Intelligent archive abstract generation system and method based on natural language processing technology

Abstract

The invention relates to the technical field of intelligent abstract generation, in particular to an intelligent archive abstract generation system and method based on a natural language processing technology. The system comprises a multi-layer semantic generation core unit which performs semantic hierarchical segmentation analysis on an original file text to construct a multi-layer semantic graph, and constructs an abstract generation model to generate abstract content; the knowledge graph fusion engine unit associates and extracts terms in an original file text, constructs an entity mapping relation, and sets semantic graph node expression weights in an abstract generation model according to semantic association confidence scores; a paging index cache retrieval unit performs fragment division processing on each node in the multi-layer semantic graph, and constructs an abstract content fragment index structure containing a node set; and the version tracing transaction management unit carries out version recording on generation and modification operations of the abstract contents. The invention discloses an intelligent archive abstract generation system which is constructed by fusing a multi-layer semantic graph and is associated with a knowledge graph.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent summary generation, and in particular to a system and method for generating intelligent archive summaries based on natural language processing technology. Background Art

[0002] With the increasing amount of unstructured text materials such as government archives, enterprise and institution documents, and academic data, intelligent archival summarization technology, as an important means to improve information reading efficiency and support document management automation, is gradually becoming an important research direction in the field of natural language processing. Currently, automatic summarization methods based on deep learning have been widely used in news, medical, legal and other scenarios. They mainly rely on text-level sequence encoders to generate fixed-length summaries. However, this traditional method often has difficulty in fine-tuning the complex semantic structure of the original archives and lacks the ability to identify and express key knowledge units such as terms, entities, and indicators. As a result, the generated summaries have certain deficiencies in accuracy, structural clarity, and traceability.

[0003] The existing technology still has the following two key problems in practical applications: First, traditional summary generation methods are mostly based on single-plane semantic modeling or shallow representation, lack the ability to model the hierarchical semantic structure in archival texts, and cannot effectively identify and retain the connections between multi-level and multi-path semantic units, resulting in a lack of semantic integrity and structural logic in the summary content; Second, current summary systems generally ignore deep integration with external knowledge graphs, lack quantitative expression and dynamic control mechanisms for the strength of semantic connections between archival terms and external entities, resulting in insufficient attention to key terms in the generated summary, affecting the professionalism and interpretability of the content, and unable to support operations such as version management and on-demand query of summary content. Summary of the Invention

[0004] The purpose of the present invention is to provide a system and method for generating intelligent archive summaries based on natural language processing technology to solve the two key problems raised in the above background technology.

[0005] To achieve the above objectives, the present invention provides a system for generating intelligent archive summaries based on natural language processing technology, comprising:

[0006] The multi-layer semantic generation core unit is used to perform semantic hierarchical segmentation analysis on the original text of the archive to construct a multi-layer semantic graph. The structural relationship node scores are used to evaluate the nodes in the multi-layer semantic graph. Based on the multi-layer semantic graph and the node scores, a summary generation model is constructed, and the summary content is generated by combining the structural path relationships of each node.

[0007] The knowledge graph fusion engine unit is used to associate and extract terms from the original text of the archive, build entity mapping relationships between terms and external knowledge graphs, calculate semantic association confidence scores based on the semantic relevance between terms and entity nodes in the external knowledge graph, set the expression weights of semantic graph nodes in the summary generation model based on the semantic association confidence scores, and simultaneously build an extended citation matrix and integrate it into the summary generation model;

[0008] A paged index cache retrieval unit is used to segment each node in the multi-layer semantic graph, construct a summary content fragment index structure containing a node set, and filter out matching summary content using a matching page block filter operation based on the user-input summary query condition;

[0009] The version tracing transaction management unit is used to record the generation and modification operations of the summary content, and to construct a multi-summary version index linked list structure to trace and manage the summary content.

[0010] The multi-layer semantic generation core unit includes a semantic graph construction module;

[0011] The semantic graph construction module is used to perform semantic hierarchical segmentation analysis on the original text of the archive, and construct a multi-layer semantic graph based on the logical connection structure between sentences, the cluster boundaries of subject words, and the pragmatic relationship between paragraphs;

[0012] Semantic level segmentation analysis includes syntactic dependency analysis, semantic role labeling, and context level segmentation. It divides the original text of the archive into several semantic units according to the grammatical structure relationship, and identifies the subject-verb-object structure and context dependency in each semantic unit to construct a text structure map with semantic nesting level.

[0013] The text structure graph of the semantic nesting level is used to construct a multi-layer semantic graph. The specific method is as follows:

[0014] The semantic layering is constructed based on the path depth of syntactic dependency analysis. Semantic units within the same path depth are classified into the same semantic layer. Intra-layer edges are established based on the semantic dependency relationship between predicate argument structures, and cross-layer edges are established based on the contextual relationship between events.

[0015] Among them, the nodes of the multi-layer semantic graph are semantic units, the intra-layer edges of the multi-layer semantic graph are syntactic dependencies between nodes in the same semantic layer, and the cross-layer edges of the multi-layer semantic graph represent the progressive relationship of event contexts in different semantic layers.

