345 results about "Text retrieval" patented technology

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 discloses a multi-mode fusion product document and source code association retrieval method based on a knowledge graph, and relates to the technical field of software engineering and artificial intelligence. The method comprises the steps that source codes are preprocessed, and code structure information and business semantics are mapped in combination with a predefined business term dictionary; performing controlled induction on a code file and a product document by utilizing a large model, extracting business terms, logic intentions and a subject relationship, fusing with original codes, and establishing a vector retrieval index system; further analyzing a code structure by using an abstract syntax tree, and extracting an entity and a calling relationship; semantic enhancement and relation normalization are performed in combination with the large model, entities and relations are stored in a graph database, and a knowledge graph is formed; and performing parallel processing on user query based on a full-text retrieval index, a vector retrieval index system and a knowledge graph, and finally generating a product concept. According to the method, the retrieval speed, the semantic depth and the logical reasoning ability can be considered at the same time, and the retrieval accuracy is improved.

The invention discloses a water conservancy design archive retrieval system and method based on a local lightweight large model, and the method comprises the steps: S1, constructing a Python automatic preprocessing assembly line, extracting texts for PDF and Word multi-format archives, correcting metadata, and outputting standardized data; s2, constructing a full-text retrieval and semantic retrieval dual-mode cross-document retrieval service by relying on a Weavi ate local vector database and a lightweight text embedding model; s3, analyzing a user query intention through a local large model, synchronously triggering metadata accurate retrieval and content semantic retrieval, and generating a structured result; and S4, integrating the core module into a local area network Web platform, adopting Docker containerization deployment, and combining an RBAC permission model and JWT authentication to guarantee security. The system comprises a preprocessing module, a cross-document retrieval module, an intelligent agent module and a background management module, and collaboration is achieved through a standardized API. According to the method, the problem of archive fragmentation is solved, multi-mode retrieval breaks through keyword limitation, an intelligent agent reduces manual intervention, a localized architecture prevents secret-related leakage, background management adapts to an existing I T environment, and full-process intelligent archive service is provided for water conservancy design.

The invention relates to a retrieval method and device based on static word embedding, computer equipment, a computer readable storage medium and a computer program product. The method comprises the steps of performing word segmentation on an original text to obtain a first word segmentation result, training a static word embedding model by utilizing the first word segmentation result to obtain word vectors, and generating a synonym word library; the method comprises the following steps: establishing a full-text inverted index by utilizing an original text, and expanding query words by utilizing a synonym library in a retrieval stage; encoding the original text into a semantic vector by using a semantic generation model, and constructing a vector index based on the semantic vector; based on a to-be-queried text in a user query request, performing retrieval by using the full-text inverted index to obtain a first candidate document, and performing retrieval by using the vector index to obtain a second candidate document; performing fusion processing on the first candidate document and the second candidate document to obtain a target candidate document; and inputting the target candidate document into the text generation model to obtain a retrieval result. By adopting the method, the accuracy of text retrieval can be improved.

The invention discloses a file retrieval method, device and equipment based on multi-modal AI and a storage medium, relates to the technical field of data retrieval, and aims to solve the problem that traditional file retrieval is low in efficiency and precision. The method comprises the following steps: performing content analysis on a multi-modal file including a picture file, a video file and a document file of text and / or visual information, and extracting a text semantic feature, a picture visual feature and a video time sequence feature; mapping the text semantic feature, the picture visual feature and the video time sequence feature to a cross-modal semantic space used for representing a high-dimensional vector associated with different modal features, and generating a cross-modal association vector; according to the modal type of the retrieval information, a corresponding retrieval module in a mixed retrieval engine is called to conduct retrieval in a cross-modal semantic space, an initial retrieval result is obtained, and the mixed retrieval engine comprises a text retrieval module, a visual retrieval module and a cross-modal fusion module; and performing dynamic weight distribution sorting on the initial retrieval result, and outputting a target matching result.

