Document link construction and evolution relation tracking method based on fragment-level semantic alignment

Abstract

The invention discloses a document link construction and evolution relation tracking method based on fragment-level semantic alignment, which relates to the technical field of document management, and comprises the following steps: carrying out structured analysis on a plurality of input documents, and dividing the document content into a plurality of semantic fragments; performing vectorization processing on each semantic fragment, and constructing a document-fragment-vector semantic mapping relationship; different document fragment-level semantic alignment is realized through cross-document semantic vector similarity calculation, a position offset tolerance algorithm and context association analysis, and a semantic link network is established; fusing the link relationship, the timestamp, the version information and the author information, and constructing a document evolution diagram taking the documents or fragments as nodes and evolution links as edges to record semantic flow paths between the documents; and outputting the semantic link network and the document evolution diagram in a structured form. The method can solve the problems that semantic relations between documents are missing, evolution paths are invisible, content reuse cannot be traced, and the structure induction capability is insufficient.

Description

Technical Field

[0001] The present invention relates to the technical field of document management, and in particular to a method for document link construction and evolution relationship tracking based on segment-level semantic alignment. Background Art

[0002] With the acceleration of digital office work and information accumulation, enterprises, institutions, and organizations have accumulated vast amounts of document data with varying versions, diverse structures, and complex sources. These documents often have complex relationships involving content reuse, version modifications, clause changes, and logical evolution. However, in traditional file management systems, documents are typically organized as independent static files, relying on simple categorization based on directory structures, file names, or timestamps, without a deep understanding of the inherent semantic connections between documents.

[0003] While some existing systems support document versioning, these systems often rely on manually specified version chains or file naming conventions, making it difficult to automatically identify content similarities, inheritance relationships, or expression variations. Furthermore, there is a lack of effective analysis tools for situations where the same business content appears repeatedly in multiple documents, has different expressions, or is migrated across projects. This leads to knowledge redundancy, chaotic management, and opaque content reuse, hindering the structured utilization and intelligent management of organizational content assets.

[0004] In recent years, the development of natural language processing technology, particularly large language models, has enabled the semantic representation and similarity calculation of unstructured documents, making it possible to analyze semantic associations and track evolutionary relationships between documents. However, a systematic approach is still lacking to segment and align document content at a fine-grained level, thereby constructing a stable and scalable semantic link graph to track the origin, evolution, and diffusion of document content. Summary of the Invention

[0005] The present invention aims to solve the problems in the prior art of missing semantic relationships between documents, invisible evolution paths, untraceable content reuse, and insufficient structural induction capabilities, and provides a document link construction and evolution relationship tracking method based on fragment-level semantic alignment.

[0006] The present invention provides a document link construction and evolution relationship tracking method based on segment-level semantic alignment, which solves the above technical problems by adopting the following technical solutions:

[0007] A document link construction and evolution relationship tracking method based on segment-level semantic alignment includes the following steps:

[0008] S1. Perform structural analysis on multiple input documents and divide the document content into multiple semantic segments;

[0009] S2. Vectorize each semantic segment and build a semantic mapping relationship between document, segment, and vector.

[0010] S3. Through cross-document semantic vector similarity calculation, position offset tolerance algorithm and context association analysis, semantic alignment of different document segments is achieved, and a semantic link network with four types of link relationship labels: semantic reuse, semantic variation, structural reorganization and expression transformation is established;

[0011] S4. Integrate link relationships, timestamps, version information, and author information to construct a document evolution graph with documents or fragments as nodes and evolution links as edges to record the semantic flow paths between documents.

[0012] S5. Output the semantic link network and document evolution graph in a structured form.

[0013] Optionally, step S1 specifically includes:

[0014] S1.1. Receive multiple structured or semi-structured documents as input and use the document parsing engine to clean, restore the structure, and restore the logical hierarchy of the documents, including operations such as title extraction, paragraph segmentation, number recognition, and table disassembly.

[0015] S1.2. After the document content is parsed, each document is divided into multiple semantic segments with complete meanings according to the document structure and language organization rules. Each semantic segment is accompanied by the metadata of the document it belongs to and can be independently processed as a content unit.

[0016] Further optionally, the step S2 specifically includes:

[0017] Using a pre-trained language model or a fine-tuned large language model, each semantic segment is converted into a multi-dimensional dense semantic vector to capture its contextual meaning, semantic relations, and pragmatic features;

[0018] During the vector conversion process, the context window, document type label, and contextual prompts are integrated as additional conditions to construct a semantic mapping relationship with a three-layer structure of "document-fragment-vector".

[0019] Further optionally, step S3 specifically includes:

[0020] S3.1. For vectorized semantic segments, use cosine similarity or vector distance to calculate the similarity between segment vectors across documents and select potentially semantically related segment pairs.

