Efficient interaction method and system for connecting large model and multi-source data

Abstract

The invention belongs to the technical field of artificial intelligence, and discloses an efficient interaction method and system for connecting a large model and multi-source data, and the method comprises the steps: constructing a dynamic analysis template comprising a format constraint sub-template and a semantic mapping sub-template by collecting structured, semi-structured and non-structured multi-source heterogeneous data, carrying out data hierarchical analysis based on the standard data unit to obtain a standard data unit; constructing a multi-source data semantic association graph based on semantic tags in the standard data unit, generating a multi-source data interaction sequence, and inputting attention weight by updating a large model to obtain an interaction response result; the method has the advantages that efficient input of multi-source heterogeneous data is achieved, a structured semantic interaction basis is provided, dynamic segmentation adapts to different data formats and semantic features, interaction consistency is ensured through semantic association graph analysis, interaction comprehensiveness is improved through multi-modal feature extraction, response reliability is enhanced through interaction sequence optimization, and the method is suitable for large-scale popularization and application. And technical support is provided for application of a large model to a complex data environment.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and more specifically, to an efficient interaction method and system for connecting large models with multi-source data. Background Technology

[0002] With the rapid development of artificial intelligence (AI) technology, large-scale models have been widely applied in various fields such as natural language processing, image recognition, and intelligent decision-making. In the modern information environment, data exhibits multi-source heterogeneous characteristics, including structured data (such as database records), semi-structured data (such as XML / JSON files), and unstructured data (such as text and images). Efficiently connecting large-scale models with multi-source data has become a key challenge for the practical application of AI. In recent years, various data processing and model interaction technologies have emerged, aiming to improve the ability of large-scale models to understand and process complex data. However, existing technologies still have many shortcomings in practical applications.

[0003] Traditional methods for large-scale model data interaction often lack a unified and efficient processing framework when dealing with heterogeneous data. For example, in enterprise intelligent analytics scenarios, data may simultaneously originate from structured business databases, semi-structured API interfaces, and unstructured documents. Existing methods typically employ fragmented processing strategies, failing to effectively integrate the format characteristics and semantic connotations of different data types, resulting in incomplete or redundant information input into the large model. Furthermore, existing technologies rely excessively on preset templates in data parsing, lacking dynamic adaptability and struggling to cope with changes in data format and semantics. This can lead to the loss or misunderstanding of important information, impacting the response quality of the large model.

[0004] Meanwhile, existing technologies have significant shortcomings in semantic association analysis of multi-source data. The key to large-scale models understanding data lies in capturing the semantic connections between different data sources, such as the relationship between customer records in business data and user feedback on social media. However, existing systems lack the ability to model such cross-source semantic relationships, often treating data from different sources in isolation and failing to construct a comprehensive view reflecting real-world semantic relationships. This lack of semantic integration in interaction can lead to one-sided or contradictory responses from large models, especially in complex scenarios requiring comprehensive judgment, severely limiting the model's practicality.

[0005] Furthermore, existing large-model interaction methods often employ static input strategies, making it difficult to adaptively adjust based on model responses. For example, the system may not be able to dynamically optimize data input methods or reorganize information structures based on the initial response of the large model, resulting in low interaction efficiency. Regarding attention mechanisms, existing technologies mostly use fixed weight allocation, lacking dynamic assessment of data importance and unable to flexibly adjust attention points based on semantic changes during the interaction process. This lack of feedback optimization in the interaction mechanism not only reduces the system's response quality but may also lead to wasted computational resources and even misleading results due to the omission of key information.

[0006] In view of this, the present invention proposes an efficient interaction method and system for connecting large models and multi-source data to solve the above problems. Summary of the Invention

[0007] To overcome the aforementioned deficiencies of the prior art and to achieve the above objectives, the present invention provides the following technical solution: An efficient interaction method for connecting large models with multi-source data includes: Step S1: Collect data from the data source connected to the target large model to obtain multi-source heterogeneous data, which includes three types of data: structured data, semi-structured data, and unstructured data. Step S2: Based on the format and semantic features of each type of data in the multi-source heterogeneous data, construct a dynamic parsing template, which includes a format constraint sub-template and a semantic mapping sub-template; Step S3: Based on the constructed dynamic parsing template, perform hierarchical parsing on the corresponding multi-source heterogeneous data to obtain the standard data units corresponding to each type of data. The standard data units include formatted data fragments and semantic tags. Step S4: Construct a multi-source data semantic association graph based on the obtained semantic tags, and generate a multi-source data interaction sequence based on the topological structure of the multi-source data semantic association graph and the obtained formatted data fragments; Step S5: Input the multi-source data interaction sequence into the target large model, and obtain the interaction response results between the corresponding large model and the multi-source data interaction sequence by updating the input attention weight of the target large model. The interaction response results include classification results, generation results and inference results. Step S6: Optimize the generation strategy of the corresponding multi-source data interaction sequence based on the semantic consistency between the interaction response results and the semantic association graph of multi-source data.

[0008] Furthermore, the process of constructing a dynamically parsed template includes: Format features are extracted from various types of data collected from multiple sources and heterogeneous data, and format constraint sub-templates are constructed based on the extracted format features; the format constraint sub-templates include delimiter matching rules and nesting level parsing rules; the format features include data delimiters, nesting levels, and field lengths; Simultaneously, semantic features are extracted from each data type in the collected multi-source heterogeneous data, and a semantic mapping sub-template is constructed based on the semantic features. The semantic mapping sub-template includes keyword semantic vectors and entity relationship weights. The format constraint sub-template is combined with the semantic mapping sub-template to obtain a dynamic parsing template for the corresponding data type.

[0009] Furthermore, the process of performing hierarchical parsing of the corresponding multi-source heterogeneous data based on the constructed dynamic parsing template includes: Based on the constructed format constraint sub-template, the type data is segmented by format features to obtain initial data fragments; and based on the constructed semantic mapping sub-template, semantic features are extracted from the corresponding initial data fragments to obtain semantic labels. Based on the correspondence between the initial data fragments and semantic tags, standard data units are constructed; and the constructed standard data units are stored in a hierarchical manner.

[0010] Furthermore, the process of constructing a semantic association graph of multi-source data includes: The semantic labels within the standard data unit are used as nodes to construct an initial semantic node set; the semantic association strength between any two nodes in the initial semantic node set is obtained, and a directed edge between semantic labels is constructed based on the semantic association strength. The initial set of semantic nodes and the directed edges are subjected to topology optimization to obtain a semantic association graph of multi-source data; the topology optimization includes dividing the semantic subgraph based on the initial set of semantic nodes and performing local clustering on the divided semantic subgraph.