The invention discloses a retrieval enhancement question and answer generation method and system based on a large language model in combination with a multi-agent cooperation mechanism. The method comprises the steps of S1, text preprocessing and semantic difference enhancement and amplification; s2, checking and complementing user questions; s3, evaluating problem complexity and selecting a generator; s4, text retrieval and answer generation; the system comprises an information enhancement processing module, a query understanding optimization module, a question complexity intelligent evaluation module and a collaborative answer generation module, and is used for realizing the method. According to the method, a multi-agent collaborative RAG framework of an information enhancement agent, an interaction analysis agent and a problem complexity evaluation agent is introduced, and semantic difference enhancement and amplification, query integrity verification and complementation and a dynamic generation strategy based on problem complexity are performed after hierarchical document partitioning are combined; and the performance of the system in the aspects of semantic distinguishing capability, query understanding precision, generation efficiency and reliability is comprehensively improved.

The invention discloses a large model retrieval enhancement generation method and system, and the method comprises the steps: obtaining the comprehensive document data of a power grid field, carrying out the partitioning processing of the comprehensive document data, and obtaining the partitioned text data; performing deep analysis on the block text data to generate a thinking chain enhanced text; encoding the thinking chain enhanced text based on a hierarchical weighted encoding method to generate a thinking chain enhanced semantic vector; inputting the block text data into a pre-trained large language model to obtain a generated task perception representation vector; splicing the thinking chain enhanced semantic vector and the generated task perception representation vector to obtain a fusion vector, and constructing a vector database in the power grid field based on the fusion vector; responding to the text retrieval signal, obtaining question text data of the target user, analyzing the question text data to obtain a question fusion vector, comparing the question fusion vector with a vector database, and obtaining retrieval data of the question text data based on a comparison result.

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

The invention relates to the technical field of remote sensing image analysis and cross-modal retrieval. The invention discloses a remote sensing image text retrieval method based on a remote sensing multi-modal basic model, which applies the large-scale pre-training capability of a CLIP large model to semantic alignment of a remote sensing image and a text by finely adjusting the CLIP large model. By introducing the visual saliency calculation module and the visual block fine-grained selection integration module, the problems of multi-scale targets and redundant information in the remote sensing image are effectively solved, fine-grained semantic alignment between the image and the text is realized, and the retrieval accuracy is improved. Particularly, under the condition that the image contains a plurality of salient targets and redundant regions, the cross-modal semantic alignment fine-grained filtering method provided by the invention can accurately identify key information blocks in the image and perform fine matching with text description.

The invention provides a cross-modal joint contrast learning method and device and electronic equipment, and the method comprises the steps: constructing a sample data set which covers a plurality of task types, such as a text retrieval image, an image retrieval text, a text retrieval text, an image retrieval image, an image-text joint retrieval image and an image-text joint retrieval text; and generating prompt words for identifying task types for each task sample, splicing the prompt words with sample data to form task input, and determining modal types of a retrieval object and a retrieval target. A retrieval object and a retrieval target are respectively input into encoders of corresponding modes to extract features, the features are mapped to the same semantic space through a unified projection layer to obtain an embedded vector, and the parameters of the encoders and the projection layer are optimized by utilizing a contrast learning loss function based on the similarity of the retrieval object and the retrieval target, so that multi-task unified training is realized. Various cross-modal retrieval tasks can be supported in a unified semantic space at the same time, and the overall retrieval effect is improved on the premise of ensuring multi-task performance balance.

The invention discloses a server hardware testing terminal and method, and relates to the technical field of server hardware testing, and the method comprises the steps: obtaining server hardware configuration parameters and storage equipment specification information, collecting the number of processor cores, memory capacity and storage equipment read-write speed reference data through a system monitoring interface, building a hardware performance baseline file, and storing the baseline file in a server; obtaining a complete hardware resource list and a performance index range; the method comprises the following steps: constructing a multi-dimensional workload model according to information retrieval scene features, creating query request sets of different scales by adopting a random number generator, simulating a mixed load mode of full-text retrieval and database query, and determining resource consumption weights and execution time distribution of various query operations; and analyzing query request distribution characteristics in the workload model through a dynamic load balancing algorithm, if query requests are concentrated in a specific time period, adjusting a load distribution strategy, obtaining a uniformly distributed test load sequence, and judging the impact degree of a load peak value on hardware resources.