[0016] Preferably, the multi-layer semantic generation core unit includes a summary modeling control module;

[0017] The summary modeling control module constructs a summary generation model based on the structural relationship of each node in the multi-layer semantic graph and the structural relationship node score;

[0018] The structural relationships of nodes in the multi-layer semantic graph include the syntactic dependency relationships between semantic units in the same layer and the event context progression relationships between semantic units across layers. The structural relationship node score is weighted based on the structural depth hierarchical factor, syntactic connection centrality factor, and context aggregation density factor of each node in the semantic graph.

[0019] Among them, the structural depth hierarchical factor represents the semantic level of the node in the multi-layer semantic graph; the syntactic connection centrality factor represents the number of inbound and outbound edges of the node in the multi-layer semantic graph; and the context aggregation density factor represents the density of co-occurring nodes in the context sliding window.

[0020] Preferably, the summary modeling control module includes:

[0021] Among them, the window-limited attention mechanism is used to limit the local window range calculated by the attention mechanism based on the path distance between nodes in the multi-layer semantic graph when constructing a summary generation model; the guided semantic output control structure is used to prioritize the generation of semantic unit content with high semantic weight based on the path centrality index and structural relationship nodes of each node in the multi-layer semantic graph during the summary generation process.

[0022] Preferably, the knowledge graph fusion engine unit includes an entity mapping construction module;

[0023] The entity mapping construction module is used to identify terms from the original text of the archive and construct a mapping path between the nodes in the multi-layer semantic graph and the entity nodes in the external knowledge graph based on the structural features between the terms and the entity nodes in the external knowledge graph. The specific steps include:

[0024] S21.1. Extract term nodes from the original text of the archive and calculate the contextual semantic vector of each term node; extract attribute description information of entity nodes in the external knowledge graph and construct the corresponding entity node semantic vector;

[0025] S21.2. Based on the similarity between the contextual semantic vector of the term node and the semantic vector of the entity node, select candidate node pairs that meet a preset semantic similarity threshold to construct a candidate mapping set;

[0026] S21.3. Construct a context path subgraph of the candidate term node in the multi-layer semantic graph and a structural neighborhood subgraph of the candidate entity node in the external knowledge graph;

[0027] S21.4. Perform structural embedding encoding on the context path subgraph and the structural neighborhood subgraph to obtain path structural feature representations, and calculate a joint confidence score of the mapping path based on the structural similarity between the path structural feature representations and the semantic similarity;

[0028] S21.5. Node pairs whose joint confidence scores exceed the preset score threshold are determined to be matching nodes, and their path mapping structure is constructed to form a node mapping path set, and finally the mapping path between the nodes in the multi-layer semantic graph and the entity nodes in the external knowledge graph is obtained.

[0029] Preferably, the knowledge graph fusion engine unit further includes an association weight control module;

[0030] The association weight control module calculates the semantic association confidence score based on the semantic relevance between the term and the entity node in the external knowledge graph, and sets the expression weight of the semantic graph node in the summary generation model according to the semantic association confidence score. Specifically, it includes the following steps:

[0031] S22.1. Extract entity definitions, attribute labels, and adjacency relationships of entity nodes in the external knowledge graph and construct a multi-dimensional vector embedding representation of the entity nodes;

[0032] S22.2. Calculate the semantic relevance score between the contextual semantic vector of the term node and the semantic vector of the entity node using a bidirectional attention mechanism-based approach. Combined with the structural position indicators of the term node in the multi-layer semantic graph, calculate the semantic association confidence score.

[0033] S22.3. Based on the semantic association confidence score, assign a semantic graph node expression weight in the summary generation model to the multi-layer semantic graph node corresponding to the term node.

[0034] Preferably, the extended reference matrix is ​​a structured semantic reference matrix between term nodes and external knowledge graph entity nodes, which is used to describe the external knowledge graph entity path relationship referenced by nodes in the semantic hierarchy structure in the multi-layer semantic graph;

[0035] The specific method for constructing the extended citation matrix is ​​as follows: extract the term node set in the multi-layer semantic graph and the candidate entity node set in the external knowledge graph; based on the entity mapping path and the semantic association confidence score, calculate the joint weight value of each pair of nodes in the term node set and the candidate entity node set, and take the joint weight value greater than the preset citation confidence threshold to construct the extended citation matrix;

[0036] The joint weight value greater than the preset reference confidence threshold indicates whether the term node references the candidate entity node, and the size of the joint weight value represents the reference strength. A larger joint weight value indicates a greater reference strength.

[0037] Preferably, the paging index cache retrieval unit includes a semantic segment division module, a segment index construction module and a condition matching retrieval module;

[0038] The semantic fragment partitioning module performs fragment clustering on each node in the multi-layer semantic graph based on the node structure path relationship and semantic clustering features, and generates multiple summary fragment subgraphs.

[0039] The fragment index construction module is used to encode each summary fragment subgraph into fragment metadata containing node index numbers and path start and end marks, and to build a paged index table structure for users to locate and call on demand;

[0040] The condition matching retrieval module constructs a query semantic vector based on the user-input summary query condition, and performs similarity scoring matching with each summary fragment subgraph in the paginated index table structure to select the target summary fragment that meets the matching threshold.

