Semantic knowledge block division method and system based on AUTOSAR protocol document
By standardizing AUTOSAR specification documents and dividing them into semantic subgraphs, the problem of context fragmentation caused by traditional block-based strategies is solved. This enables the aggregation of logically related information and the construction of high-fidelity knowledge sources, and is applicable to the semantic knowledge block division of AUTOSAR specification documents.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- WUHAN KOTEI INFORMATICS
- Filing Date
- 2026-03-13
- Publication Date
- 2026-05-08
AI Technical Summary
Traditional chunking strategies lead to the loss of logical association information across pages and modalities in AUTOSAR specification documents, resulting in context fragmentation problems.
By standardizing the AUTOSAR specification documents, multimodal features are extracted to generate high-dimensional embedding vectors and visual embedding vectors, a document knowledge graph is constructed, and knowledge blocks are divided using the semantic subgraph partitioning method to ensure that logically related information is aggregated in the same block.
It solves the context fragmentation problem caused by traditional segmentation strategies, provides high-quality, high-fidelity knowledge sources, reduces the risk of downstream systems generating illusions, and improves the accuracy of information retrieval and reasoning.
Smart Images
Figure CN121997943A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of document processing technology, and in particular to a method and system for semantic knowledge block partitioning of documents based on the AUTOSAR specification. Background Technology
[0002] Automotive Open Systems Architecture (AUTOSAR) is a core standard for software development in the global automotive industry. Its official specification documents are the fundamental basis for engineers to conduct system design, software development, and functional safety analysis. These documents are highly complex, usually published in PDF format, and contain a large number of technical terms, definition lists, parameter tables, and visual elements to illustrate complex logic, such as system architecture diagrams, software component interaction diagrams, state machine diagrams, and sequence diagrams.
[0003] To construct the knowledge source for a Retrieval Augmented Generation (RAG) system, documents need to be segmented. Segmentation strategies include fixed-size segments, recursive character segments, or sentence / paragraph-based semantic segments. However, traditional segmentation strategies often result in the loss of logical connections across pages and modalities in AUTOSAR-reduced documents. Therefore, addressing the context fragmentation problem caused by traditional segmentation strategies has become a pressing issue.
[0004] The above content is only used to help understand the technical solution of the present invention and does not represent an admission that the above content is prior art. Summary of the Invention The main objective of this invention is to provide a method and system for semantic knowledge block partitioning based on AUTOSAR specification documents, aiming to solve the context fragmentation problem caused by traditional partitioning strategies.
[0005] To achieve the above objectives, this invention provides a method for semantic knowledge block partitioning based on AUTOSAR specification documents. The method includes: S1, standardize the AUTOSAR specification document; S2, extract multimodal features from the processed AUTOSAR specification document to generate multiple high-dimensional embedding vectors and multiple visual embedding vectors; S3, a formal knowledge base based on AUTOSAR domain ontology, constructs a document knowledge graph based on multiple high-dimensional embedding vectors and multiple visual embedding vectors; S4, the document knowledge graph is divided into semantic knowledge blocks using the semantic subgraph partitioning method.
[0006] Optionally, S1 includes: S1.1, parse the AUTOSAR specification document to obtain the original document data; S1.2 Clean the original document data and perform unified character encoding on the cleaned original document data.
[0007] Optionally, S2 includes: S2.1 Extract text features from the processed AUTOSAR specification document using the feature extractor of the LayoutLMv3 architecture; S2.2, Generate multiple high-dimensional embedding vectors based on the text features; S2.3, extract key visual features from the processed AUTOSAR specification document using a visual encoder based on the DiT architecture; S2.4, Generate multiple visual embedding vectors based on the key visual features.
[0008] Optionally, before step S3, the following steps are included: Define multiple entities from the AUTOSAR standard and determine the semantic relationships between them; A formal knowledge base for AUTOSAR domain ontology is constructed based on multiple entities and the semantic relationships between them.
[0009] Optionally, S3 includes: S3.1, multiple high-dimensional embedding vectors and multiple visual embedding vectors are respectively used as graph nodes; S3.2, Generate multiple sets of node pairs based on each graph node; S3.3, traverse multiple pairs of nodes and treat the traversed pairs of nodes as pairs of nodes to be processed; S3.4, The node pairs to be processed are matched with multiple sets of entity pairs in the formal knowledge base using entity linking technology; S3.5 After a successful match, the semantic relationship between the node pairs to be processed and the successfully matched entity pairs is obtained through a relation classifier, and the process returns to S3.3 until the semantic relationship between all node pairs is obtained. S3.6 Construct a document knowledge graph based on the semantic relationships between each graph node and all node pairs.