The invention discloses a multi-modal data processing method and device based on a large model, equipment and a medium. Performing feature extraction and association mining on the multi-modal data to obtain multi-modal data features; according to a retrieval sequence of the multiple text matching methods, sequentially using the multiple text matching methods to perform text retrieval on the multi-modal data features in a pre-constructed knowledge base to obtain a first associated sub-graph; based on the semantic similarity between the multi-modal data features and multi-modal data in a pre-constructed knowledge base and a similarity threshold value, determining a second associated sub-graph; performing standard evaluation on the plurality of candidate results in the first associated sub-graph and the second associated sub-graph, and sorting the plurality of candidate results according to an evaluation result to obtain a retrieval result; inputting the retrieval result and the multi-modal data features into a large model for information analysis to obtain prompt information and a target tool; and sending the prompt information and the target tool to the terminal, and receiving a data processing result of the terminal.

The embodiment of the invention discloses a question answering method and device based on retrieval enhancement generation, equipment and a medium, and relates to the technical field of large language model question answering. The method comprises the steps of obtaining a target question input by a user; on the basis of keywords in the target question, keyword retrieval is carried out on each text in a pre-established knowledge base; each text in the knowledge base corresponds to a text vector; retrieving a text vector similar to the target vector in a pre-established knowledge base, and determining a text corresponding to the retrieved text vector; the target vector is a vector generated based on a target problem; and processing the target question and a retrieval result obtained in the knowledge base based on the large language model to obtain an answer corresponding to the target question. According to the technical scheme, text retrieval and vector retrieval are carried out in the knowledge base, the retrieval result with higher timeliness and higher accuracy is obtained, and then the answer output by the large language model can be obtained based on the retrieval result and the target question.

The invention discloses a natural language and machine vision-based alignment method, which comprises the following steps of: providing a three-level alignment architecture, and respectively extracting local-global features of a visual scene and grammar-semantic features of a natural language by adopting a double-flow feature extraction network; through a space-time attention enhanced cross-modal alignment module, a dynamic gating mechanism is adopted to complete feature space adaptive projection; a joint optimization strategy is constructed based on comparative learning, and vision and language embedding space consistency is optimized by using a multi-granularity comparative loss function. And meanwhile, semantic topology constraint and visual causal reasoning are introduced, so that the calculation complexity is reduced, and the task robustness is improved. Through experimental test data, the Top-1 accuracy rate of the method in image-text retrieval reaches 92.3%, the visual question and answer F1 value is 83.4%, the model parameter quantity is reduced by 40%, the method can be widely applied to the fields of intelligent interaction systems, automatic driving scene understanding, industrial quality inspection knowledge base construction and the like, and the semantic perception and reasoning ability of a multi-modal system is remarkably improved.

The present invention relates to the technical field of artificial intelligence, and provides a diversity-enhanced text retrieval-augmented generation method and system. The method comprises: using knowledge data to construct a local knowledge base; acquiring a user question and the desired number of retrieval results; in a retrieval stage, using a large language model to paraphrase the user question to obtain a paraphrased question, then expanding the retrieval range, expanding the number of retrieval results, and performing vector retrieval in the knowledge base on the user question and the paraphrased question to obtain retrieval results; calculating the similarity among the retrieval results, and screening for diverse retrieval results on the basis of the similarity calculation result; and integrating the screened retrieval results and the user question by means of prompt engineering, inputting the integrated result into the large language model, and taking a model generation result as an answer to be returned to the user, thereby providing the user with richer and more informative search results, and improving the richness and accuracy of content generated therefrom.

The invention discloses an image-text retrieval method based on feature collaboration and adaptive attention adjustment, which comprises the following steps of: firstly, generating a regional feature set of an image and a text feature set corresponding to a sentence, and generating an enhanced regional feature set by adopting a global-local feature collaboration enhancement module; performing interactive matching on word features in the text feature set and the enhanced regional feature set to obtain a comprehensive image feature vector concerned by each word, and performing similarity calculation to obtain a similarity score between an image and a sentence; and meanwhile, an adaptive cross-modal attention regulator module is adopted to update a comprehensive image feature vector concerned by each word, and triple loss based on the most difficult negative sample is applied to training of a target function. According to the method, through cooperation of local and global features of the image, region features are enhanced, channel weights and attention distribution of image regions and word pairs are optimized, and then the cross-modal semantic alignment capability between the image and the text is improved.

