculates data of each column of the Bloom filter index table through consensus to obtain a message verification code, and the message verification code is stored on the block chain. According to the method, the index is constructed through symmetric encryption and the Bloom filter, and block chain evidence storage and FSS distributed storage are combined, so that data privacy protection and index credibility are realized. Through verification and tampering prevention, the problem that the existing FSS lacks verification and auditing mechanisms is solved.

The invention discloses a document retrieval method and device, equipment, a medium and a program product, and relates to the technical field of document processing. The method comprises the steps of obtaining a query statement of a document retrieval party, at least two candidate documents and candidate abstract vectors and candidate full-text vectors of the candidate documents; performing keyword extraction on the query statement to obtain a query keyword, and performing vectorization on the query statement to obtain a query statement vector; according to the query keyword, the query statement vector, each candidate document and a candidate abstract vector of each candidate document, performing coarse screening on each candidate document to obtain at least two coarse screening documents; and according to the query statement vector and the candidate full-text vector of each coarse screening document, performing fine arrangement on each coarse screening document to obtain a target retrieval document. According to the technical scheme provided by the embodiment of the invention, the accuracy of document retrieval is improved.

The invention provides an electric power document keyword extraction method based on Prompt and a knowledge graph, relates to the technical field of electric power document processing, and constructs a lightweight multi-level index knowledge graph in the electric power field by combining entity type and relation type division based on an electric power industry standard document and an electric power field corpus. The method comprises the following steps: performing vector modeling on a power document, constructing a multi-level index from an entity to a vector, realizing standardized semantic modeling and efficient hybrid retrieval of a power document field background, and obtaining a topic vector and a core paragraph of the power document in combination with power key information; according to the method, entity types are indexed in a knowledge graph by using subject vectors, similar entities are obtained to form knowledge sub-graphs, so that multilayer Prompt is obtained to guide a large language model to extract keywords, then knowledge graph similarity constraints are introduced to decode the output of the large language model, the keyword recognition capability in the power field is improved, and the keyword recognition efficiency is improved. And the accuracy of keyword type identification and the normalization of term naming are both considered.

The invention relates to the technical field of information management, and discloses a project cloud platform data management method based on a process market, and the method comprises the steps: obtaining process demand information inputted by an enterprise user, and obtaining an enterprise name of the enterprise user; crawling enterprise process information through a web crawler tool, and performing multi-level classification to construct a process market platform; obtaining the business scope and the enterprise scale of the enterprise user according to the enterprise name, and determining the enterprise type of the enterprise user according to the business scope; performing keyword extraction on the process demand information input by the enterprise user to determine a target keyword of the process demand information; and matching a demand flow list in a flow market platform according to the enterprise type, the enterprise scale and the target keyword, and pushing the demand flow list to the enterprise user. According to the method, an intelligent process market platform is constructed, keyword extraction and analysis are carried out on demands of enterprise users, automatic aggregation and accurate matching of process resources are realized, and the process adaptation accuracy is remarkably improved.

The invention discloses an AI large model-based arrival person screening method and device, a medium and equipment, and belongs to the field of screening, and the method comprises the steps of firstly obtaining screening conditions input by a user, including text content, image data and qualification information requirements, and to-be-screened arrival person data; next, performing word segmentation, keyword extraction and semantic matching on the text content of the person arrival data by utilizing a preset semantic analysis model in combination with BiLSTM, CRF and LDA technologies, and outputting a semantic score; meanwhile, note styles are recognized through the text classification model, logic judgment is conducted, and styles and logic verification scores are obtained. In addition, element detection and style verification are carried out on the image data through the multi-modal recognition model, and image matching scores are output; the qualification scoring module calculates qualification scores according to a preset weight formula. And finally, fusing the multi-dimensional scores to generate a comprehensive score, and carrying out accurate screening on the arriving persons according to the comprehensive score.

The invention discloses a file analysis and knowledge recall method and device for a large model question-answering system, equipment and a storage medium. The method comprises the following steps: analyzing a document, and extracting fine-grained information; receiving a natural language query request, and generating a question text containing similar questions; performing keyword extraction and weight distribution on the question text to generate keyword weight mapping; calculating similarity scores of different documents and question texts by using keyword weights based on a BM25 algorithm, and preliminarily screening documents associated with similar questions; in the primarily screened document, accurately querying a specific fragment; and smoothing the specific segments, and returning the final result after sorting the final result according to the comprehensive scores of the segments. According to the method and the device, the operation cost of the question-answering system is reduced, the analysis capability of fine-grained information is improved, the user can accurately position required specific information during query, and the retrieval efficiency and accuracy are improved.