[0021] S3.2, analyzing the structural position differences of segments in the document through a position shift tolerance algorithm, allowing preset range shifts to match semantically related segments across positions;

[0022] S3.3. Introduce context compatibility analysis and historical version information to verify the contextual semantic consistency and evolutionary relationship of the fragment pairs and eliminate cross-context mismatches;

[0023] S3.4. Based on the semantic proximity, positional differences, and contextual features of the segment pairs, determine and label the four types of link relationships: semantic reuse, semantic variation, structural reorganization, and expression transformation;

[0024] S3.5. Integrate the annotated semantic fragment pairs into a graph structure, using fragments as nodes and link relationships as edges to build a cross-document semantic link network.

[0025] Further optionally, step S4 specifically includes:

[0026] S4.1. Collect the annotated semantic link relationship data and extract the timestamp, version information, and author information corresponding to each link;

[0027] S4.2. Treat the document and its segmented semantic fragments as nodes in the graph structure, assign a unique identifier to each node, and associate node attributes with it.

[0028] S4.3. Use semantic links as edges in the graph structure, define the direction of the edge based on the link type, and annotate the edge with timestamps, version information, and author information;

[0029] S4.4. Integrate nodes and edges into a document evolution graph through a graph database or visualization tool, support node scaling and edge attribute filtering operations, and intuitively present the semantic flow path between documents or fragments.

[0030] Preferably, step S4 is performed to develop a multi-dimensional query function, allowing users to filter the document evolution graph by time stage, link relationship type or author, and generate a customized semantic flow visualization view.

[0031] Further optionally, step S5 is executed to output the semantic link network and the document evolution graph in a structured form. The output content includes: a link relationship table between documents, evolution graph structure data, a visual map, a node summary, a similarity score, a link relationship label, and source document reference information for system call or manual review.

[0032] Preferably, step S5 is executed to output the semantic link network and the document evolution graph in the format of JSON, GraphML or CSV to generate an interactive visual graph. The user selects any document, clause or paragraph through the visual interactive interface to dynamically view its semantic evolution path, upstream and downstream links and the number of citations / reuses. The graph zooming, path expansion and traceability jump operations are supported to enhance the intuitive understanding of document relationships.

[0033] The document link construction and evolution relationship tracking method based on segment-level semantic alignment of the present invention has the following beneficial effects compared with the prior art:

[0034] 1. The present invention establishes a three-level structure of "document-fragment-vector" by semantic segmentation and fragment-level modeling of document content, realizes fine-grained content understanding and unified representation of complex documents, and improves the resolution and alignment accuracy of semantic analysis; through semantic alignment, it supports the construction of multiple types of document link relationships such as "semantic reuse chain", "evolution variant chain" and "structural migration chain", reflecting stronger semantic perception and link structure expression capabilities; by introducing metadata such as time, version, and author, a semantic evolution graph is constructed based on the identification of link relationships, realizing the full-cycle tracking capability of content from initial expression to multi-version derivation, which is suitable for content traceability, version auditing and risk propagation path analysis. It supports the output of link networks and document evolution graphs as structured graph data, and performs multi-dimensional screening, path traceability, and reference relationship navigation through interactive views, thereby improving users' understanding of the relationship between documents and operational efficiency;

[0035] 2. This invention can enhance the ability to perceive the semantic structure between documents, improve the traceability and explainability of document evolution links, support the automatic generation of knowledge structures and networked content management, improve document retrieval efficiency and context aggregation capabilities, and is independent of specific formats, structures, or languages. It can be widely applied to multilingual, multimodal, and multi-business document processing platforms, and supports integration with knowledge graph systems, enterprise content management systems (ECM), and regulatory comparison platforms.

[0036] 3. The present invention automatically constructs a semantic link network in multi-source, multi-version documents and tracks their content evolution path. It is suitable for scenarios such as intelligent document management systems, knowledge graph construction platforms, policy and regulation comparison systems, and enterprise content archiving platforms. It has the beneficial effects of improving the depth of semantic understanding between documents, enhancing version traceability, and optimizing content organizational structure. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Attachment Figure 1 This is a flow chart of the method of embodiment 1 of the present invention. DETAILED DESCRIPTION

[0038] In order to make the technical solution, the technical problems solved and the technical effects of the present invention more clear, the technical solution of the present invention is clearly and completely described below in conjunction with specific embodiments.

[0039] Example 1:

[0040] Combined with attachment Figure 1 This embodiment proposes a document link construction and evolution relationship tracking method based on segment-level semantic alignment, which includes the following steps:

[0041] S1. Perform structural analysis on multiple input documents and divide the document content into multiple semantic segments, including:

[0042] S1.1. Receive multiple structured or semi-structured documents as input and use the document parsing engine to clean the document content, restore its structure, and restore its logical hierarchy. This includes operations such as title extraction, paragraph segmentation, number recognition, and table disassembly.