[0011] Furthermore, the process of obtaining multi-source data interaction sequences includes: The semantic association graph of the multi-source data is traversed to obtain multiple semantic association paths, which include a sequence of nodes sorted by the weight of the directed edges. Based on the node sequence within the semantic association path, formatted data fragments of corresponding standard data units are extracted to generate an initial interactive data block; and the initial interactive data block is semantically compressed to obtain a compressed interactive data block. Based on the edge weights of each semantic association path in the multi-source data semantic association graph, the compressed interactive data blocks are sorted to generate a multi-source data interaction sequence.

[0012] Furthermore, the process of obtaining the interactive response result includes: For each compressed interactive data block in the obtained multi-source data interaction sequence, extract its corresponding semantic feature vector; Based on the type of data corresponding to the compressed interactive data block, calculate the field weights of the compressed interactive data block corresponding to structured data, the hierarchical weights of the compressed interactive data block corresponding to semi-structured data, and the semantic weights of the compressed interactive data block corresponding to unstructured data. Based on the correlation between the semantic feature vector and the input layer neurons of the large model, and combined with the field weights, hierarchical weights, and semantic weights, the attention weight of each interactive data block is calculated. Graph structure enhancement is performed on the semantic feature vector of the compressed interactive data block to obtain the enhanced semantic feature vector, and the attention weights are updated based on the dot product of the enhanced semantic feature vector and the weight vector of the input layer neuron. The multi-source data interaction sequence is weighted based on the attention weights to obtain the initial response result of the large model, and the semantic consistency of the initial response result is checked to obtain the final interaction response result.

[0013] Furthermore, the process of optimizing the generation of the corresponding multi-source data interaction sequence includes: Based on the matching degree between the semantic vector corresponding to the interactive response result and the semantic vector corresponding to each semantic label in the semantic association graph of multi-source data, redundant interactive data blocks in the multi-source data interaction sequence are identified. Based on the confidence distribution corresponding to the interaction response results, low semantic importance compressed interaction data blocks in the multi-source data interaction sequence are identified, and corresponding data optimization suggestions are generated through the internal gradient analysis of the large model. The semantic association paths corresponding to redundant interactive data blocks and compressed interactive data blocks with low semantic importance are pruned, and the multi-source data interaction sequence is regenerated based on the pruned multi-source data semantic association graph. The generation strategy for multi-source data interaction sequences is updated by comparing the semantic consistency between the regenerated multi-source data interaction sequences and the corresponding interaction response results of the original multi-source data interaction sequences.

[0014] Furthermore, the process of performing semantic consistency verification includes: Obtain the semantic vector of the initial response result; and obtain the cosine similarity between the semantic vector corresponding to the initial response result and the semantic vector corresponding to the original multi-source data interaction sequence; Set a preset consistency threshold; if the consistency level is less than the preset consistency threshold, reacquire the response results of the large model and repeat this process; if the consistency level is not less than the preset consistency threshold, output the corresponding initial response results as the final interactive response results.

[0015] Furthermore, the process of partitioning the semantic subgraph based on the initial set of semantic nodes and performing local clustering on the partitioned semantic subgraph includes: Cluster the nodes corresponding to each semantic label in the initial semantic node set to obtain their corresponding local clustering coefficients; based on the local clustering coefficients, divide the initial directed weighted graph corresponding to the initial semantic node set into multiple semantic subgraphs; Obtain the semantic importance score of each node in the semantic subgraph; simultaneously obtain the directed edge weights corresponding to the nodes and calculate the semantic centrality based on them; perform local clustering of the nodes in the semantic subgraph according to the semantic centrality to establish a structured representation of the semantic cluster, which includes cluster members, boundaries and representative node information.

[0016] An efficient interactive system for connecting large models and multi-source data includes: The data acquisition module is used to acquire data from the data source connected to the target large model to obtain multi-source heterogeneous data, which includes three types of data: structured data, semi-structured data, and unstructured data. The template construction module is used to construct a dynamic parsing template based on the format and semantic features of various types of data in multi-source heterogeneous data. The dynamic parsing template includes a format constraint sub-template and a semantic mapping sub-template. The data parsing module performs hierarchical parsing of the corresponding multi-source heterogeneous data based on the constructed dynamic parsing template, and obtains the standard data units corresponding to each type of data. The standard data units include formatted data fragments and semantic tags. The topology analysis module is used to construct a multi-source data semantic association graph based on the obtained semantic tags, and generate a multi-source data interaction sequence based on the topological structure of the multi-source data semantic association graph and the obtained formatted data fragments. The data interaction module is used to input the multi-source data interaction sequence into the target large model, and obtain the interaction response results between the corresponding large model and the multi-source data interaction sequence by updating the input attention weight of the target large model. The interaction response results include classification results, generation results and inference results. The interactive feedback module is used to optimize the generation strategy of the corresponding multi-source data interaction sequence based on the semantic consistency between the interactive response results and the semantic association graph of multi-source data.

[0017] This invention provides an efficient interaction method and system for connecting large models and multi-source data, along with its technical effects and advantages: This invention addresses the challenges of large models processing multi-source heterogeneous data, improving the adaptability and accuracy of data interaction and providing strong support for the application of large models in complex data environments. By constructing dynamic parsing templates and semantic association graphs, the system can accurately identify semantic relationships between different types of data, avoiding information comprehension biases or omissions caused by differences in data formats and structures. This not only improves the ability of large models to process complex data but also enhances the consistency and reliability of response results, enabling the model to more effectively integrate multi-source information and generate more comprehensive and accurate analysis results. Furthermore, the dynamic attention weight adjustment and interaction sequence optimization mechanisms enable the system to adaptively adjust according to actual needs, providing more accurate decision support for professional users in different fields. This intelligent data interaction framework not only improves the accuracy and reliability of large-scale model applications, but also provides a technical foundation for cross-domain knowledge integration and innovative exploration, promoting the expansion of artificial intelligence technology to deeper and broader application scenarios. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of an efficient interaction method for connecting large models and multi-source data according to the present invention; Figure 2 This is a schematic diagram of an efficient interactive system for connecting large models and multi-source data according to the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Example 1 Please see Figure 1 As shown in this embodiment, an efficient interaction method for connecting large models and multi-source data includes: Step S1: Collect data from the data source connected to the target large model to obtain multi-source heterogeneous data, which includes three types of data: structured data, semi-structured data, and unstructured data. Step S2: Based on the format and semantic features of each type of data in the multi-source heterogeneous data, construct a dynamic parsing template, which includes a format constraint sub-template and a semantic mapping sub-template; Step S3: Based on the constructed dynamic parsing template, perform hierarchical parsing on the corresponding multi-source heterogeneous data to obtain the standard data units corresponding to each type of data. The standard data units include formatted data fragments and semantic tags. Step S4: Construct a multi-source data semantic association graph based on the obtained semantic tags, and generate a multi-source data interaction sequence based on the topological structure of the multi-source data semantic association graph and the obtained formatted data fragments; Step S5: Input the multi-source data interaction sequence into the target large model, and obtain the interaction response results between the corresponding large model and the multi-source data interaction sequence by updating the input attention weight of the target large model. The interaction response results include classification results, generation results and inference results. Step S6: Optimize the generation process of the corresponding multi-source data interaction sequence based on the semantic consistency between the interaction response results and the semantic association graph of multi-source data.