The invention discloses a multi-agent multi-modal table question and answer method and system based on a cognitive collaboration mechanism. The method and system are particularly suitable for extracting accurate information from a complex, unstructured or multi-modal data source and answering questions of a user. The system is composed of four core modules, namely a task ontology analysis agent, a multi-modal clue perception agent, a logic verification agent and an arbitration fusion agent, and the task ontology analysis agent, the multi-modal clue perception agent, the logic verification agent and the arbitration fusion agent are integrated by receiving user questions and multi-modal table input. The intelligent agents cooperate to complete the processes of keyword extraction, intention recognition, vision-text cross-modal analysis, clue generation, logic verification, result fusion and the like, and an efficient cognitive closed loop is formed. The system not only supports operation on standard table data, but also can process visual clues such as complex header structures, cell merging modes, font styles and the like, so that the ability of understanding fuzzy or incomplete input is improved. Successful tests in a plurality of practical application scenes prove that the method has strong fault-tolerant capability and accuracy.

The invention provides an image video retrieval method based on a domain fine-tuning large language model, which comprises the following steps: performing fine-tuning on a pre-training model to obtain a fine-tuning pre-training model for intention classification and keyword extraction; performing dynamic iteration screening on an optimal prompt template through Monte Carlo tree search in combination with a hidden Markov model (HMM); performing noise filtering on the keyword list, and predicting category labels of the filtered keywords through a conditional random field model to obtain a keyword enhancement set; combining with the user intention to generate a query condition, and obtaining a candidate resource set; and according to the similarity between the user query text and the candidate resource set, and in combination with the optimal prompt template, obtaining the resource path with the highest matching score between the user query and the candidate resource, and obtaining the retrieved image or video, so that the identification deviation possibly occurring when a general model processes proper nouns and terminologies can be effectively solved, and the user experience is improved. And the retrieval accuracy and response speed are improved, so that the retrieval accuracy and professional adaptability are improved.

The invention relates to a railway official document keyword extraction method and device and electronic equipment, and the method comprises the steps: based on a pre-constructed railway official document format rule base, extracting a key field of a fixed position from an input text through regular expression matching and position locking; a Jieba word segmentation device is used for loading a railway-specific term library for word segmentation, and a multi-word combination entity boundary is dynamically corrected through a dependency relationship rule; executing a TF-IDF algorithm on the text after word segmentation to generate an initial word weight, adjusting the weight according to the position area of the word in the official document and a preset coefficient, and performing position weighting; and combining the words of which the weights are greater than a set threshold value with the extracted key fields, and outputting a final keyword set after verification of a term library. According to the method, missing detection caused by low frequency of a traditional algorithm is avoided, splitting errors of a general word segmentation device are eliminated, the term recognition error rate is reduced, the core word sorting priority is improved, and the semantic weight of keywords is strengthened; the new term storage time is shortened, and the updating cost problem is solved.

The invention provides a key event discrimination method and system based on a hierarchical processing architecture, and the method comprises the steps: obtaining reported events from different channels, carrying out the data preprocessing of the reported events, obtaining processed event texts, and classifying the event texts; clustering the classified texts by adopting a semantic regularization agglomeration hierarchical clustering method; and according to a clustering result, utilizing an in-cluster entity keyword extraction and comparison mechanism to discriminate repeated events in the text to obtain a key event discrimination result. According to the key event identification method provided by the invention, a classification-clustering-discrimination hierarchical processing architecture is constructed, the technical problem existing in one-event multi-report screening is effectively solved, and finally, the screened repeated report events are taken as key events needing to be marked. Therefore, related managers can preferentially dispose the key events or give full attention to the key events.

The invention discloses an intelligent information processing method and system for a shipping service scene. The method comprises the following steps: screening out an information text related to the shipping service scene through multi-dimensional processing; carrying out basic field identification, geographic element standardization, special keyword extraction and multi-dimensional collaborative structured information extraction of industry label code mapping on each information text; determining a first-level classification label and a second-level classification label of the information text by adopting a two-level classification strategy combining LLM and a rule engine; the title is translated and rewritten through LLM, a multi-level structured abstract is generated, structured event element information and industry value scores are extracted, and an effective information text is obtained; and performing information cluster division on each effective information text through LLM and a density clustering algorithm, reserving an optimal information text for each information cluster, extracting a core semantic structure, performing aggregation processing on the information texts with the same core semantic structure and a secondary classification label, and generating a unified event narration text.