The invention provides a large language model-based finance and taxation data intelligent dialogue and query system, and relates to the technical field of finance and taxation data processing and natural language processing.The method comprises the following steps of: firstly, constructing a middle adaptation layer for connecting a large language model and a finance and taxation service database, and converting database information into a form which can be recognized by the large language model; and performing finance and taxation field special training on the large language model, so that the large language model can process finance and taxation problems of a user and generate a query instruction. And executing data query through the middle adaptation layer to obtain an original numerical result. The model is called to analyze the result to generate preliminary interpretation, and a finance and tax rule text retrieval library is combined for correction to form a final result and push the final result to the user. The finance and taxation data query efficiency and accuracy are improved, and comprehensive and professional finance and taxation information is provided for users.

The invention discloses a remote sensing cross-modal image-text retrieval method and device based on knowledge distillation, and relates to the technical field of cross-modal retrieve.The remote sensing cross-modal image-text retrieval method comprises the steps that a retrieval problem is received, the retrieval problem is preprocessed, a retrieval problem text is obtained, the retrieval problem text is input into a preset retrieval model, and a retrieval result is obtained; and outputting the target remote sensing image. According to the method and the device, the technical problem of relatively low accuracy of remote sensing cross-modal image-text retrieval in related technologies is solved.

The invention belongs to the technical field of cross-modal image-text retrieval, and discloses an image-text cross-modal retrieval method based on an embedded sparse gate expert hybrid model, which comprises the following steps of: cross-modal multi-scale modeling: extracting multi-scale image features of an image by utilizing a cavity space pyramid pooling module ASPP, dynamically weighting text features by utilizing a multi-scale activation factor, and carrying out multi-scale multi-scale modeling; image-text cross-modal multi-scale semantic alignment is realized; multi-scale cross-modal feature fusion: designing a multi-scale cross-modal router, fusing image and text features through cross attention, and extracting cross-modal joint features in a scale division manner through an expert network; and two-way triple loss calculation: adopting a two-way triple loss function, and combining with intra-scale and cross-scale constraint optimization feature space to realize joint optimization of multi-scale and cross-modal levels to obtain a final cross-modal retrieval result. According to the method and the device, the precision and the efficiency of cross-modal image-text retrieval are improved.

The invention relates to a remote sensing image-text retrieval method based on knowledge enhancement and an asymmetric structure, and belongs to the technical field of remote sensing image-text cross-modal retrieval. Comprising the following steps: inputting a to-be-retrieved remote sensing image and text into a trained vision-language basic model with knowledge enhancement and an asymmetric structure, and realizing cross-modal retrieval through model processing; and obtaining a retrieval result, obtaining matching retrieval output of the remote sensing image and the text, and completing a cross-modal retrieval task. According to the method, the problem that the retrieval performance is limited due to the fact that significant information asymmetry exists between the remote sensing image and the text modality at present is solved. According to the method, the cross-modal asymmetric Kolmogorov-Arnod adapter fine tuning method is designed, so that efficient modal fine-grained shared feature learning is realized; meanwhile, knowledge is extracted from ConceptNet and a remote sensing knowledge graph, and a knowledge enhancement sentence is generated to enrich text semantics, so that a semantic gap between modals is bridged, and remote sensing image-text retrieval performance is improved.

The invention provides a multi-dimensional right dynamic control method and system based on full-text retrieval. The method comprises the following steps: configuring a right requirement mark of each business data object in an information system; each permission requirement mark is converted into a permission feature character string, and the permission feature character string is used as a full-text retrieval field to be associated with a business data object; in the operation process of the information system, collecting a permission matching log in a user access request, and counting the permission popularity of each permission feature character string; according to the permission popularity, identifying each target permission feature character string as a permission hotspot set; combining the index entries corresponding to the data objects belonging to the same authority hotspot set to the same index fragment; according to the method, the query condition is generated according to the feature information of the target access user, full-text retrieval is executed on the updated index fragment structure, access data allowed by the target user is obtained, permission hotspots can be accurately recognized and adjusted, and flexibility and high efficiency of data access are guaranteed.