[0021] This invention achieves efficient input to large models through multi-source heterogeneous data acquisition and processing. It constructs dynamic parsing templates and semantic association graphs to provide a structured foundation for multi-source data interaction. Interaction sequence generation based on semantic association strength endows the input with semantic organization capabilities. Dynamically segmented data units can adapt to the formats and semantic features of different data types. Combining format constraints and semantic mapping for cross-modal feature extraction improves the comprehensiveness of data interaction. Semantic association graph analysis considers semantic connections between data to ensure interaction consistency. Deep variation feature and semantic consistency analysis overcomes the limitations of traditional single-feature interaction. Multi-source data interaction sequences enhance the reliability and accuracy of large model responses. Optimized generation strategies based on interaction response results improve system adaptability and efficiency.

[0022] It should be further explained that, in the specific implementation process, the acquisition of multi-source heterogeneous data includes the following steps: Structured data is obtained from the data source connected to the target large model to obtain a structured dataset. Field type identification and data cleaning are performed on the structured dataset to obtain standardized structured data. Semi-structured data is obtained from the data source connected to the target large model to obtain a semi-structured dataset. Label extraction and nested hierarchical analysis are performed on the semi-structured dataset to obtain standardized semi-structured data. Unstructured data is obtained from the data source connected to the target large model to obtain an unstructured dataset. The unstructured dataset is then segmented and key information is extracted to obtain standardized unstructured data. Metadata annotation is performed on standardized structured data, standardized semi-structured data, and standardized unstructured data to obtain multi-source heterogeneous data with metadata tags. The multi-source heterogeneous data with metadata tags is then organized according to the collection time and data source.

[0023] In this embodiment, data acquisition terminals are first deployed on various data sources (such as databases, API interfaces, file systems, etc.) connected to the target large model to capture various types of structured data. The structured data is obtained through SQL queries, database connectors, or structured API calls to form a structured dataset containing clearly defined fields and records. The obtained structured dataset is then processed to identify the data type of each field (such as numeric, character, time, etc.) and remove outliers, missing values, and duplicate records to ensure data quality and form structured data. Simultaneously, semi-structured data is collected from the target data source. The semi-structured data is obtained through XML / JSON parsers, web crawlers, or document APIs to generate a semi-structured dataset containing tags and hierarchical structures. The obtained semi-structured dataset is then labeled, key nodes and attributes are identified, the nesting hierarchy of the data is analyzed, and a tree structure representation is constructed to form semi-structured data. Unstructured data is collected from relevant data sources. The unstructured data is obtained through text extractors, image processing APIs or audio transcription services to form an unstructured dataset containing natural language text, images, audio and other content. The obtained unstructured dataset is segmented into paragraphs or semantic units to extract key information (such as topics, entities, relationships, etc.) and form unstructured data. Then, by employing metadata annotation technology, unified metadata tags are added to the three types of data, including data source information, timestamps, data types, and topic classifications. This gives different types of data a unified metadata framework, forming multi-source heterogeneous data with metadata tags. Finally, the multi-source heterogeneous data with metadata tags is organized according to collection time and data source to build time-series indexes and data source indexes, forming a structured multi-source heterogeneous dataset. These multi-source heterogeneous data together constitute the basic data source for large-scale model interaction, providing a comprehensive data foundation for subsequent dynamic parsing template construction and semantic analysis.

[0024] It should be further explained that, in the specific implementation process, the construction process of the dynamically parsed template includes the following steps: Format features are extracted from various types of data collected from multiple sources and heterogeneous data, and format constraint sub-templates are constructed based on the extracted format features; the format constraint sub-templates include delimiter matching rules and nesting level parsing rules; the format features include data delimiters, nesting levels, and field lengths; Simultaneously, semantic features are extracted from each data type in the collected multi-source heterogeneous data, and a semantic mapping sub-template is constructed based on the semantic features. The semantic mapping sub-template includes keyword semantic vectors and entity relationship weights. The format constraint sub-template is combined with the semantic mapping sub-template to obtain a dynamic parsing template for the corresponding data type.

[0025] In this embodiment, firstly, format analysis (such as field splitters and type identifiers) is applied to structured data. By identifying data delimiters (such as commas, tabs, semicolons, etc.), the distribution of field lengths and value ranges are analyzed to extract data format features. For semi-structured data, a pre-built structure parser (such as an XML / JSON parser) is applied to extract nesting level information by identifying tag hierarchy, nesting relationships, and tree structure features. For unstructured data, text format analysis (such as paragraph analysis and syntax parsing) is applied to identify the structural division of content and data format. Based on this, format features corresponding to each type of data are formed, including data delimiters, nesting levels, and field lengths. Based on the extracted format features, different delimiter matching rules and nested hierarchical parsing rules are set for different types of data; for example: regular expression patterns are built for structured data to accurately match field delimiters and data formats; tree-like parsing rules are built for semi-structured data, including node path expressions and nested hierarchical navigation rules; content segmentation rules are built for unstructured data, such as segmentation patterns based on paragraph marks, chapter divisions, or semantic boundaries; and these rules are combined to form format constraint sub-templates; Simultaneously, key fields are identified in structured data to analyze the semantic meaning of field names and values; semantic analysis of labels is performed on semi-structured data to identify hierarchical relationships and semantic connections between semantic labels; natural language processing techniques (such as named entity recognition, relation extraction, and dependency parsing) are applied to unstructured data to extract keyword distribution, entity relationships, and contextual dependency structures from the text, obtaining semantic features of each type of data. Then, keywords are converted into semantic vector representations to capture the semantic information and contextual relationships of words; graph networks are applied to model entity relationships, learning the strength and semantic importance of relationships between entities, calculating contextual dependency, and using it as entity relationship weights to quantify the importance and dependence of words or entities in the context, combining these semantic representations to form semantic mapping sub-templates; pre-trained language models and knowledge graphs are used to assist in the semantic extraction process of each type of data to improve the semantic richness and accuracy of features; Finally, based on the type data and feature distribution, the weight ratio of format constraints and semantic mapping is dynamically adjusted; and based on this, the parsing results of the format constraint sub-template and the semantic results of the semantic mapping sub-template are integrated; a type data-specific dynamic parsing template is constructed, which includes format constraint rules and semantic mapping relationships, forming a complete dynamic parsing framework. This dual template construction method based on format features and semantic features can accurately capture the structural features and semantic connotations of different types of data, providing precise guidance for subsequent hierarchical parsing and improving the system's adaptability to complex heterogeneous data.