[0041] Preferably, in the version tracing transaction management unit, the multi-summary version index linked list structure is a unidirectional or bidirectional chain structure composed of multiple summary version nodes, which is used to record historical summary versions in an orderly manner, enable traceable management, and track version content differences after the user operates on the summary content;

[0042] The specific method for constructing a multi-summary version index linked list structure is as follows: after each operation and editing of the summary content, a summary version node is constructed, which includes the summary compressed content, timestamp, unique identifier and original semantic fragment mapping record; if this is the first time the system generates summary content, the summary version node constructed for the first time is used as the starting node of the linked list; otherwise, the current summary version node is linked to the previous summary version node, and a bidirectional index pointer is set; between each pair of consecutive summary version nodes, a summary difference description vector based on word-level difference recognition technology is used to construct the vector, and it is attached to the index structure between the pair of linked list nodes, and finally a multi-summary version index linked list structure is obtained.

[0043] In another aspect, the present invention provides a method for generating intelligent summaries of archives based on natural language processing technology, which is used in the above-mentioned intelligent summaries of archives based on natural language processing technology, comprising the following steps:

[0044] S10.1. Perform semantic hierarchical segmentation analysis on the original text of the archive to construct a multi-layer semantic graph. Score the structural relationships of each node in the multi-layer semantic graph. Build a summary generation model based on the multi-layer semantic graph and the node scores. Generate summary content based on the structural path relationships of each node.

[0045] S10.2. Associate and extract terms from the original text of the archive, construct entity mapping relationships between the terms and the external knowledge graph, calculate semantic association confidence scores based on the semantic relevance between the terms and the entity nodes in the external knowledge graph, set the expression weights of the semantic graph nodes in the summary generation model based on the semantic association confidence scores, and simultaneously construct an extended citation matrix and integrate it into the summary generation model;

[0046] S10.3. Segment each node in the multi-layer semantic graph, construct a summary content fragment index structure containing a node set, and use a matching page block filtering operation to filter out matching summary content based on the summary query condition input by the user;

[0047] S10.4. Record the generation and modification operations of the summary content, and construct a multi-summary version index linked list structure to trace and manage the summary content.

[0048] Compared with the prior art, the above technical solution of the present invention has the following beneficial technical effects:

[0049] 1. In this invention, a summary generation model is constructed based on a multi-layer semantic graph and structural relationship node scoring, which can achieve hierarchical parsing and structural path control of semantic units in the original text of the archive, and improve the generation quality of the summary content in terms of semantic structure integrity, information focus, and expression controllability;

[0050] 2. In the present invention, by constructing an entity mapping relationship between terms and external knowledge graphs and setting the expression weights of semantic graph nodes based on the semantic association confidence score, the importance of key node expressions in the summary generation model is regulated, and the construction of an extended citation matrix is ​​combined to further enhance the contextual semantic coverage capability between archival terms and graph entities. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 A functional block diagram of an embodiment of the present invention;

[0052] Figure 1: 1. Multi-layer semantic generation core unit; 11. Semantic graph construction module; 12. Summary modeling control module; 2. Knowledge graph fusion engine unit; 21. Entity mapping construction module; 22. Association weight control module; 3. Paging index cache retrieval unit; 31. Semantic fragment division module; 32. Fragment index construction module; 33. Condition matching retrieval module; 4. Version tracing transaction management unit. DETAILED DESCRIPTION

[0053] Example 1, as Figure 1 As shown, a system for generating intelligent archive summaries based on natural language processing technology is provided, including:

[0054] Multi-layer semantic generation core unit 1 is used to perform semantic hierarchical segmentation analysis on the original text of the archive to construct a multi-layer semantic graph, use the structural relationship node score of each node in the multi-layer semantic graph to build a summary generation model based on the multi-layer semantic graph and node score, and generate summary content based on the structural path relationship of each node;

[0055] The multi-layer semantic generation core unit 1 includes a semantic graph construction module 11;

[0056] The semantic graph construction module 11 is used to perform semantic hierarchical segmentation analysis on the original text of the archive, and to construct a multi-layer semantic graph based on the logical connection structure between sentences, the clustering boundaries of subject words, and the pragmatic relationship between paragraphs;

[0057] Semantic level segmentation analysis includes syntactic dependency analysis, semantic role labeling, and context level segmentation. It divides the original text of the archive into several semantic units according to the grammatical structure relationship, and identifies the subject-verb-object structure and context dependency in each semantic unit to construct a text structure map with semantic nesting level.

[0058] The text structure graph of the semantic nesting level is used to construct a multi-layer semantic graph. The specific method is as follows:

[0059] The semantic layering is constructed based on the path depth of syntactic dependency analysis. Semantic units within the same path depth are classified into the same semantic layer. Intra-layer edges are established based on the semantic dependency relationship between predicate argument structures, and cross-layer edges are established based on the contextual relationship between events.

[0060] Among them, the nodes of the multi-layer semantic graph are semantic units, the intra-layer edges of the multi-layer semantic graph are syntactic dependencies between nodes in the same semantic layer, and the cross-layer edges of the multi-layer semantic graph represent the progressive relationship of event contexts in different semantic layers.