[0043] S1.2. After the document content is parsed, each document is divided into multiple semantic segments with complete meanings according to the document structure and language organization rules. Each semantic segment is accompanied by the metadata of the document it belongs to and can be independently processed as a content unit.

[0044] For example, in a contract document, a clause is a fragment; in a report, a discussion paragraph or definition statement constitutes an independent fragment. Each fragment is accompanied by metadata about the document in which it is located, including file name, creation time, and chapter location.

[0045] S2. Vectorize each semantic segment and construct a semantic mapping relationship between document, segment, and vector. Specifically, it includes:

[0046] Use pre-trained language models (such as BERT, SBERT, BGE, etc.) or fine-tuned large language models to convert each semantic segment into a multi-dimensional dense semantic vector to capture its contextual meaning, semantic relations, and pragmatic features;

[0047] During the vector conversion process, the context window, document type label, and contextual prompts are integrated as additional conditions to construct a semantic mapping relationship with a three-layer structure of "document-fragment-vector", providing basic semantic representation support for tasks such as cross-document content comparison, clustering, and link construction.

[0048] S3. Through cross-document semantic vector similarity calculation, position offset tolerance algorithm and contextual association analysis, semantic alignment of different document segments is achieved, and a semantic link network with four types of link relationship labels: semantic reuse, semantic variation, structural reorganization and expression transformation is established. Specifically, the following are included:

[0049] S3.1. For vectorized semantic segments, use cosine similarity or vector distance to calculate the similarity between segment vectors across documents and select potentially semantically related segment pairs.

[0050] S3.2, analyzing the structural position differences of segments in the document through a position shift tolerance algorithm, allowing preset range shifts to match semantically related segments across positions;

[0051] S3.3. Introduce context compatibility analysis and historical version information to verify the contextual semantic consistency and evolutionary relationship of the fragment pairs and eliminate cross-context mismatches;

[0052] S3.4. Based on the semantic proximity, positional differences, and contextual features of the segment pairs, determine and label the four types of link relationships: semantic reuse, semantic variation, structural reorganization, and expression transformation;

[0053] S3.5. Integrate the annotated semantic fragment pairs into a graph structure, using fragments as nodes and link relationships as edges to build a cross-document semantic link network.

[0054] S4. Integrate link relationships, timestamps, version information, and author information to construct a document evolution graph with documents or fragments as nodes and evolution links as edges to record the semantic flow paths between documents. Specifically, it includes:

[0055] S4.1. Collect the annotated semantic link relationship data and extract the timestamp, version information, and author information corresponding to each link;

[0056] S4.2. Treat the document and its segmented semantic fragments as nodes in the graph structure, assign a unique identifier to each node, and associate node attributes with it.

[0057] S4.3. Use semantic links as edges in the graph structure, define the direction of the edge based on the link type, and annotate the edge with timestamps, version information, and author information;

[0058] S4.4. Integrate nodes and edges into a document evolution graph (Document Evolution Graph) through a graph database or visualization tool. This supports node scaling and edge attribute filtering, and intuitively presents the semantic flow path between documents or fragments.

[0059] This step also develops a multi-dimensional query function, allowing users to filter the document evolution graph by time stage, link relationship type or author, and generate a customized semantic flow visualization view.

[0060] S5. Output the semantic link network and document evolution graph in a structured form. The output content includes: inter-document link relationship table, evolution graph structure data, visualization graph, node summary, similarity score, link relationship label, and source document reference information for system call or manual review.

[0061] This step specifically outputs the semantic link network and document evolution graph in JSON, GraphML, or CSV format, generating an interactive visualization. Users can select any document, clause, or paragraph through the interactive visualization interface to dynamically view its semantic evolution path, upstream and downstream links, and citation / reuse counts. Graph zooming, path expansion, and source tracing operations are supported, enhancing intuitive understanding of document relationships. This visualization mechanism can significantly enhance the understandability and operability of document management systems, supporting real-world applications such as intelligent retrieval, evolutionary review, and knowledge reorganization.

[0062] In summary, the document link construction and evolution relationship tracking method based on fragment-level semantic alignment of the present invention is used to automatically construct a semantic link network in multi-source and multi-version documents and track their content evolution paths. This can solve the problems of missing semantic relationships between documents, invisible evolution paths, untraceable content reuse, and insufficient structural induction capabilities. It is suitable for scenarios such as intelligent document management systems, knowledge graph construction platforms, policy and regulation comparison systems, and enterprise content archiving platforms, and has the beneficial effects of improving the depth of semantic understanding between documents, enhancing version traceability capabilities, and optimizing content organizational structures.