[0026] It should be further explained that, in the specific implementation process, the process of performing hierarchical parsing of the corresponding multi-source heterogeneous data based on the constructed dynamic parsing template includes: Based on the constructed format constraint sub-template, the type data is segmented by format features to obtain initial data fragments; and based on the constructed semantic mapping sub-template, semantic features are extracted from the corresponding initial data fragments to obtain semantic labels. Based on the correspondence between the initial data fragments and semantic tags, standard data units are constructed; and the constructed standard data units are stored in a hierarchical manner. The hierarchical storage includes storing the standard data units in different cache layers based on the priority of the semantic tags, wherein the priority is determined by the semantic importance score of the semantic tags, and the semantic importance score is calculated by the frequency of occurrence and contextual dependency of the semantic tags in the multi-source heterogeneous data.

[0027] In this embodiment, data types are segmented based on format constraint templates to obtain initial data fragments. Format feature segmentation divides the data into basic units according to the identified delimiters and nested structures; these units represent the smallest processing granularity of the data. The segmentation process adopts a bottom-up strategy, first identifying the most basic field delimiters, and then processing nested structures level by level to ensure the completeness and accuracy of the segmentation. For example, delimiter matching rules in the format constraint sub-template are applied to structured data to split the data into multiple field values ​​according to field delimiters; nested hierarchical parsing rules are applied to semi-structured data to parse XML / JSON and other data into sets of nodes and attributes; and content segmentation is applied to unstructured data to divide text into paragraphs, sentences, or semantic units. Based on the constructed semantic mapping template, semantic features are extracted from the corresponding initial data fragments to obtain semantic labels. Semantic labels include entity categories (such as people, organizations, places, times, etc.) and relationship categories (such as subordinate relationships, causal relationships, temporal relationships, etc.). The extraction process utilizes pre-trained language models and knowledge graphs to map data fragments to a semantic space and identify their semantic categories and relational attributes. For example, the semantic mapping template is applied to the obtained initial data fragments to identify key concepts and semantic units, and entity categories (such as names, places, times, organizations, etc.) are marked. Relation extraction techniques are used to identify the relationship categories between entities (such as subordinate relationships, causal relationships, etc.) to obtain the semantic labels of the initial data fragments. Based on the correspondence between initial data fragments and semantic tags, standard data units are constructed. The construction process establishes the correspondence between fields and entities / relationships by analyzing the mapping relationship between initial data fragments and semantic tags. The integrity index of each field is calculated, including the field missing rate (proportion of null values) and redundancy rate (proportion of duplicate information). An integrity threshold is set (e.g., missing rate <10% and redundancy rate <20%), and initial data fragments that meet the threshold are selected. The selected set of fields is combined with the corresponding semantic tags to construct standard data units. Each standard data unit contains formatted data fragments and corresponding semantic tags. Next, by analyzing the frequency and contextual dependency of each semantic tag in the corresponding multi-source heterogeneous data, and by weighting and summing the frequency and contextual dependency, a semantic importance score is obtained to determine the priority of the semantic tags; the formula for calculating the contextual dependency is as follows: In the formula, This represents the degree of contextual dependency corresponding to the t-th semantic representation; , represents the probability of the occurrence of the t0th semantic tag given that the tth semantic tag has occurred; N represents the total number of semantic tags; t0 represents the index of the other semantic tags excluding the tth semantic tag; This represents the total number of times the semantic tag t appears; This indicates the number of times semantic labels t0 and t co-occur. , representing the probability of the occurrence of the t0th semantic tag given that the t-th semantic tag has occurred. This represents the probability of the t-th semantic tag occurring given that the t0-th semantic tag has occurred. This represents the probability of the occurrence of the t0th semantic tag; The design incorporates a multi-tiered caching architecture, including a high-priority caching layer (for storing standard data units corresponding to highly important semantic tags), a medium-priority caching layer, and a low-priority caching layer. Standard data units are allocated to the appropriate caching layer based on the priority of the semantic tags. Caching strategies are optimized, such as using faster storage media or higher access frequencies for the high-priority layer. This hierarchical storage method based on semantic importance enables efficient management and access to standard data units from heterogeneous data, providing a structured and prioritized data foundation for subsequent semantic association graph construction and interaction sequence generation, thereby improving the efficiency of large models interacting with multi-source data.

[0028] It should be further explained that, in the specific implementation process, the construction of the multi-source data semantic association graph includes the following steps: The semantic labels within the standard data unit are used as nodes to construct an initial semantic node set; the semantic association strength between any two nodes in the initial semantic node set is obtained, and a directed edge between semantic labels is constructed based on the semantic association strength. Based on the initial set of semantic nodes and the directed edges, topology optimization is performed to obtain a semantic association graph of multi-source data; the topology optimization includes dividing the semantic subgraph based on the initial set of semantic nodes and performing local clustering on the divided semantic subgraph.

[0029] In this embodiment, semantic labels in all standard data units are first collected, duplicate labels are removed, unique identifiers are established, and an initial set of semantic nodes is formed. The set of semantic nodes consists of several nodes, and the nodes contain attributes such as label ID, semantic category, semantic vector, and semantic importance score. For each node in the set, its semantic vector is extracted, and the cosine similarity between any two node semantic vectors is calculated. The co-occurrence frequency of any two nodes in multi-source heterogeneous data is analyzed, that is, the frequency at which two labels appear simultaneously in the same standard data unit or context. A weighted formula is designed, such as semantic association strength = α × cosine similarity + β × normalized co-occurrence frequency, where α and β are weight parameters, to obtain the semantic association strength between the corresponding nodes. The weight parameters are dynamically adjusted according to the data characteristics. Typically, the semantic similarity weight is 0.6-0.7, and the co-occurrence frequency weight is 0.3-0.4, to ensure that the association strength considers both semantic and statistical factors.