[0010] Optionally, obtaining the semantic relationship between the node pair to be processed and the successfully matched entity pair through a relation classifier includes: The nodes to be processed are concatenated with the successfully matched entity pairs to obtain joint features; The semantic relationship between the joint features and the successfully matched entity pairs is input into the relation classifier, which outputs the semantic relationship between the node pairs to be processed.
[0011] Optionally, S4 includes: S4.1 Select a seed node from multiple graph nodes; S4.2, Based on the seed node, perform a directed graph traversal of the document knowledge graph to determine the semantic cohesion of the current growing subgraph; S4.3 When the semantic cohesion is less than a preset threshold, the document knowledge graph is divided into semantic knowledge blocks according to the current growing subgraph.
[0012] Optionally, after S4, the following steps are included: S5, Sequentially segmented semantic knowledge blocks, generate retrieval embedding vectors based on the sequential semantic knowledge blocks; S6, the serialized semantic knowledge block and the corresponding retrieval embedding vector are stored in the vector database for knowledge indexing.
[0013] Furthermore, to achieve the above objectives, this invention also proposes a semantic knowledge block partitioning system based on AUTOSAR specification documents, wherein the semantic knowledge block partitioning system based on AUTOSAR specification documents includes: The processing module is used for standardizing AUTOSAR specification documents; The extraction module is used to extract multimodal features from the processed AUTOSAR specification document to generate multiple high-dimensional embedding vectors and multiple visual embedding vectors. The building module is used to construct a document knowledge graph based on a formal knowledge base of AUTOSAR domain ontology, using multiple high-dimensional embedding vectors and multiple visual embedding vectors. The partitioning module is used to partition the document knowledge graph into semantic knowledge blocks using the semantic subgraph partitioning method.
[0014] Furthermore, to achieve the above objectives, the present invention also proposes a semantic knowledge block partitioning device based on AUTOSAR specification documents. The device includes: a memory, a processor, and a semantic knowledge block partitioning program based on AUTOSAR specification documents stored in the memory and executable on the processor. The semantic knowledge block partitioning program based on AUTOSAR specification documents is configured to implement the steps of the semantic knowledge block partitioning method based on AUTOSAR specification documents as described above.
[0015] Furthermore, to achieve the above objectives, the present invention also proposes a storage medium storing a semantic knowledge block partitioning program based on AUTOSAR specification documents. When the semantic knowledge block partitioning program based on AUTOSAR specification documents is executed by a processor, it implements the steps of the semantic knowledge block partitioning method based on AUTOSAR specification documents as described above.
[0016] This invention first standardizes the AUTOSAR specification document, then extracts multimodal features from the processed document to generate multiple high-dimensional embedding vectors and multiple visual embedding vectors. Next, based on a formal knowledge base of the AUTOSAR domain ontology, a document knowledge graph is constructed using these high-dimensional and visual embedding vectors. Finally, the document knowledge graph is divided into semantic knowledge blocks using a semantic subgraph partitioning method. Compared to traditional block partitioning strategies that result in the loss of logical connections across pages and modalities in the AUTOSAR specification document, this invention, by partitioning the constructed document knowledge graph into semantic knowledge blocks using a semantic subgraph partitioning method, ensures that all logically related multimodal information is aggregated within the same knowledge block, thus solving the context fragmentation problem caused by traditional knowledge block partitioning methods. Attached Figure Description
[0017] Figure 1 This is a schematic diagram of the structure of the semantic knowledge block partitioning device based on the AUTOSAR specification document in the hardware operating environment involved in the embodiments of the present invention; Figure 2 This is a flowchart illustrating the first embodiment of the semantic knowledge block partitioning method for AUTOSAR-based documents according to the present invention. Figure 3 This is a schematic diagram of the relation prediction process in the first embodiment of the semantic knowledge block partitioning method for AUTOSAR-represented documents of the present invention. Figure 4 This is a schematic diagram of the overall process of the first embodiment of the semantic knowledge block partitioning method for AUTOSAR specification documents of the present invention. Figure 5 This is a structural block diagram of the first embodiment of the semantic knowledge block partitioning system for AUTOSAR-based documents according to the present invention.
[0018] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation
[0019] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.