The invention relates to an enterprise risk control dynamic portrait generation method and device based on multi-modal data fusion. The method comprises the steps that structured data, unstructured texts and time sequence behavior logs are collected and preprocessed, and the preprocessing comprises standardization processing, NLP keyword extraction and time sequence behavior feature extraction; then aligning the heterogeneous data by using a Transform architecture, generating a unified feature vector, and realizing cross-modal feature alignment; and on the basis, generating a dynamic risk portrait, calculating a dynamic behavior entropy value to quantify an operation anomaly degree, and generating a risk level map in combination with risk label matching. Through multi-modal data fusion, dynamic feature alignment and entropy quantification, the problems of single data, rule lagging and cross-modal association deficiency of traditional risk control are solved, the comprehensiveness, dynamicity and accuracy of risk identification are remarkably improved, accurate identification of composite risks such as business development places and actual travel itineraries can be realized, and the risk identification efficiency is improved. And the risk control level of enterprises is effectively improved.

The invention belongs to the field of computer vision and industrial design assistance, particularly relates to an industrial drawing similarity detection method based on deep learning, and aims to realize efficient and accurate similarity analysis of industrial drawings. Comprising the following steps: collecting an industrial drawing data set, and carrying out target detection labeling and feature extraction on each drawing; a YOLO model is adopted to detect a target, and a non-maximum suppression method is adopted to remove duplication so as to obtain each visual angle of the part in the drawing and an overall marking frame; recognizing characters in the drawing, and obtaining part names and material information through keyword extraction and domain dictionary filtering; local and global feature vectors are extracted, multi-modal feature alignment is realized through comparative learning, and part semantic representation based on image-text association is constructed. And storing the extracted image and OCR text feature vectors into a Faiss index database, establishing a multi-modal mapping relation of local and global feature vectors and text semantics, extracting the image and text features of a to-be-detected drawing, and outputting a similarity result through similarity retrieval of a vector database.

The invention relates to the technical field of literature retrieval, in particular to a two-step literature retrieval method and system based on keyword extension. Comprising the following steps: calibrating an initial word, inputting a user keyword, querying a preset high-frequency subject word bank, and executing a calibration rule; preliminary retrieval: performing retrieval in an AuthorKeywords field of a target database by using the calibrated seed keyword to obtain a preliminary retrieval result set; and extracting and purifying extended keywords. According to the two-step method provided by the invention, through automatic expansion and parallel field retrieval, wider coverage can be realized in a single process, it is expected that operation rounds required by a user for repeatedly trying different keyword combinations can be obviously reduced, and the automatic expansion keyword extraction, filtering and multi-field parallel retrieval mechanism can improve the efficiency of the user. According to the method provided by the invention, a relatively wide literature range can be covered by single execution, so that a relatively comprehensive retrieval result can be expected to be obtained more efficiently through a structured automatic process, and the time cost of repeated trial and error and screening of researchers is saved.

The invention relates to the technical field of data processing, in particular to a multi-modal medical data intelligent association analysis system based on deep learning. The system comprises a data acquisition module used for acquiring a plurality of first data and a plurality of second data of a target patient, the first data being structured medical data, and the second data being unstructured medical data; the keyword extraction module is used for extracting a plurality of keywords of each piece of second data; the clustering module is used for clustering the multiple pieces of second data based on the multiple keywords of each piece of second data to obtain multiple class clusters; the time sequence analysis module is used for performing time sequence analysis on the treatment time period based on the plurality of class clusters so as to determine a plurality of diagnosis and treatment time slices; and the correlation analysis module is used for performing correlation analysis on the second data and the first data in each diagnosis and treatment time slice according to the plurality of class clusters so as to obtain a correlation analysis result. According to the invention, the association analysis effect of the multi-modal medical data can be improved.