[0063] The above specific examples are used to illustrate the principles and implementation methods of the present invention in detail. These examples are only used to help understand the core technical content of the present invention. Based on the above specific embodiments of the present invention, any improvements and modifications made by those skilled in the art without departing from the principles of the present invention should fall within the scope of patent protection of the present invention.

Claims

1. A document link construction and evolution relationship tracking method based on segment-level semantic alignment, characterized by: The steps include: S1. Perform structural analysis on multiple input documents and divide the document content into multiple semantic segments; S2. Vectorize each semantic segment and build a semantic mapping relationship between document, segment, and vector. S3. Through cross-document semantic vector similarity calculation, position offset tolerance algorithm and context association analysis, semantic alignment of different document segments is achieved, and a semantic link network with four types of link relationship labels: semantic reuse, semantic variation, structural reorganization and expression transformation is established; S4. Integrate link relationships, timestamps, version information, and author information to construct a document evolution graph with documents or fragments as nodes and evolution links as edges to record the semantic flow paths between documents. S5. Output the semantic link network and document evolution graph in a structured form.

2. The document link construction and evolution relationship tracking method based on segment-level semantic alignment according to claim 1 is characterized in that: The step S1 specifically includes: S1.

1. Receive multiple structured or semi-structured documents as input and use the document parsing engine to clean, restore the structure, and restore the logical hierarchy of the documents, including operations such as title extraction, paragraph segmentation, number recognition, and table disassembly. S1.

2. After the document content is parsed, each document is divided into multiple semantic segments with complete meanings according to the document structure and language organization rules. Each semantic segment is accompanied by the metadata of the document it belongs to and can be independently processed as a content unit.

3. The document link construction and evolution relationship tracking method based on segment-level semantic alignment according to claim 2 is characterized in that: The step S2 specifically includes: Using a pre-trained language model or a fine-tuned large language model, each semantic segment is converted into a multi-dimensional dense semantic vector to capture its contextual meaning, semantic relations, and pragmatic features; During the vector conversion process, the context window, document type label, and contextual prompts are integrated as additional conditions to construct a semantic mapping relationship with a three-layer structure of "document-fragment-vector".

4. The document link construction and evolution relationship tracking method based on segment-level semantic alignment according to claim 3 is characterized in that: The step S3 specifically includes: S3.

1. For vectorized semantic segments, use cosine similarity or vector distance to calculate the similarity between segment vectors across documents and select potentially semantically related segment pairs. S3.2, analyzing the structural position differences of segments in the document through a position shift tolerance algorithm, allowing preset range shifts to match semantically related segments across positions; S3.

3. Introduce context compatibility analysis and historical version information to verify the contextual semantic consistency and evolutionary relationship of the fragment pairs and eliminate cross-context mismatches; S3.

4. Based on the semantic proximity, positional differences, and contextual features of the segment pairs, determine and label the four types of link relationships: semantic reuse, semantic variation, structural reorganization, and expression transformation; S3.

5. Integrate the annotated semantic fragment pairs into a graph structure, using fragments as nodes and link relationships as edges to build a cross-document semantic link network.

5. The document link construction and evolution relationship tracking method based on segment-level semantic alignment according to claim 4 is characterized in that: The step S4 specifically includes: S4.

1. Collect the annotated semantic link relationship data and extract the timestamp, version information, and author information corresponding to each link; S4.

2. Treat the document and its segmented semantic fragments as nodes in the graph structure, assign a unique identifier to each node, and associate node attributes with it. S4.

3. Use semantic links as edges in the graph structure, define the direction of the edge based on the link type, and annotate the edge with timestamps, version information, and author information; S4.

4. Integrate nodes and edges into a document evolution graph through a graph database or visualization tool, support node scaling and edge attribute filtering operations, and intuitively present the semantic flow path between documents or fragments.

6. The document link construction and evolution relationship tracking method based on segment-level semantic alignment according to claim 5 is characterized in that: Execute step S4 to develop a multi-dimensional query function, allowing users to filter the document evolution graph by time stage, link relationship type, or author, and generate a customized semantic flow visualization view.

7. The document link construction and evolution relationship tracking method based on segment-level semantic alignment according to claim 5 is characterized in that: Execute step S5 to output the semantic link network and document evolution graph in a structured form. The output content includes: inter-document link relationship table, evolution graph structure data, visualization map, node summary, similarity score, link relationship label, and source document reference information for system call or manual review.

8. The document link construction and evolution relationship tracking method based on segment-level semantic alignment according to claim 7 is characterized in that: Execute step S5 to output the semantic link network and document evolution graph in JSON, GraphML or CSV format to generate an interactive visual graph. Users can select any document, clause or paragraph through the visual interactive interface to dynamically view its semantic evolution path, upstream and downstream links and citation / reuse times. It supports graph zooming, path expansion and traceability jump operations, improving the intuitive understanding of document relationships.

Images