[0020] Reference Figure 1 , Figure 1 This is a schematic diagram of the device structure for semantic knowledge block partitioning based on AUTOSAR specification documents in the hardware operating environment involved in the embodiments of the present invention.
[0021] like Figure 1As shown, the semantic knowledge block partitioning device based on the AUTOSAR specification document may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to implement communication between these components. The user interface 1003 may include a display screen or an input unit such as a keyboard; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wireless-Fidelity (Wi-Fi) interface). The memory 1005 may be high-speed random access memory (RAM) or stable non-volatile memory (NVM), such as a disk storage device. The memory 1005 may also optionally be a storage system independent of the aforementioned processor 1001.
[0022] Those skilled in the art will understand that Figure 1 The structure shown does not constitute a limitation on the semantic knowledge block partitioning device based on the AUTOSAR specification document, and may include more or fewer components than shown, or combine certain components, or have different component arrangements.
[0023] like Figure 1 As shown, the memory 1005, which serves as a storage medium, may include an operating system, a network communication module, a user interface module, and a semantic knowledge block partitioning program based on the AUTOSAR specification document.
[0024] exist Figure 1 In the semantic knowledge block partitioning device based on AUTOSAR specification documents shown, the network interface 1004 is mainly used for data communication with the network server; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and memory 1005 in the semantic knowledge block partitioning device based on AUTOSAR specification documents of the present invention can be set in the semantic knowledge block partitioning device based on AUTOSAR specification documents. The semantic knowledge block partitioning device based on AUTOSAR specification documents calls the semantic knowledge block partitioning program based on AUTOSAR specification documents stored in the memory 1005 through the processor 1001, and executes the semantic knowledge block partitioning method based on AUTOSAR specification documents provided in the embodiments of the present invention.
[0025] This invention provides a method for semantic knowledge block partitioning of documents based on AUTOSAR specifications, referring to... Figure 2, Figure 2 This is a flowchart illustrating the first embodiment of the semantic knowledge block partitioning method for AUTOSAR specification documents according to the present invention.
[0026] In this embodiment, the semantic knowledge block partitioning method based on AUTOSAR specification documents includes the following steps: S1, standardizes the AUTOSAR specification document.
[0027] It is easy to understand that the execution subject of this embodiment can be a semantic knowledge block partitioning system based on AUTOSAR specification documents with functions such as data processing, network communication and program execution, or other computer devices with similar functions. This embodiment does not limit it.
[0028] Furthermore, the AUTOSAR specification document is parsed to obtain the original document data; the original document data is cleaned, and the cleaned original document data is uniformly encoded.
[0029] It should be noted that the AUTOSAR specification document is a multimodal specification document guided by the AUTOSAR Domain Ontology (ADO), and the AUTOSAR specification document is a PDF file.
[0030] In the specific implementation, Python libraries (such as PyMuPDF and pandas) are used to extract all the original content (i.e., the original document data) from the PDF file, including the text stream, embedded images, and the row and column structure and cell content of tables. The extracted content is then cleaned, standardized in character encoding, and the page number and coordinate location information of each content element (text block, image, table) in the AUTOSAR specification document are recorded.
[0031] S2 extracts multimodal features from the processed AUTOSAR specification document to generate multiple high-dimensional embedding vectors and multiple visual embedding vectors.
[0032] Furthermore, textual features are extracted from the processed AUTOSAR reduction document using a feature extractor based on a multimodal pre-trained model (LayoutLMv3) architecture; multiple high-dimensional embedding vectors are generated based on the textual features; key visual features are extracted from the processed AUTOSAR reduction document using a visual encoder based on the DiT framework; and multiple visual embedding vectors are generated based on the key visual features.
[0033] In the specific implementation, for text blocks and table units, a feature extractor based on the LayoutLMv3 architecture is used to generate high-dimensional embedding vectors that simultaneously encode text semantics, 2D position, font style, and other text features. For images and charts (such as architecture diagrams and state machine diagrams), a visual encoder based on DiT (Data-efficient Image Transformer) is used to extract key visual features such as structure and shape to generate visual embedding vectors.
[0034] S3 is a formal knowledge base based on AUTOSAR domain ontology that constructs a document knowledge graph based on multiple high-dimensional embedding vectors and multiple visual embedding vectors.
[0035] In this embodiment, it is also necessary to predefine multiple entities in the AUTOSAR standard and determine the semantic relationships between each entity (i.e., all legal relationships with explicit engineering semantics); and construct a formal knowledge base of the AUTOSAR domain ontology based on the multiple entities and the semantic relationships between them.