[0061] In this embodiment, the path depth of syntactic dependency analysis refers to the shortest path length required for a semantic unit to start from the syntactic root node and reach the node through the dependency edge in the syntactic dependency tree. The path depth is expressed as the minimum number of edges in the graph structure, which is used to reflect the subordinate level of the node in the overall syntactic structure.

[0062] In this embodiment, the multi-layer semantic generation core unit 1 includes a summary modeling control module 12;

[0063] The summary modeling control module 12 constructs a summary generation model based on the structural relationship of each node in the multi-layer semantic graph and the structural relationship node score;

[0064] The structural relationships of nodes in the multi-layer semantic graph include the syntactic dependency relationships between semantic units in the same layer and the event context progression relationships between semantic units across layers. The structural relationship node score is weighted based on the structural depth hierarchical factor, syntactic connection centrality factor, and context aggregation density factor of each node in the semantic graph.

[0065] Among them, the structural depth hierarchical factor represents the semantic level of the node in the multi-layer semantic graph; the syntactic connection centrality factor represents the number of inbound and outbound edges of the node in the multi-layer semantic graph; and the context aggregation density factor represents the density of co-occurring nodes in the context sliding window.

[0066] In this embodiment, the syntactic dependency relationship between semantic units at the same level is based on the main component connection structure of dependency grammar, which is used to express the grammatical control and modification direction between the subject, predicate, object and modifiers in the same semantic level, thereby reflecting the primary and secondary structure and logical cohesion relationship between semantic units;

[0067] The event contextual progression between cross-layer semantic units is based on the sequential pointing structure of event trigger words and contextual causal words. It is used to represent the contextual presupposition and causal preparation of the upper-layer event on the lower-layer semantic units, reflecting the pragmatic evolution path of the semantic connection chain in summary generation.

[0068] In this embodiment, the basic concept of the structural depth hierarchical factor is that the closer the node is to the top layer, the stronger its control and generative representativeness over the entire semantic organizational structure. The hierarchical weight is represented by the reverse normalized value of the layer number, which is defined as follows:

[0069] Assume that the multi-layer semantic graph has L layers and node v q The layer is l q , then its structural depth stratification factor is:

[0070]

[0071] Among them, f str (v q ) is the node v q The structural depth hierarchical factor, and f str (v q )∈[0,1], the closer to the first layer, the larger the value;

[0072] The syntactic connection centrality factor is used to measure the "connection activity" of a node in the structural connection graph, representing the strength of its syntactic association with other semantic units. It is defined as follows:

[0073] Let node v q The outgoing edge set is O(v q ), the set of incoming edges is I(v q ), then its syntactic connection centrality factor is:

[0074]

[0075] Among them, f syn (v q ) is the node v q The syntactic connection centrality factor of v j is the jth node; V is the set of all nodes; the larger the syntactic connection centrality factor, the more likely the node is a syntactic connection hub node and has a higher summary representativeness;

[0076] The context aggregation density factor measures the semantic density of a node in its context neighborhood and is defined as follows:

[0077] For node v q Set the context window W q , which contains all the nodes that are semantically related to it in the sliding time window / structure window, denoted as N c (v q ), then its context aggregation density factor is:

[0078]

[0079] Among them, f ctx (v q ) is the node v q Context aggregation density factor;

[0080] Finally, node v q The structural relationship node scores are:

[0081] S str (v q )=α1f str (v q )+α2f syn (v q )+α3f ctx (v q );

[0082] Among them, S str (v q ) is the node v q The structural relationship node score; α1 is the structural depth stratification factor weight; α2 is the syntactic connection centrality factor weight; α3 is the context aggregation density factor weight.

[0083] In this embodiment, the summary modeling control module 12 includes a window-limited attention mechanism and a guided semantic output control structure;

[0084] Among them, the window-limited attention mechanism is used to limit the local window range of the attention mechanism calculation based on the path distance between nodes in the multi-layer semantic graph when constructing the summary generation model, and only perform attention allocation calculation within the set of nodes that meets the preset path length threshold, so as to improve the semantic focusing ability between nodes and suppress cross-segment semantic interference; the guided semantic output control structure is used to prioritize the generation of semantic unit content with high semantic weight based on the path centrality index and structural relationship nodes of each node in the multi-layer semantic graph during the summary generation process.

[0085] In this embodiment, the window refers to the path perception domain size set to limit the scope of the attention mechanism during the construction of the summary generation model. It is used to limit the range of other node sets that any node in the multi-layer semantic graph can focus on when calculating attention. This window is not a fixed-position sliding window in traditional text sequences, but a graph path window based on the multi-layer semantic graph structure. The path distance between nodes in the multi-layer semantic graph refers to the shortest structural path length between two nodes in the semantic graph, usually measured by the shortest path hop count in the graph, that is, the number of edges that a node needs to traverse from one node to another. The local window range refers to the set of nodes in the multi-layer semantic graph whose path distance does not exceed a preset threshold with a central node as the reference point.

[0086] The path centrality index of each node in the multi-layer semantic graph is used to measure the importance of the node as a node in the semantic information transmission path in the graph. The path centrality index is calculated as the frequency of the node appearing in all shortest paths, specifically:

[0087]

[0088] Among them, C pc (v q ) is the node v q Path centrality index of Represents node v q1 To node v q2 and passes through node v q The number of all shortest paths; Represents node v q1 To node v q2 The number of all shortest paths.