[0030] Then, the contextual relationships of semantic tags in the original data are analyzed, such as determining the dependency direction between semantic tags through dependency parsing or sequence positional relationships; Define direction determination rules, such as information flow (from descriptive tags to specific instances) or causal relationships (from cause to effect); determine the direction based on the direction determination rules, and create directed edges for each pair of related semantic tags based on them; The calculated semantic association strength is normalized (e.g., Min-Max normalization or Softmax normalization) and converted into a weight value between 0 and 1; the normalized weight is assigned to each directed edge to obtain the initial directed weighted graph; Based on the initial set of semantic nodes and the directed edges, topology optimization is performed to obtain a multi-source data semantic association graph. Specifically, based on the initial set of semantic nodes, the initial directed weighted graph is divided into semantic subgraphs with tightly connected internal connections and sparse external connections, and local clustering is performed on the internal nodes in the divided semantic subgraphs. By clustering nodes, the overall complexity and edge density of the graph are reduced. The optimized graph structure retains the most important semantic relationships while reducing redundant connections, forming a multi-source data semantic association graph. This graph is a directed graph structure with semantic labels as nodes and semantic association strength as edge weights, accurately representing the semantic relationship network in multi-source heterogeneous data. Through this semantic association graph construction method, semantic connections in different data sources and types can be captured, providing a structured semantic navigation framework for subsequent interaction sequence generation and improving the ability of large models to understand the semantic structure of data.

[0031] It should be further explained that, in the specific implementation process, the acquisition of multi-source data interaction sequences includes the following steps: The semantic association graph of the multi-source data is traversed to obtain multiple semantic association paths, which include a sequence of nodes sorted by the weight of the directed edges. Based on the node sequence within the semantic association path, formatted data fragments of corresponding standard data units are extracted to generate an initial interactive data block; and the initial interactive data block is semantically compressed to obtain a compressed interactive data block. Based on the edge weights of each semantic association path in the multi-source data semantic association graph, the compressed interactive data blocks are sorted to generate a multi-source data interaction sequence.

[0032] In this embodiment, firstly, a graph traversal is performed on the constructed multi-source data semantic association graph (implemented by depth-first search (DFS), breadth-first search (BFS), random walk, etc.). Starting from the high-score nodes based on the semantic importance scores of the semantic tags, the path is explored along the directed edges; and the path selection is based on the edge weights (semantic association strength), prioritizing the traversal of edges with higher weights; thus, multiple paths with high semantic association are obtained, each path consisting of a series of nodes sorted by association strength, forming a set of semantically associated paths. Then, for each semantically associated path, the corresponding semantic label is matched in the standard data unit for the node sequence; and formatted data fragments are extracted from the matched standard data units; the data fragments on the same path are combined into an initial interactive data block, while maintaining the order relationship of the node sequence, to form a structured initial interactive data representation; the initial interactive data block contains the original formatted data and necessary metadata (such as data source, semantic labels, etc.). Next, the semantic vector of the obtained initial interactive data block is reduced in dimensionality based on the pre-selected variational autoencoder, retaining the main components of the semantic vector, removing redundant and noise information, and generating semantically compressed interactive data blocks. During the dimensionality reduction process, the degree of dimensionality reduction needs to be adjusted according to the semantic information retention rate. Usually, 80% to 90% of the semantic information is retained, ensuring semantic integrity while reducing the amount of data.

[0033] Finally, the compressed interactive data blocks are sorted according to the weights of each semantic association path in the multi-source data semantic association graph. The sorting is based on the product of the weights of the semantic association paths and the semantic importance scores of the compressed interactive data blocks, resulting in a comprehensive priority score. All compressed interactive data blocks are sorted according to their comprehensive priority scores, with high-priority data blocks placed at the beginning of the sequence. The sorted compressed interactive data blocks are then organized to form a multi-source data interaction sequence with semantic order. This sequence is sorted according to the strength of semantic association and contains the most relevant and important information extracted from multi-source heterogeneous data. This method of generating interaction sequences based on semantic association graphs can provide large models with structured, high-information-density, and semantically coherent input data, improve the efficiency of large models in understanding and processing multi-source data, and provide a foundation for generating high-quality interactive responses.

[0034] It should be further explained that, in the specific implementation process, the process of obtaining the interactive response result includes the following steps: For each compressed interactive data block in the obtained multi-source data interaction sequence, extract its corresponding semantic feature vector; Based on the type of data corresponding to the compressed interactive data block, calculate the field weights of the compressed interactive data block corresponding to structured data, the hierarchical weights of the compressed interactive data block corresponding to semi-structured data, and the semantic weights of the compressed interactive data block corresponding to unstructured data. Based on the correlation between the semantic feature vector and the input layer neurons of the large model, and combined with the field weights, hierarchical weights, and semantic weights, the attention weight of each interactive data block is calculated. Graph structure enhancement is performed on the semantic feature vector of the compressed interactive data block to obtain the enhanced semantic feature vector, and the attention weights are updated based on the dot product of the enhanced semantic feature vector and the weight vector of the input layer neuron. The multi-source data interaction sequence is weighted based on the attention weights to obtain the initial response result of the large model, and the semantic consistency of the initial response result is checked to obtain the final interaction response result.