[0036] There may be one or more semantic relationships between entities. The formal knowledge base stores multiple pairs of entity pairs with semantic relationships.
[0037] It should also be noted that the AUTOSAR domain ontology, in the form of a knowledge graph, precisely defines the entities in the AUTOSAR standard (such as SoftwareComponent, RunnableEntity, PortInterface, etc.) and all the legal relationships between them with clear engineering semantics (such as implements, isAllocatedTo, hasConstraint, etc.).
[0038] Furthermore, the processing method for constructing a document knowledge graph based on a formal knowledge base of AUTOSAR domain ontology using multiple high-dimensional embedding vectors and multiple visual embedding vectors is as follows: Multiple high-dimensional embedding vectors and multiple visual embedding vectors are respectively used as graph nodes; multiple sets of node pairs are generated based on each graph node; multiple sets of node pairs are traversed, and the traversed node pairs are used as node pairs to be processed; the node pairs to be processed are matched with multiple sets of entity pairs in the formal knowledge base using entity linking technology; after a successful match, the semantic relationship between the node pairs to be processed and the successfully matched entity pairs is obtained through a relation classifier, and the process of traversing multiple sets of node pairs and using the traversed node pairs as node pairs to be processed is repeated until the semantic relationship between all node pairs is obtained; the document knowledge graph is constructed based on the semantic relationship between multiple graph nodes and all node pairs.
[0039] In the specific implementation, since nodes and entities exist in the form of vectors, the cosine similarity between nodes and vectors can be calculated. Then, the calculated cosine similarities are sorted by value, and the entity pair corresponding to the maximum similarity value is selected. It is then determined whether the maximum similarity value is greater than a preset threshold. If the maximum similarity value is greater than the preset threshold, the match is considered successful; if it is less than the preset threshold, the match is considered unsuccessful. The semantic relationship between the node pairs to be processed needs to be manually labeled or the node pairs to be processed are directly input into the relationship classifier to predict the semantic relationship between the node pairs to be processed.
[0040] Furthermore, the processing method for obtaining the semantic relationship between the node pair to be processed and the successfully matched entity pair through the relation classifier is as follows: the node pair to be processed and the successfully matched entity pair are concatenated to obtain joint features; the joint features and the semantic relationship between the successfully matched entity pair are input into the relation classifier, and the most suitable semantic relationship is selected from all the semantic relationships between the successfully matched entity pairs so that the relation classifier outputs the semantic relationship between the node pair to be processed.
[0041] In this embodiment, reference Figure 3 , Figure 3 This diagram illustrates the relation prediction process of the first embodiment of the semantic knowledge block partitioning method based on AUTOSAR specification documents according to the present invention. For a pair of nodes (Node A, Node B), the system first matches the node pair (Node A, Node B) with entities in the pre-built AUTOSAR Domain Ontology (ADO) using entity linking technology. After a successful match, the embedding vectors of nodes A and B are concatenated with the embedding vectors of the linked ADO entities (e.g., RunnableEntity and Task in ADO) to form a knowledge-enhanced joint feature. Finally, the semantic relationship between RunnableEntity and Task is used as a valid relation candidate set. This joint feature and the valid relation candidate set (e.g., isAllocatedTo) are input into a relation classifier (e.g., a multilayer perceptron). The output category of this classifier is strictly limited to the relation types predefined in ADO. In this way, the system predicts the most probable semantic relation edge connecting nodes A and B that conforms to the AUTOSAR specification.
[0042] It's important to note that when predicting the relationship between any two nodes, the model not only analyzes their features but also actively queries ADO (Aspect-Oriented Documentation). The system first uses entity linking technology to match the node content with entities (i.e., concepts) in ADO, then retrieves all legal relationship types between the two concepts from ADO as a candidate set. This significantly narrows the prediction space. Subsequently, the relationship classification model's task becomes searching for evidence in the document's multimodal features to confirm the most probable relationship in the candidate set. This ensures that the graph construction process is strongly constrained by domain knowledge, guaranteeing its logical correctness.
[0043] S4, the document knowledge graph is divided into semantic knowledge blocks using the semantic subgraph partitioning method.
[0044] Furthermore, seed nodes are selected from multiple graph nodes; a directed graph traversal is performed on the document knowledge graph based on the seed nodes to determine the semantic cohesion of the current growing subgraph; when the semantic cohesion is less than a preset threshold or reaches a preset traversal depth, the expansion stops, and the document knowledge graph is divided into semantic knowledge blocks according to the current growing subgraph.