[0089] Knowledge graph fusion engine unit 2 is used to associate and extract terms from the original text of the archive, build entity mapping relationships between terms and external knowledge graphs, calculate semantic association confidence scores based on the semantic relevance between terms and entity nodes in the external knowledge graph, set semantic graph node expression weights in the summary generation model based on the semantic association confidence scores, and simultaneously build an extended citation matrix and integrate it into the summary generation model;

[0090] In this embodiment, the knowledge graph fusion engine unit 2 includes an entity mapping construction module 21;

[0091] The entity mapping construction module 21 is used to identify terms from the original text of the archive and construct a mapping path between the nodes in the multi-layer semantic graph and the entity nodes in the external knowledge graph based on the structural features between the terms and the entity nodes in the external knowledge graph. Specifically, the module includes the following steps:

[0092] S21.1. Extract term nodes from the original text of the archive and calculate the contextual semantic vector of each term node; extract attribute description information of entity nodes in the external knowledge graph and construct the corresponding entity node semantic vector;

[0093] S21.2. Based on the similarity between the contextual semantic vector of the term node and the semantic vector of the entity node, select candidate node pairs that meet a preset semantic similarity threshold to construct a candidate mapping set;

[0094] S21.3. Construct a context path subgraph of the candidate term node in the multi-layer semantic graph and a structural neighborhood subgraph of the candidate entity node in the external knowledge graph;

[0095] S21.4. Perform structural embedding encoding on the context path subgraph and the structural neighborhood subgraph to obtain path structural feature representations, and calculate a joint confidence score of the mapping path based on the structural similarity between the path structural feature representations and the semantic similarity;

[0096] S21.5. Node pairs whose joint confidence scores exceed the preset score threshold are determined to be matching nodes, and their path mapping structure is constructed to form a node mapping path set, and finally the mapping path between the nodes in the multi-layer semantic graph and the entity nodes in the external knowledge graph is obtained.

[0097] In this embodiment, the external knowledge graph refers to a data structure set that expresses terms, entities, and the semantic relationships between them in the external domain in the form of a structured graph. It is usually derived from an existing authoritative database or automatically extracted by a graph construction tool. It contains entity nodes such as people, places, and institutions, and attribute edges such as belongs to, contains, and involves.

[0098] The attribute description information of entity nodes in the external knowledge graph refers to the descriptive text data, category labels, upper and lower entity information and associated edge weight features carried by each entity node in the graph. It is used to represent the semantic boundaries and contextual features of the node and serves as the basic content for constructing the entity semantic representation vector;

[0099] The preset semantic similarity threshold is the lower limit used to determine whether the archive term node and the graph entity node meet the initial matching relationship. This threshold is a similarity score boundary set empirically or automatically selected through cross-validation, and represents the minimum acceptance standard for cosine similarity between semantic vectors.

[0100] The context path subgraph is a local path structure graph consisting of a set of adjacent nodes extending outward from the archival term node to a fixed step size or semantic level in the multi-layer semantic graph. It is used to capture the structural nesting, co-occurrence links and semantic propagation characteristics of the term in the original context.

[0101] A structural neighborhood subgraph is a local subgraph structure formed by expanding outward from the target entity node based on the node connection relationship in the knowledge graph. It contains the entity's directly related nodes, semantic attribute edges, and their adjacency relationships. It is usually controlled within two-order expansion and is used to construct the structural context representation of the graph entity.

[0102] In S21.4 and S21.5 of this embodiment, the joint confidence score of the mapping path is calculated and its path mapping structure is constructed. The specific method is as follows:

[0103] Extract the context path subgraph of the term nodes in the multi-layer semantic graph, construct the structural adjacency relationship set of the term nodes in the original archive text, and assign a semantic weight to each adjacent node in the set. The semantic weight is calculated based on its position depth in the path and the path frequency:

[0104]

[0105] in, For node v q The semantic weight of For node v q The path frequency; For node v q The position depth; λ is the path depth penalty factor;

[0106] Extract the structural neighborhood subgraph of the target entity in the knowledge graph, build a structural attribute context set for the entity, calculate the attribute description vector for each adjacent entity in the set, and use the word vector representation method to generate a semantic embedding matrix;

[0107] Calculate the semantic correlation matrix R between the term context subgraph and the entity structure neighborhood subgraph; comprehensively calculate the joint confidence score based on the maximum matching value and the average matching path in the correlation matrix R; compare the joint confidence score with the preset semantic similarity threshold. When the joint confidence score is greater than or equal to the preset semantic similarity threshold, it is determined that there is a valid mapping relationship between the term node and the entity node, and it is recorded as a mapping path.

[0108] In this embodiment, the knowledge graph fusion engine unit 2 further includes an association weight control module 22;

[0109] The association weight control module 22 calculates the semantic association confidence score based on the semantic relevance between the term and the entity node in the external knowledge graph, and sets the expression weight of the semantic graph node in the summary generation model according to the semantic association confidence score. Specifically, the module includes the following steps:

[0110] S22.1. Extract entity definitions, attribute labels, and adjacency relationships of entity nodes in the external knowledge graph and construct a multi-dimensional vector embedding representation of the entity nodes;

[0111] S22.2. Calculate the semantic relevance score between the contextual semantic vector of the term node and the semantic vector of the entity node using a bidirectional attention mechanism-based approach. Combined with the structural position indicators of the term node in the multi-layer semantic graph, calculate the semantic association confidence score.