The invention relates to the technical field of judicial informatization and artificial intelligence, and particularly provides an intelligent case processing method and system for litigation mediation and a medium. The method comprises the steps that litigation data of appeals are obtained and filled into an appeal form, the compliance state of the appeal form is judged, if the appeal form is compliant, keyword extraction is conducted on the appeal form, vector conversion processing is conducted, an appeal semantic vector is obtained, the appeal semantic vector is input into a preset appeal classification model to be processed, and the appeal data of the appeal form is obtained. Obtaining complaint and call category feature data, generating a mediation scheme according to the complaint and call category feature data, performing mediation, and if the mediation is successful, generating a mediation document and sending the mediation document to a case party; according to the method, the complaint form is automatically generated, the compliance state verification is performed, the complaint category feature data is determined through keyword extraction and vector representation, and then the conciliation scheme is generated, so that the whole-process intelligent processing of the litigation conciliation case is realized, and the efficiency and success rate of litigation conciliation are improved.

The invention provides a hospital health education intelligent question and answer method based on a large language model and a knowledge base, and belongs to the technical field of intelligent question and answer, and the method comprises the steps: collecting hospital health education related document data for preprocessing, and converting a text block into a vector form through a vector embedding model, storing in an AI native database in combination with metadata, and constructing an index to form a hospital health education knowledge base; receiving health consultation questions input by a user, performing semantic analysis and keyword extraction, and retrieving related knowledge fragments from a hospital health education knowledge base; carrying out fusion reordering on the retrieved knowledge fragments, calculating correlation scores, and selecting the first K knowledge fragments with the highest scores; and substituting the selected knowledge fragments and the user question into a preset cue word engineering template, generating a cue word input large language model, generating an answer for the user question, and feeding back the answer to the user. The acceptability and practicability of the patient on health knowledge are enhanced, and the efficiency of hospital health education and the satisfaction degree of the patient are improved.

The invention relates to the technical field of legal consultation services, in particular to an intelligent question and answer matching method based on machine learning, which comprises the following steps of: acquiring legal consultation text recognition subject behavior objects and articles, analyzing and carding statement logic in sections, generating a semantic hierarchical table, extracting key phrases and integrating semantic units, and constructing a semantic network to generate a mapping table. Comparing question answer semantic logic to generate a matching corresponding set, reviewing offset revised text to generate a revised set, and rechecking consistency arrangement to generate an intelligent matching result set. According to the method, legal text logic relations are analyzed and processed through semantic layering, question and answer matching precision is optimized, keyword extraction and semantic network construction are combined, semantic consistency recognition and matching flexibility is enhanced, semantic offset and logic inconsistency are corrected, result connection coherence is ensured, mismatching is reduced compared with the prior art, question and answer accuracy is improved, and the method is suitable for popularization and application. Legal consultation is more intelligent and accurate, and a reliable matching result is provided.

The invention provides an intelligent network operation and maintenance method and system based on performance analysis and a large language model, and belongs to the field of network operation and maintenance. The method comprises the following steps: firstly, embedding a large language model into a network application performance analysis system; performance analysis platform authentication information and an API path are configured, a semantic analysis template, an inference rule base, a graph model and an alarm threshold value are initialized, and a report generation template and a scheduling strategy are configured; receiving an operation and maintenance problem, a query request or a diagnosis demand input by a user in a natural language form; based on a semantic analysis template configured in the large language model, intention recognition and keyword extraction are carried out, a special language DSL in the field of structured query is generated, network application performance analysis is carried out, performance feedback data is obtained, index analysis is carried out, an analysis result is fed back to the large language model, a natural language is generated according to the feedback result, and the natural language is stored in a database. And displaying according to the report generation template. The problem analysis efficiency, the automation degree and the network operation and maintenance efficiency and effect are improved.

The invention provides a vehicle-mounted log uploading method and system, vehicle-mounted equipment and a storage medium, and relates to the technical field of data processing.The method comprises the steps that log data of a target vehicle is obtained and subjected to structured processing, and structured log data comprising multiple structured log texts are obtained; based on a preset keyword library, performing keyword extraction on each structured log text through a keyword extraction algorithm to obtain a plurality of keywords; based on a natural language processing algorithm, determining the semantic similarity of any two keywords; based on the semantic similarities, performing clustering analysis on the plurality of structured log texts to obtain a plurality of log clusters; obtaining an abnormal score of each log cluster according to a preset scoring factor corresponding to each log cluster; according to each abnormal score, screening out at least one target log cluster from the plurality of log clusters, the target log cluster comprising at least one target log; and uploading each target log to a cloud server. According to the invention, the uploading efficiency of the vehicle-mounted log is improved.