[0045] It should be understood that the currently generated subgraph is a knowledge subgraph accessed from the seed node to the current node (for example, starting from each seed node, the algorithm performs a bounded breadth-first search (BFS) along edges with strong semantic types (such as implements) in the graph, and the traversal process aggregates all logically closely related nodes, including knowledge subgraphs composed of text descriptions, parameter tables, graph references, etc.). The seed node can be selected by the user, or a selection rule can be preset by the user, such as the strongest relation edge.
[0046] In its implementation, the "chunking" problem is redefined as a "community discovery" problem on the document graph. Starting from a seed node (such as a SoftwareComponent node), it traverses the graph along edges with strong semantics, aggregating all logically closely related multimodal nodes (regardless of their physical location in the document) to form a self-contained, context-complete "semantic knowledge block" (i.e., a knowledge subgraph). This "knowledge block partitioning" method ensures that all logically related multimodal information is aggregated within the same knowledge block. This completely solves the context fragmentation problem caused by traditional chunking methods, providing downstream RAG systems with an unprecedented high-quality, high-fidelity knowledge source, thereby greatly reducing the risk of LLM illusions.
[0047] Because the output semantic knowledge blocks are structured, they contain explicit logical relationships between entities. This enables downstream intelligent question-answering systems not only to perform information retrieval but also to conduct reliable multi-hop reasoning based on this structured knowledge, answering more complex and in-depth engineering questions.
[0048] It should also be noted that the reference Figure 4 , Figure 4 This is a schematic diagram of the overall process of the first embodiment of the semantic knowledge block partitioning method based on AUTOSAR specification documents of the present invention. The semantic knowledge blocks are serialized, and retrieval embedding vectors are generated based on the serialized semantic knowledge blocks. The serialized semantic knowledge blocks and the corresponding retrieval embedding vectors are stored in a vector database for knowledge indexing.
[0049] In the specific implementation, each knowledge block (subgraph) is serialized into a standardized data structure (such as JSON), which fully preserves all its nodes, edges, multimodal content and metadata.
[0050] The serialized knowledge blocks are then fed into a persistent storage and indexing system for use by downstream RAG applications. The specific implementation is as follows: Generate an embedding vector for fast retrieval for each knowledge block (e.g., using the embedding vector of its seed node or the embedding of its content summary).
[0051] Store the serialized representation (JSON) of the knowledge block along with its embedding vector in a vector database (such as FAISS or Weaviate).
[0052] When the RAG system receives a user query, it first performs a similarity search in the vector database to retrieve the most relevant knowledge block. Then, it passes the complete, structured multimodal content of the knowledge block to the Large Language Model (LLM) as context to generate a high-fidelity answer.
[0053] It should also be noted that this solution efficiently and accurately transforms unstructured AUTOSAR specification documents into a structured knowledge base, significantly reducing manual construction costs and error rates, while ensuring the logical rigor and maintainability of the knowledge base.
[0054] In this embodiment, the AUTOSAR specification document is first standardized. Then, multimodal features are extracted from the processed AUTOSAR specification document to generate multiple high-dimensional embedding vectors and multiple visual embedding vectors. Next, a document knowledge graph is constructed based on a formal knowledge base of the AUTOSAR domain ontology, using the multiple high-dimensional embedding vectors and multiple visual embedding vectors. Finally, the document knowledge graph is divided into semantic knowledge blocks using a semantic subgraph partitioning method. Compared to the loss of logical association information across pages and modalities in the AUTOSAR specification document caused by traditional block partitioning strategies, this embodiment uses a semantic subgraph partitioning method to divide the constructed document knowledge graph into semantic knowledge blocks, ensuring that all logically related multimodal information is aggregated in the same knowledge block, thus solving the context fragmentation problem caused by traditional knowledge block partitioning methods.
[0055] Reference Figure 5 , Figure 5 This is a structural block diagram of the first embodiment of the semantic knowledge block partitioning system for AUTOSAR-based documents according to the present invention.