[0112] S22.3. Based on the semantic association confidence score, assign a semantic graph node expression weight in the summary generation model to the multi-layer semantic graph node corresponding to the term node.

[0113] In this embodiment, in an entity node in an external knowledge graph, the entity definition refers to the textual description information of the standardized archival terminology represented by the graph node, which serves as the primary semantic expression of the entity node; the attribute label refers to the set of attributes associated with the entity node, such as category, source, timestamp, legal classification number, etc.; the adjacency relationship represents the semantic or logical connection between the entity node and other entity nodes;

[0114] The multi-dimensional vector embedding of entity nodes adopts a combined feature learning mechanism based on the Transformer encoder to extract cross-domain entity semantics and perform multi-dimensional vector embedding.

[0115] In this embodiment, the extended reference matrix is ​​a structured semantic reference matrix between term nodes and external knowledge graph entity nodes, which is used to describe the external knowledge graph entity path relationship referenced by nodes in the semantic hierarchy structure in the multi-layer semantic graph;

[0116] The specific method for constructing the extended citation matrix is ​​as follows: extract the term node set in the multi-layer semantic graph and the candidate entity node set in the external knowledge graph; based on the entity mapping path and the semantic association confidence score, calculate the joint weight value of each pair of nodes in the term node set and the candidate entity node set, and take the joint weight value greater than the preset citation confidence threshold to construct the extended citation matrix;

[0117] The joint weight value greater than the preset reference confidence threshold indicates whether the term node references the candidate entity node, and the size of the joint weight value represents the reference strength. A larger joint weight value indicates a greater reference strength.

[0118] The paging index cache retrieval unit 3 is used to segment each node in the multi-layer semantic graph, construct a summary content segment index structure containing a node set, and use a matching page block filtering operation to filter out matching summary content based on the user input summary query condition;

[0119] In this embodiment, the page index cache retrieval unit 3 includes a semantic segment division module 31, a segment index construction module 32 and a condition matching retrieval module 33;

[0120] The semantic segmentation module 31 performs segment clustering on each node in the multi-layer semantic graph based on the node structure path relationship and semantic clustering features to generate multiple summary segment subgraphs.

[0121] The segment index construction module 32 is used to encode each summary segment subgraph into segment metadata including node index numbers and path start and end marks, and to construct a page index table structure for users to locate and call on demand;

[0122] The condition matching retrieval module 33 constructs a query semantic vector based on the summary query condition input by the user, and performs similarity scoring matching with each summary segment subgraph in the paging index table structure to select target summary segments that meet the matching threshold.

[0123] The version tracing transaction management unit 4 is used to record the generation and modification operations of the summary content, and to construct a multi-summary version index linked list structure to trace and manage the summary content;

[0124] In this embodiment, in the version tracing transaction management unit 4, the multi-summary version index linked list structure is a unidirectional or bidirectional chain structure composed of multiple summary version nodes. It is used to record historical summary versions in an orderly manner, manage them retroactively, and track version content differences after the user operates on the summary content.

[0125] The specific method for constructing a multi-summary version index linked list structure is as follows: after each operation and editing of the summary content, a summary version node is constructed, which includes the summary compressed content, timestamp, unique identifier and original semantic fragment mapping record; if this is the first time the system generates summary content, the summary version node constructed for the first time is used as the starting node of the linked list; otherwise, the current summary version node is linked to the previous summary version node, and a bidirectional index pointer is set; between each pair of consecutive summary version nodes, a summary difference description vector based on word-level difference recognition technology is used to construct the vector, and it is attached to the index structure between the pair of linked list nodes, and finally a multi-summary version index linked list structure is obtained.

[0126] In this embodiment, the multi-summary version index linked list structure is further provided with a version rollback interface for supporting users to query the target version in the linked list by version number or timestamp and perform summary content rollback or difference comparison operations.

[0127] In a second embodiment, the present invention proposes a method for generating an intelligent summary of an archive based on natural language processing technology, which is used in the intelligent summary generation system of an archive based on natural language processing technology in the first embodiment, and includes the following steps:

[0128] S10.1. Perform semantic hierarchical segmentation analysis on the original text of the archive to construct a multi-layer semantic graph. Score the structural relationships of each node in the multi-layer semantic graph. Build a summary generation model based on the multi-layer semantic graph and the node scores. Generate summary content based on the structural path relationships of each node.

[0129] S10.2. Associate and extract terms from the original text of the archive, construct entity mapping relationships between the terms and the external knowledge graph, calculate semantic association confidence scores based on the semantic relevance between the terms and the entity nodes in the external knowledge graph, set the expression weights of the semantic graph nodes in the summary generation model based on the semantic association confidence scores, and simultaneously construct an extended citation matrix and integrate it into the summary generation model;

[0130] S10.3. Segment each node in the multi-layer semantic graph, construct a summary content fragment index structure containing a node set, and use a matching page block filtering operation to filter out matching summary content based on the summary query condition input by the user;

[0131] S10.4. Record the generation and modification operations of the summary content, and construct a multi-summary version index linked list structure to trace and manage the summary content.