[0035] In this embodiment, semantic vector analysis is first performed on each compressed interactive data block in the multi-source data interaction sequence to extract its semantic representation in context awareness; and word embedding representations (such as Word2Vec, GloVe, etc.) of the semantic tags corresponding to the compressed interactive data blocks are obtained; the corresponding semantic representations and word embedding representations are converted into encoded vectors and concatenated to obtain semantic feature vectors; Then, for the compressed interactive data blocks corresponding to structured data, the completeness of their fields is analyzed, and the proportion of non-empty fields and data quality indicators are calculated to obtain field completeness scores, which are then normalized into field weights. For the compressed interactive data blocks corresponding to semi-structured data, the nesting depth of their XML / JSON structures is analyzed to obtain hierarchical weights; the deeper the hierarchy, the more specific or important the information may be. For the compressed interactive data blocks corresponding to unstructured data, the distribution entropy of their semantic vectors is calculated to quantify the richness and diversity of information, thus obtaining semantic weights. Next, the input layer structure of the target large model is analyzed, the weight vectors of neurons in the input layer are extracted, and the dot product of the semantic feature vector corresponding to the compressed interactive data block and the input layer weight vector is calculated to obtain the correlation between the semantic feature vector and the input layer neurons of the large model. Combined with the weights specific to the data type (including field weights, hierarchical weights or semantic weights), the initial attention weights corresponding to each compressed interactive data block are calculated by weighted summation. The semantic feature vectors of the compressed interactive data blocks are augmented using graph structure enhancement to obtain enhanced semantic feature vectors. Graph structure enhancement utilizes the topological information of the semantic association graph of multi-source data and fuses the features of related nodes through a message passing mechanism to enrich the semantic representation of each compressed interactive data block. During the enhancement process, the system dynamically adjusts the message passing weights according to the strength of the relationships between nodes to ensure that important relationships are fully reflected. The enhanced semantic feature vectors contain richer contextual information, which helps to improve the interaction effect with large models. The correlation between the enhanced semantic feature vectors and the weight vectors of the input layer neurons is recalculated to update the attention weights. The message passing mechanism refers to an existing technical method that fuses the features of related nodes using the topological information of the semantic association graph of multi-source data during the graph structure enhancement process. Then, the updated attention weights are applied to the multi-source data interaction sequence to achieve weighted input of the data. Compressed interaction data blocks with high attention receive more attention from the large model. The weighted interaction sequence is then input into the target large model to obtain the initial response results, including classification results, generation results, and inference results. Finally, the semantic vector of the initial response result is obtained (refer to the semantic vector acquisition process described above); the cosine similarity between the semantic vector corresponding to the initial response result and the semantic vector corresponding to the original multi-source data interaction sequence is obtained to quantify the degree of semantic consistency; a preset consistency threshold is set (e.g., 0.7 or 0.8); if the consistency degree is less than the preset consistency threshold, the attention weights are dynamically adjusted through gradient backpropagation, and the response result of the large model is re-obtained, and this process is repeated until the consistency requirement is met to obtain the final interaction response result; if the consistency degree is not less than the preset consistency threshold, the corresponding initial response result is output as the final interaction response result; through this method of dynamically adjusting attention weights, the system can optimize the attention allocation of the large model to multi-source data, improve the semantic consistency between the model output and the input data, ensure the quality and relevance of the interaction response, and provide reliable feedback signals for subsequent interaction strategy optimization.

[0036] It should be further explained that, in the specific implementation process, the optimization of the generation process of the corresponding multi-source data interaction sequence includes the following steps: Based on the matching degree between the semantic vector corresponding to the interactive response result and the semantic vector corresponding to each semantic label in the semantic association graph of multi-source data, redundant interactive data blocks in the multi-source data interaction sequence are identified. Based on the confidence distribution corresponding to the interaction response results, low semantic importance compressed interaction data blocks in the multi-source data interaction sequence are identified, and corresponding data optimization suggestions are generated through the internal gradient analysis of the large model. The semantic association paths corresponding to redundant interactive data blocks and compressed interactive data blocks with low semantic importance are pruned, and the multi-source data interaction sequence is regenerated based on the pruned multi-source data semantic association graph. The generation strategy for multi-source data interaction sequences is updated by comparing the semantic consistency between the regenerated multi-source data interaction sequences and the corresponding interaction response results of the original multi-source data interaction sequences.

[0037] In this embodiment, firstly, the semantic vector corresponding to the interaction response result is obtained, and the semantic vectors of each semantic label in the multi-source data semantic association graph are extracted; the Euclidean distance between the semantic vector corresponding to the interaction response result and the semantic vectors corresponding to each semantic label is obtained; wherein, the smaller the Euclidean distance, the higher the matching degree. Set a matching threshold, compare the obtained Euclidean distance with the corresponding matching threshold, and identify semantic tags that highly match the interaction response results based on the comparison results; and treat the compressed data blocks corresponding to non-highly matched semantic tags in the multi-source data interaction sequence as potential redundant data blocks. Then, by analyzing the interaction response results of the large model (such as the category probability distribution of the classification results, the token prediction probability of the generated results, and the logical link strength of the inference results), low-confidence regions are identified. Low-confidence regions refer to the input parts that are uncertain or difficult to handle by the large model. Through techniques such as backpropagation or attention visualization, the attention allocation of each compressed interactive data block of the input within the large model is analyzed. Several compressed interactive data blocks with lower attention weights are identified and obtained. Compressed interactive data blocks can be regarded as having lower semantic importance. Furthermore, based on internal gradient and attention analysis, data optimization suggestions are generated, such as adjusting the weights of specific semantic mapping relationships and modifying the parsing strategy for specific types of data. Next, redundant interactive data blocks and compressed interactive data blocks with low semantic importance are marked, and the semantic association paths corresponding to these data blocks are traced. Directed edges with weights less than a preset weight threshold are identified in these paths (the threshold is set based on the statistical distribution of edge weights, usually 60% to 70% of the distribution mean, to ensure that only edges that contribute less are pruned). Edges with weights less than the preset weight threshold are deleted from the semantic association graph of multi-source data, thereby achieving pruning optimization of the graph structure. Finally, a new multi-source data interaction sequence is generated based on the pruned semantic association graph and input into the large model to obtain new interaction response results. The semantic consistency between the new interaction response results and the original interaction response results is obtained by using the cosine similarity between semantic vectors. If the consistency remains within a preset consistency range, it indicates that the pruning is effective and redundant information has been successfully removed. Based on this, the weights of the semantic association paths are adjusted to update the generation strategy of the multi-source data interaction sequence and improve the interaction efficiency. The generation strategy includes adjusting the weight calculation formula of the semantic association paths, optimizing the path traversal algorithm, and improving the data block sorting mechanism.

[0038] An embodiment of the present invention further includes: updating the dynamic parsing template based on the semantic consistency of the interaction response results, the specific implementation process of which includes: Based on the semantic consistency of the interactive response results, annotation errors in the semantic tags within the standard data units are identified. Annotation errors refer to situations where the entity category of a semantic tag is inconsistent with the semantic category of the interactive response result. This inconsistency may stem from inaccurate initial annotations or changes in data features. The identification process compares the tag category with the semantic category of the corresponding content in the response, calculates the inconsistency rate, and tags exceeding a preset threshold are marked as potential errors.

[0039] The algorithm obtains dynamic parsing templates corresponding to semantic tags with annotation errors, and adjusts the keyword semantic vectors and entity relation weights in the semantic mapping sub-templates within the corresponding dynamic parsing templates. The adjustment process uses a backpropagation algorithm to fine-tune the keyword semantic vectors to make them more consistent with the correct semantic categories. At the same time, it optimizes the entity relation weights through the edge weight update rules of the graph neural network. The edge weight update rules include weights that strengthen correct relations and weights that weaken incorrect relations.

[0040] Based on the format integrity of the interactive response result, identify parsing errors in the formatted data fragments in the standard data unit; parsing errors include missing fields (should-have fields are not recognized) or redundancy (illegal fields are incorrectly included); the identification method is to detect differences by comparing the actual structure of the formatted data fragment with the expected structure; record the error type, frequency and distribution characteristics to form an error feature map, providing detailed basis for adjusting the format constraint sub-template.