The invention discloses an online diagnosis and treatment data supervision method and system based on artificial intelligence, which is used for carrying out real-time intelligent supervision on free text diagnosis and treatment records generated by an online medical platform, and specifically comprises the following steps: carrying out semantic coding on text fields such as basic information, chief complaint, present medical history, past history and diagnosis of a patient; executing necessary filling and logic consistency verification based on a predefined multi-level rule base; clustering analysis is carried out on the semantic vectors by adopting a density clustering algorithm and the like, and semantic abnormity or duplicate records are identified; filtering junk texts and test data by using keyword extraction and named entity recognition technologies; and finally, outputting a structured supervision report and adding a quality label to each record. Through the method, various quality and compliance problems in the diagnosis and treatment data can be found and labeled in real time, and powerful support is provided for compliance supervision of an internet medical platform, improvement of service quality and guarantee of data credibility; meanwhile, a rule-driven and AI model-driven'double-track 'framework is adopted in the method, and an active learning mechanism is combined, so that the system can continuously perform self-learning and iterative evolution and continuously adapt to a new data mode and a new supervision requirement.

The invention relates to the technical field of archive management, and particularly provides an archive task processing method, device and equipment and a readable storage medium, and the method comprises the steps: carrying out the primary processing of a to-be-processed archive through a preset archive processing large model, and obtaining the global information of the to-be-processed archive, the primary processing comprises at least one of global understanding, context analysis, key information extraction, semantic relation extraction and cross-archive information integration; the global information is subjected to final processing through a preset file processing small model, a processing result is obtained, and final processing comprises at least one of keyword extraction, file classification and file abstract generation; and processing a target archive task according to the affair processing result, the target archive task including archive management, information query or decision support of the to-be-processed archive. Through the method, the effect of efficiently and accurately processing the file task can be achieved.

The invention discloses an intelligent archive keyword indexing method based on multi-modal semantic association, and relates to the technical field of archive keyword indexing. The method comprises the following steps: initially performing text segmentation of different target detection on text modal information; semantic expansion information block segmentation of corresponding target detection is carried out on image, audio and video modal information in an associated manner in sequence, and keyword extraction weight value statistics of positive samples with similar semantics and negative samples with opposite semantics is carried out on segmented sub-information blocks, and the segmented sub-information blocks are combined; the method comprises the following steps of: segmenting information blocks of image, audio and video modal information which are respectively used as initial processing objects in sequence, combining segmented text sub-information I, image sub-information II, audio sub-information III and video sub-information IV, and carrying out keyword extraction weight value statistics; and summarizing the test sample set obtained by statistics to realize keyword indexing through an established file keyword extraction model test. The keyword indexing accuracy can be improved.

The invention provides a work order clustering and theme extraction method, system and device and a medium, and belongs to the technical field of natural language processing and data mining. The method comprises the steps of obtaining original work order data, performing data cleaning, extracting work order abstracts from the original work order data by using a text abstract generation technology according to preset abstract constraints, and generating an abstract set; inputting the work order abstract into a language model, generating a semantic vector of a preset dimension, constructing a vector index through an FAISS library, and converting the abstract set into a set of numerical vectors; clustering the numerical vectors by using an improved K-means algorithm, and outputting K groups of work order digests with balanced quantity; for each group of work order abstracts, screening representative abstracts based on text similarity; extracting keywords from the representative abstracts by adopting a keyword extraction technology, and generating a keyword list of each group of work order abstracts; and for the keyword list of each group of work order abstracts, generating a theme tag by combining keywords.

The invention relates to a mixed RAG retrieval method and device based on constraint keyword extraction. The method comprises the following steps: extracting a binding keyword from a retrieval problem of a user; the constraint keyword is a structured information unit playing a role in limiting or filtering a retrieval result corresponding to the retrieval problem; constructing a query statement corresponding to the retrieval problem based on the extracted constraint keyword; performing data filtering on data in a vector database by utilizing the constructed query statement, and screening out first target data in the vector database; the first target data comprises a target document fragment and a target document vector; calculating the similarity between every two first target data and the retrieval problem, and determining a preset first quantity of first target data with the highest similarity with the retrieval problem as a retrieval result. By the adoption of the method, the problems that an existing RAG retrieval method is too large in calculation amount and low in retrieval result precision can be solved.