[0132] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings, but the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A system for generating intelligent summaries of archives based on natural language processing technology, characterized in that: include: The multi-layer semantic generation core unit (1) is used to perform semantic hierarchical segmentation analysis on the original text of the archive to construct a multi-layer semantic graph, use the structural relationship node score of each node in the multi-layer semantic graph, build a summary generation model based on the multi-layer semantic graph and the node score, and generate summary content in combination with the structural path relationship of each node; The knowledge graph fusion engine unit (2) is used to associate and extract terms from the original text of the archive, build an entity mapping relationship between the terms and the external knowledge graph, calculate the semantic association confidence score based on the semantic relevance between the terms and the entity nodes in the external knowledge graph, set the expression weight of the semantic graph node in the summary generation model according to the semantic association confidence score, and build an extended reference matrix to be integrated into the summary generation model; The paging index cache retrieval unit (3) is used to perform segmentation processing on each node in the multi-layer semantic graph, construct a summary content segment index structure including a node set, and use a matching page block filtering operation to filter out matching summary content based on a user input summary query condition; The version tracing transaction management unit (4) is used to record the generation and modification operations of the summary content and to construct a multi-summary version index linked list structure to trace and manage the summary content.

2. The intelligent archive summary generation system based on natural language processing technology according to claim 1 is characterized in that: The multi-layer semantic generation core unit (1) includes a semantic graph construction module (11); The semantic graph construction module (11) is used to perform semantic hierarchical segmentation analysis on the original text of the archive, and to construct a multi-layer semantic graph based on the logical connection structure between sentences, the clustering boundaries of subject words and the pragmatic relationship of paragraphs; Semantic level segmentation analysis includes syntactic dependency analysis, semantic role labeling, and context level segmentation. It divides the original text of the archive into several semantic units according to the grammatical structure relationship, and identifies the subject-verb-object structure and context dependency in each semantic unit to construct a text structure map with semantic nesting level. The text structure graph of the semantic nesting level is used to construct a multi-layer semantic graph. The specific method is as follows: The semantic layering is constructed based on the path depth of syntactic dependency analysis. Semantic units within the same path depth are classified into the same semantic layer. Intra-layer edges are established based on the semantic dependency relationship between predicate argument structures, and cross-layer edges are established based on the contextual relationship between events. Among them, the nodes of the multi-layer semantic graph are semantic units, the intra-layer edges of the multi-layer semantic graph are syntactic dependencies between nodes in the same semantic layer, and the cross-layer edges of the multi-layer semantic graph represent the progressive relationship of event contexts in different semantic layers.

3. The intelligent archive summary generation system based on natural language processing technology according to claim 2 is characterized in that: The multi-layer semantic generation core unit (1) includes a summary modeling control module (12); The summary modeling control module (12) constructs a summary generation model based on the structural relationship of each node in the multi-layer semantic graph and the structural relationship node score; The structural relationships of nodes in the multi-layer semantic graph include the syntactic dependency relationships between semantic units in the same layer and the event context progression relationships between semantic units across layers. The structural relationship node score is weighted based on the structural depth hierarchical factor, syntactic connection centrality factor, and context aggregation density factor of each node in the semantic graph. Among them, the structural depth hierarchical factor represents the semantic level of the node in the multi-layer semantic graph; the syntactic connection centrality factor represents the number of inbound and outbound edges of the node in the multi-layer semantic graph; and the context aggregation density factor represents the density of co-occurring nodes in the context sliding window.

4. The intelligent archive summary generation system based on natural language processing technology according to claim 3 is characterized in that: The summary modeling control module (12) includes: Among them, the window-limited attention mechanism is used to limit the local window range calculated by the attention mechanism based on the path distance between nodes in the multi-layer semantic graph when constructing a summary generation model; the guided semantic output control structure is used to prioritize the generation of semantic unit content with high semantic weight based on the path centrality index and structural relationship nodes of each node in the multi-layer semantic graph during the summary generation process.

5. The intelligent archive summary generation system based on natural language processing technology according to claim 4 is characterized in that: The knowledge graph fusion engine unit (2) includes an entity mapping construction module (21); The entity mapping construction module (21) is used to identify terms from the original text of the archive, and to construct a mapping path between the nodes in the multi-layer semantic graph and the entity nodes in the external knowledge graph based on the structural features between the terms and the entity nodes in the external knowledge graph. Specifically, the module includes the following steps: S21.

1. Extract term nodes from the original text of the archive and calculate the contextual semantic vector of each term node; extract attribute description information of entity nodes in the external knowledge graph and construct the corresponding entity node semantic vector; S21.

2. Based on the similarity between the contextual semantic vector of the term node and the semantic vector of the entity node, select candidate node pairs that meet a preset semantic similarity threshold to construct a candidate mapping set; S21.

3. Construct a context path subgraph of the candidate term node in the multi-layer semantic graph and a structural neighborhood subgraph of the candidate entity node in the external knowledge graph; S21.

4. Perform structural embedding encoding on the context path subgraph and the structural neighborhood subgraph to obtain path structural feature representations, and calculate a joint confidence score of the mapping path based on the structural similarity between the path structural feature representations and the semantic similarity; S21.