[0041] Obtain the dynamic parsing template corresponding to the formatted data fragment with parsing errors, and adjust the delimiter matching rules and nested parsing rules in the format constraint sub-template. The adjustment process first statistically analyzes the distribution characteristics of the set of parsing errors, such as the frequency distribution of error location and error type. Then, it updates the regular expression of the delimiter matching rule to optimize pattern recognition capabilities. Finally, it improves the nested parsing rules by retraining the tree-structured parsing model.

[0042] Based on the analysis of interaction response results (such as the category probability distribution of classification results, the token prediction probability of generated results, and the logical link strength of inference results), proactive optimization suggestions are generated for the dynamic parsing template. These proactive optimization suggestions refer to template improvement schemes actively proposed based on the interaction response results, including suggestions for prioritizing the delimiter matching rules of format constraint sub-templates and suggestions for enhancing the entity relationship weights of semantic mapping sub-templates. Suggestion generation employs a reinforcement learning method, using the quality of interaction response results as a reward signal to guide the exploration of better template configurations.

[0043] It should be further explained that the specific implementation of semantic subgraph partitioning based on the initial semantic node set, and the process of local clustering of the partitioned semantic subgraphs, includes: Clustering is performed on the nodes corresponding to each semantic label in the initial semantic node set to obtain their corresponding local clustering coefficients. The local clustering coefficient is an indicator of the tightness of the neighborhood connections between nodes, calculated as the ratio of the number of directed edges between a node's neighbors to the total number of possible directed edges. The corresponding mathematical formula is: In the formula, This represents the local clustering coefficient corresponding to the i-th node; This represents the number of nodes that have an edge with the i-th node; This represents the number of nodes adjacent to the i-th node; Based on the local clustering coefficients, the initial directed weighted graph corresponding to the initial set of semantic nodes is divided into multiple semantic subgraphs. The partitioning process adopts the principle of maximizing the sum of local clustering coefficients within the subgraphs, and finds the optimal partitioning scheme by iterative optimization using optimization algorithms (such as greedy algorithms, simulated annealing, etc.). The final boundary of the semantic subgraph is determined by iteratively adjusting the node allocation. The initial set of semantic nodes is divided into multiple semantic subgraphs with high internal connection density. For each node within a semantic subgraph, its semantic importance score in the graph is calculated, such as through centrality metrics (PageRank, betweenness centrality, etc.) or semantic importance scores. The weights of the directed edges corresponding to the nodes are obtained, and semantic centrality is calculated based on them. The metrics comprehensively consider the node's connection strength and semantic importance. The corresponding mathematical calculation formula is as follows: ;in, It is a node With neighboring nodes The weight of the directed edges between them It is a node The semantic importance score; Based on the semantic centrality, nodes within the semantic subgraph are locally clustered. The clustering process is based on the semantic similarity and connection patterns between nodes, merging nodes gradually from bottom to top. Specifically, by identifying the node with the highest semantic centrality in each semantic cluster, it is set as the representative node or central node of the cluster, establishing a structured representation of the semantic cluster, including cluster members, boundaries, and representative node information. Through this community detection and local clustering method based on local clustering coefficients and semantic centrality, the system can identify cohesive semantic concept groups in the semantic association graph, reduce the complexity of the graph, and preserve the core semantic structure. This provides an optimized semantic navigation framework for subsequent interaction sequence generation, improving the efficiency of large models in understanding and processing the semantic structure of multi-source data.

[0044] Example 2 Please see Figure 2 As shown, parts not described in detail in this embodiment are described in Embodiment 1. This embodiment provides an efficient interactive system for connecting large models and multi-source data, including: The data acquisition module is used to acquire data from the data source connected to the target large model to obtain multi-source heterogeneous data, which includes three types of data: structured data, semi-structured data, and unstructured data. The template construction module is used to construct a dynamic parsing template based on the format and semantic features of various types of data in multi-source heterogeneous data. The dynamic parsing template includes a format constraint sub-template and a semantic mapping sub-template. The data parsing module performs hierarchical parsing of the corresponding multi-source heterogeneous data based on the constructed dynamic parsing template, and obtains the standard data units corresponding to each type of data. The standard data units include formatted data fragments and semantic tags. The topology analysis module is used to construct a multi-source data semantic association graph based on the obtained semantic tags, and generate a multi-source data interaction sequence based on the topological structure of the multi-source data semantic association graph and the obtained formatted data fragments. The data interaction module is used to input the multi-source data interaction sequence into the target large model, and obtain the interaction response results between the corresponding large model and the multi-source data interaction sequence by updating the input attention weight of the target large model. The interaction response results include classification results, generation results and inference results. The interactive feedback module is used to optimize the generation strategy of the corresponding multi-source data interaction sequence based on the semantic consistency between the interactive response results and the semantic association graph of multi-source data.

[0045] The modules are connected via wired and / or wireless means to enable data transmission between them.

[0046] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0047] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0048] In the description of this invention, it should be understood that the terms "first," "second," etc., are used only for distinguishing descriptions and should not be construed as indicating or implying relative importance.

[0049] In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0050] In the description of this invention, "several" means one or more, and "a large number" means two or more.

[0051] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0052] All formulas in this manual are dimensionless and calculated numerically. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0053] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. An efficient interaction method for connecting large models and multi-source data, characterized in that, include: Step S1: Collect data from the data source connected to the target large model to obtain multi-source heterogeneous data, which includes three types of data: structured data, semi-structured data, and unstructured data. Step S2: Based on the format and semantic features of each type of data in the multi-source heterogeneous data, construct a dynamic parsing template, which includes a format constraint sub-template and a semantic mapping sub-template; Step S3: Based on the constructed dynamic parsing template, perform hierarchical parsing on the corresponding multi-source heterogeneous data to obtain the standard data units corresponding to each type of data. The standard data units include formatted data fragments and semantic tags. Step S4: Construct a multi-source data semantic association graph based on the obtained semantic tags, and generate a multi-source data interaction sequence based on the topological structure of the multi-source data semantic association graph and the obtained formatted data fragments; Step S5: Input the multi-source data interaction sequence into the target large model, and obtain the interaction response results between the corresponding large model and the multi-source data interaction sequence by updating the input attention weight of the target large model. The interaction response results include classification results, generation results and inference results. Step S6: Optimize the generation strategy of the corresponding multi-source data interaction sequence based on the semantic consistency between the interaction response results and the semantic association graph of multi-source data.