The invention belongs to but is not limited to the technical field of artificial intelligence, and particularly relates to a multi-strategy RAG process optimization method and system for large model questioning and answering, and the method comprises the steps: carrying out the classification processing of a document of a user, and selecting semantic blocks or customizing the sizes of the blocks; receiving a text input by a user; the retrieval is divided into two parts, one part is semantic vector retrieval, and the other part is keyword retrieval; the system further screens and filters candidate text paragraphs in combination with structured meta-information on the basis of vector retrieval; after the preliminary vector retrieval, a bge-ryanker-v2-m3 model is introduced to carry out secondary sorting on the results; the model firstly receives input Prompt and converts the Prompt into vector representation; the vector is processed through a multi-layer Transform network so as to extract and model semantic features in the text; a corresponding output result is generated in combination with specific task requirements defined by Prompt; and outputting an answer generated by the large model for the user question.

A multi-granularity semantic alignment method oriented to cross-modal image-text retrieval comprises the steps that multi-layer feature representation of an image and a text is established, and a global vector and a local mark sequence are obtained respectively; capturing fine-grained features with higher information amount by using a mark screening mechanism, and inhibiting interference of redundant and invalid features on representation; under a unified semantic alignment framework, a bidirectional mapping relationship among multi-granularity semantics is constructed, and coarse-granularity, fine-granularity and cross-granularity semantic similarity alignment is comprehensively realized so as to strengthen multilayer semantic consistency constraints; and adaptively integrating the granularity similarities through a weighted fusion module, generating single instance-level similarities, and carrying out descending sorting on candidates. According to the method, the multi-granularity characteristics of the image and the text are comprehensively considered, the image-text retrieval precision in an actual scene can be improved, and the method has good application value in the fields of image-text retrieval, multi-modal content analysis and the like.

The invention provides a multi-level alignment video text retrieval method and system based on a multi-modal large model. The method comprises the steps that S1, sparse sampling is carried out to obtain a video frame sequence needing to be used for retrieval; s2, extracting video feature representation and text feature representation; s3, obtaining a video-text global level similarity; s4, selecting a plurality of video-text expert multi-level alignment networks through a routing network; s5, the total loss is obtained through calculation, the model is adjusted and updated, and a target video text retrieval model is obtained; s6, according to the video text retrieval model obtained through training, similarity calculation is conducted on videos and texts used for retrieval and video text data in a database, and videos or texts which are related to the videos and texts and have the highest similarity are obtained; by applying the technical scheme, sparse and refined execution path selection of the retrieval task is realized, and the accuracy of video text retrieval is improved through multi-level semantic alignment.

The invention discloses a method and a system for retrieving a picture file through a text based on cross-modal feature alignment, and the method comprises the steps: extracting text and picture features, and guaranteeing that a high-value mode contributes to a higher weight based on multi-modal attention weighted fusion; calculating a dynamic temperature coefficient through initial semantic similarity of positive and negative samples to construct a loss function for multi-modal contrast learning, and mapping original features of a text and an image to a unified space to obtain alignment features; according to the method, the picture archives are retrieved through texts, the image-text similarity, the time sequence weight and the core area proportion weight are comprehensively considered, optimization sorting of time sequence perception is carried out, and the most matched picture archives are obtained. According to the method, a text-image cross-modal semantic gap is solved, so that semantic alignment of two types of features in a unified space is realized; according to the method, the sample difficulty is dynamically adapted to improve the feature distinction degree; weights of texts and images are distributed according to needs so as to retain core information; according to the method, retrieval result sorting is optimized in combination with time attributes.