5. Node pairs whose joint confidence scores exceed the preset score threshold are determined to be matching nodes, and their path mapping structure is constructed to form a node mapping path set, and finally the mapping path between the nodes in the multi-layer semantic graph and the entity nodes in the external knowledge graph is obtained.

6. The intelligent archive summary generation system based on natural language processing technology according to claim 5 is characterized in that: The knowledge graph fusion engine unit (2) further includes an association weight control module (22); The association weight control module (22) calculates the semantic association confidence score based on the semantic relevance between the term and the entity node in the external knowledge graph, and sets the expression weight of the semantic graph node in the summary generation model according to the semantic association confidence score, which specifically includes the following steps: S22.

1. Extract entity definitions, attribute labels, and adjacency relationships of entity nodes in the external knowledge graph and construct a multi-dimensional vector embedding representation of the entity nodes; S22.

2. Calculate the semantic relevance score between the contextual semantic vector of the term node and the semantic vector of the entity node using a bidirectional attention mechanism-based approach. Combined with the structural position indicators of the term node in the multi-layer semantic graph, calculate the semantic association confidence score. S22.

3. Based on the semantic association confidence score, assign a semantic graph node expression weight in the summary generation model to the multi-layer semantic graph node corresponding to the term node.

7. The intelligent archive summary generation system based on natural language processing technology according to claim 6 is characterized in that: The extended reference matrix is ​​a structured semantic reference matrix between term nodes and external knowledge graph entity nodes, which is used to describe the external knowledge graph entity path relationship referenced by nodes in the semantic hierarchy structure in the multi-layer semantic graph; The specific method for constructing the extended citation matrix is ​​as follows: extract the term node set in the multi-layer semantic graph and the candidate entity node set in the external knowledge graph; based on the entity mapping path and the semantic association confidence score, calculate the joint weight value of each pair of nodes in the term node set and the candidate entity node set, and take the joint weight value greater than the preset citation confidence threshold to construct the extended citation matrix; The joint weight value greater than the preset reference confidence threshold indicates whether the term node references the candidate entity node, and the size of the joint weight value represents the reference strength. A larger joint weight value indicates a greater reference strength.

8. The intelligent archive summary generation system based on natural language processing technology according to claim 7 is characterized in that: The paging index cache retrieval unit (3) includes a semantic segment division module (31), a segment index construction module (32) and a condition matching retrieval module (33); The semantic segmentation module (31) performs segment clustering on each node in the multi-layer semantic graph based on the node structure path relationship and semantic clustering features, and generates multiple summary segment subgraphs; The fragment index construction module (32) is used to encode each summary fragment subgraph into fragment metadata including node index numbers and path start and end marks, and to construct a page index table structure for users to locate and call on demand; The condition matching retrieval module (33) constructs a query semantic vector based on the user input summary query condition, and performs similarity scoring matching with each summary fragment subgraph in the paging index table structure to select the target summary fragment that meets the matching threshold.

9. The intelligent archive summary generation system based on natural language processing technology according to claim 8 is characterized in that: In the version tracing transaction management unit (4), the multi-summary version index linked list structure is a unidirectional or bidirectional chain structure composed of multiple summary version nodes, which is used to record historical summary versions in an orderly manner, manage them retroactively, and track version content differences after the user operates on the summary content; The specific method for constructing a multi-summary version index linked list structure is as follows: after each operation and editing of the summary content, a summary version node is constructed. The summary version node includes the summary compressed content, timestamp, unique identifier, and the original semantic fragment mapping record; if this is the first time the system generates summary content, the first constructed summary version node is used as the starting node of the linked list; Otherwise, the current summary version node is linked to the previous summary version node, and a bidirectional index pointer is set; between each pair of consecutive summary version nodes, a summary difference description vector is constructed using a word-level difference recognition technology method, and attached to the index structure between the pair of linked list nodes, ultimately obtaining a multi-summary version index linked list structure.

10. A method for generating intelligent summaries of archives based on natural language processing technology, used in a system for generating intelligent summaries of archives based on natural language processing technology as claimed in any one of claims 1 to 9, characterized in that: The steps include: S10.

1. Perform semantic hierarchical segmentation analysis on the original text of the archive to construct a multi-layer semantic graph. Score the structural relationships of each node in the multi-layer semantic graph. Build a summary generation model based on the multi-layer semantic graph and the node scores. Generate summary content based on the structural path relationships of each node. S10.

2. Associate and extract terms from the original text of the archive, construct entity mapping relationships between the terms and the external knowledge graph, calculate semantic association confidence scores based on the semantic relevance between the terms and the entity nodes in the external knowledge graph, set the expression weights of the semantic graph nodes in the summary generation model based on the semantic association confidence scores, and simultaneously construct an extended citation matrix and integrate it into the summary generation model; S10.

3. Segment each node in the multi-layer semantic graph, construct a summary content fragment index structure containing a node set, and use a matching page block filtering operation to filter out matching summary content based on the summary query condition input by the user; S10.

4. Record the generation and modification operations of the summary content, and construct a multi-summary version index linked list structure to trace and manage the summary content.

Images