2. The efficient interaction method for connecting large models and multi-source data according to claim 1, characterized in that, The process of constructing a dynamically parsed template includes: Format features are extracted from various types of data collected from multiple sources and heterogeneous data, and format constraint sub-templates are constructed based on the extracted format features; the format constraint sub-templates include delimiter matching rules and nesting level parsing rules; the format features include data delimiters, nesting levels, and field lengths; Simultaneously, semantic features are extracted from each data type in the collected multi-source heterogeneous data, and a semantic mapping sub-template is constructed based on the semantic features. The semantic mapping sub-template includes keyword semantic vectors and entity relationship weights. The format constraint sub-template is combined with the semantic mapping sub-template to obtain a dynamic parsing template for the corresponding data type.

3. The efficient interaction method for connecting large models and multi-source data according to claim 2, characterized in that, The process of performing hierarchical parsing of corresponding multi-source heterogeneous data based on the constructed dynamic parsing template includes: Based on the constructed format constraint sub-template, the type data is segmented by format features to obtain initial data fragments; and based on the constructed semantic mapping sub-template, semantic features are extracted from the corresponding initial data fragments to obtain semantic labels. Based on the correspondence between the initial data fragments and semantic tags, standard data units are constructed; and the constructed standard data units are stored in a hierarchical manner.

4. The efficient interaction method for connecting large models and multi-source data according to claim 3, characterized in that, The process of constructing a semantic association graph of multi-source data includes: The semantic labels within the standard data unit are used as nodes to construct an initial semantic node set; the semantic association strength between any two nodes in the initial semantic node set is obtained, and a directed edge between semantic labels is constructed based on the semantic association strength. The initial set of semantic nodes and the directed edges are subjected to topology optimization to obtain a semantic association graph of multi-source data; the topology optimization includes dividing the semantic subgraph based on the initial set of semantic nodes and performing local clustering on the divided semantic subgraph.

5. The efficient interaction method for connecting large models and multi-source data according to claim 4, characterized in that, The process of obtaining multi-source data interaction sequences includes: The semantic association graph of the multi-source data is traversed to obtain multiple semantic association paths, which include a sequence of nodes sorted by the weight of the directed edges. Based on the node sequence within the semantic association path, formatted data fragments of corresponding standard data units are extracted to generate an initial interactive data block; and the initial interactive data block is semantically compressed to obtain a compressed interactive data block. Based on the edge weights of each semantic association path in the multi-source data semantic association graph, the compressed interactive data blocks are sorted to generate a multi-source data interaction sequence.

6. The efficient interaction method for connecting large models and multi-source data according to claim 5, characterized in that, The process of obtaining the interactive response result includes: For each compressed interactive data block in the obtained multi-source data interaction sequence, extract its corresponding semantic feature vector; Based on the type of data corresponding to the compressed interactive data block, calculate the field weights of the compressed interactive data block corresponding to structured data, the hierarchical weights of the compressed interactive data block corresponding to semi-structured data, and the semantic weights of the compressed interactive data block corresponding to unstructured data. Based on the correlation between the semantic feature vector and the input layer neurons of the large model, and combined with the field weights, hierarchical weights, and semantic weights, the attention weight of each interactive data block is calculated. Graph structure enhancement is performed on the semantic feature vector of the compressed interactive data block to obtain the enhanced semantic feature vector, and the attention weights are updated based on the dot product of the enhanced semantic feature vector and the weight vector of the input layer neuron. The multi-source data interaction sequence is weighted based on the attention weights to obtain the initial response result of the large model, and the semantic consistency of the initial response result is checked to obtain the final interaction response result.

7. The efficient interaction method for connecting large models and multi-source data according to claim 6, characterized in that, The process of optimizing the generation of the corresponding multi-source data interaction sequence includes: Based on the matching degree between the semantic vector corresponding to the interactive response result and the semantic vector corresponding to each semantic label in the semantic association graph of multi-source data, redundant interactive data blocks in the multi-source data interaction sequence are identified. Based on the confidence distribution corresponding to the interaction response results, low semantic importance compressed interaction data blocks in the multi-source data interaction sequence are identified, and corresponding data optimization suggestions are generated through the internal gradient analysis of the large model. The semantic association paths corresponding to redundant interactive data blocks and compressed interactive data blocks with low semantic importance are pruned, and the multi-source data interaction sequence is regenerated based on the pruned multi-source data semantic association graph. The generation strategy for multi-source data interaction sequences is updated by comparing the semantic consistency between the regenerated multi-source data interaction sequences and the corresponding interaction response results of the original multi-source data interaction sequences.

8. The efficient interaction method for connecting large models and multi-source data according to claim 6, characterized in that, The process of performing semantic consistency verification includes: Obtain the semantic vector of the initial response result; and obtain the cosine similarity between the semantic vector corresponding to the initial response result and the semantic vector corresponding to the original multi-source data interaction sequence; Set a preset consistency threshold; if the consistency level is less than the preset consistency threshold, reacquire the response results of the large model and repeat this process; if the consistency level is not less than the preset consistency threshold, output the corresponding initial response results as the final interactive response results.

9. The efficient interaction method for connecting large models and multi-source data according to claim 4, characterized in that, The process of partitioning a semantic subgraph based on an initial set of semantic nodes and then performing local clustering on the partitioned semantic subgraphs includes: Cluster the nodes corresponding to each semantic label in the initial semantic node set to obtain their corresponding local clustering coefficients; based on the local clustering coefficients, divide the initial directed weighted graph corresponding to the initial semantic node set into multiple semantic subgraphs; Obtain the semantic importance score of each node in the semantic subgraph; simultaneously obtain the directed edge weights corresponding to the nodes and calculate the semantic centrality based on them; perform local clustering of the nodes in the semantic subgraph according to the semantic centrality to establish a structured representation of the semantic cluster, which includes cluster members, boundaries and representative node information.

10. A highly efficient interactive system for connecting large models and multi-source data, used to implement the highly efficient interactive method for connecting large models and multi-source data as described in any one of claims 1 to 9, characterized in that, include: The data acquisition module is used to acquire data from data sources connected to the target large model to obtain multi-source heterogeneous data; The template building module is used to build dynamic parsing templates based on the format and semantic features of various types of data in multi-source heterogeneous data. The data parsing module performs hierarchical parsing of the corresponding multi-source heterogeneous data based on the constructed dynamic parsing template, and obtains the standard data units corresponding to each type of data. The topology analysis module is used to construct a semantic association graph of multi-source data based on the obtained standard data units and generate multi-source data interaction sequences. The data interaction module is used to input the multi-source data interaction sequence into the target large model, and obtain the interaction response results between the corresponding large model and the multi-source data interaction sequence by updating the input attention weight of the target large model; The interactive feedback module is used to optimize the generation strategy of the corresponding multi-source data interaction sequence based on the semantic consistency between the interactive response results and the semantic association graph of multi-source data.

Images