The invention discloses a mixed retrieval method and system based on multi-dimensional weight self-adaption, mainly relates to the technical field of mixed retrieval, and aims to solve the problems of low retrieval accuracy and poor applicability caused by adopting a fixed weight between keyword matching and semantic understanding in traditional mixed retrieval. Comprising the following steps: according to a vector weight adjustment coefficient and initial weight distribution, obtaining adjusted first weight distribution; according to a weight threshold interval corresponding to full-text retrieval and vector retrieval, adjusting the first weight distribution to obtain an adjusted second weight distribution; based on the full-text retrieval path and the vector retrieval path, obtaining a full-text retrieval sorting result and a vector retrieval sorting result by utilizing the second weight distribution value and the user query text; and integrating the full-text retrieval sorting result and the vector retrieval sorting result by adopting a deduplication merging strategy and a rescore mechanism to obtain a fusion sorting result.

The invention relates to the technical field of computer big data processing and retrieval, and discloses a data multivariate storage and full-text retrieval system which is characterized in that an isomorphic hash intake module is used for truncating an input data stream into logic data blocks, extracting lexical elements to generate isomorphic hash packets and distributing the isomorphic hash packets to a bottom layer for storage; the state monitoring module maintains heat potential energy and queries fingerprints, updates the state according to scanning feedback, and sends out a trigger signal when the heat exceeds a dimension rising threshold value; the differential projection module responds to the signal to read the isomorphic hash packet, and constructs a differential inverted index by calculating the query fingerprint and the intersection of the query fingerprint; and the cascade retrieval module executes memory search when the index is mounted, and scans the logic data block and feeds back a result to the state monitoring module if the index is not hit. According to the method, the heat potential energy of the data block is tracked through the state monitoring module, index construction is triggered only when query accumulation reaches the dimension raising threshold value, and cold start resource consumption of the system is obviously reduced.

The invention discloses a retrieval enhancement generation method and device based on an electric power knowledge graph. The method comprises the following steps: performing text retrieval on an obtained query text in an electric power knowledge base to obtain a text retrieval result related to the query text; querying the electric power knowledge graph according to the text retrieval result to obtain a knowledge graph retrieval result; performing unified integration and organization optimization on the text retrieval result and the knowledge graph retrieval result, and constructing a context prompt; and inputting the context prompt into a preset language model to generate question and answer content. According to the technical scheme, the electric power knowledge graph can be introduced into the retrieval enhancement generation process, so that the problem that an existing retrieval enhancement generation method lacks effective utilization of structured knowledge involved in the electric power field is solved, and the generation quality when a language model processes professional problems in the electric power industry is improved.

The invention provides a visual question and answer method based on a domain self-adaptive retrieval decision, which comprises the following steps of: forming an input triple (x, q, d) comprising an image, a question text and an image description text; feature modal extraction and domain identification; generating an explicit reasoning track and a preliminary answer by using a chain reasoning technology CoT; according to a preset decision rule, judging that the preliminary answer is output as a final answer or enters the next step; based on the input triad (x, q, d) and the reasoning track, image retrieval and text retrieval are executed, and an enhanced knowledge set is generated; using the enhanced knowledge set to generate a final answer through a chain reasoning technology CoT, performing credibility verification, and outputting the final answer or a preset unknown identifier according to a verification result; according to the method, through reasoning-driven adaptive retrieval and multi-modal knowledge reordering, efficient utilization and real-time supplement of external knowledge are realized, and the accuracy and robustness of visual questions and answers are effectively improved.

The invention discloses a retrieval enhancement generation method and system fusing multi-modal and hybrid retrieval, and relates to the technical field of multi-modal fusion and hybrid retrieval, and the method comprises multi-modal information fusion framework construction, row hybrid retrieval execution, adaptive weight distribution and context construction, and structured intelligent content compression and generation. According to the retrieval enhancement generation method and system fusing multi-modal and mixed retrieval, the limitation that a traditional RAG system depends on a single text vector library is broken through by integrating multi-source information, multi-modal information sources such as contract original texts, expert suggestions, dialogue history and external similar cases are integrated through a unified framework, the comprehensiveness of information coverage is ensured, and the retrieval enhancement generation efficiency is improved. According to the method, more abundant factual basis is provided for result generation, and for double guarantee of semantics and keywords, a parallel mixed retrieval method is adopted to be combined with vector retrieval and full-text retrieval, so that the defect of a single retrieval mode is avoided, and the accuracy and correlation of retrieval results are improved.