[0056] like Figure 5 As shown, the semantic knowledge block partitioning system for AUTOSAR-represented documents proposed in this embodiment of the invention includes: Processing module 5001 is used for standardization processing of AUTOSAR specification documents; Extraction module 5002 is used to extract multimodal features from the processed AUTOSAR specification document to generate multiple high-dimensional embedding vectors and multiple visual embedding vectors. Module 5003 is used to construct a document knowledge graph based on a formal knowledge base of AUTOSAR domain ontology, which is constructed from multiple high-dimensional embedding vectors and multiple visual embedding vectors. The partitioning module 5004 is used to partition the document knowledge graph into semantic knowledge blocks using the semantic subgraph partitioning method.
[0057] Other embodiments or specific implementations of the semantic knowledge block partitioning system based on AUTOSAR specification documents of the present invention can be referred to the above-described method embodiments, and will not be repeated here.
[0058] 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 system 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 system. 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 system that includes that element.
[0059] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.
[0060] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as read-only memory / random access memory, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.
[0061] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.
Claims
1. A method for semantic knowledge block partitioning of documents based on AUTOSAR specification, characterized in that, The method includes the following steps: S1, standardize the AUTOSAR specification document; S2, extract multimodal features from the processed AUTOSAR specification document to generate multiple high-dimensional embedding vectors and multiple visual embedding vectors; S3, a formal knowledge base based on AUTOSAR domain ontology, constructs a document knowledge graph based on multiple high-dimensional embedding vectors and multiple visual embedding vectors; S4, the document knowledge graph is divided into semantic knowledge blocks using the semantic subgraph partitioning method.
2. The method as described in claim 1, characterized in that, S1 includes: S1.1, parse the AUTOSAR specification document to obtain the original document data; S1.2 Clean the original document data and perform unified character encoding on the cleaned original document data.
3. The method as described in claim 1, characterized in that, The S2 includes: S2.1 Extract text features from the processed AUTOSAR specification document using the feature extractor of the LayoutLMv3 architecture; S2.2, Generate multiple high-dimensional embedding vectors based on the text features; S2.3, extract key visual features from the processed AUTOSAR specification document using a visual encoder based on the DiT architecture; S2.4, Generate multiple visual embedding vectors based on the key visual features.
4. The method as described in claim 1, characterized in that, Before S3, it includes: Define multiple entities from the AUTOSAR standard and determine the semantic relationships between them; A formal knowledge base for AUTOSAR domain ontology is constructed based on multiple entities and the semantic relationships between them.
5. The method as described in claim 4, characterized in that, The S3 includes: S3.1, multiple high-dimensional embedding vectors and multiple visual embedding vectors are respectively used as graph nodes; S3.2, Generate multiple sets of node pairs based on each graph node; S3.3, traverse multiple pairs of nodes and treat the traversed pairs of nodes as pairs of nodes to be processed; S3.4, The node pairs to be processed are matched with multiple sets of entity pairs in the formal knowledge base using entity linking technology; S3.5 After a successful match, the semantic relationship between the node pairs to be processed and the successfully matched entity pairs is obtained through a relation classifier, and the process returns to S3.3 until the semantic relationship between all node pairs is obtained. S3.6 Construct a document knowledge graph based on the semantic relationships between each graph node and all node pairs.
6. The method as described in claim 5, characterized in that, The step of obtaining the semantic relationship between the node pairs to be processed and the successfully matched entity pairs through a relation classifier includes: The nodes to be processed are concatenated with the successfully matched entity pairs to obtain joint features; The semantic relationship between the joint features and the successfully matched entity pairs is input into the relation classifier, which outputs the semantic relationship between the node pairs to be processed.
7. The method as described in claim 6, characterized in that, The S4 includes: S4.1 Select a seed node from multiple graph nodes; S4.2, Based on the seed node, perform a directed graph traversal of the document knowledge graph to determine the semantic cohesion of the current growing subgraph; S4.3 When the semantic cohesion is less than a preset threshold, the document knowledge graph is divided into semantic knowledge blocks according to the current growing subgraph.
8. The method as described in claim 1, characterized in that, Following S4, the following is included: S5, Sequentially segmented semantic knowledge blocks, generating retrieval embedding vectors based on the sequential semantic knowledge blocks; S6, the serialized semantic knowledge block and the corresponding retrieval embedding vector are stored in the vector database for knowledge indexing.
9. A semantic knowledge block partitioning system for AUTOSAR-represented documents, characterized in that, The system includes: The processing module is used for standardizing AUTOSAR specification documents; The extraction module is used to extract multimodal features from the processed AUTOSAR specification document to generate multiple high-dimensional embedding vectors and multiple visual embedding vectors. The building module is used to construct a document knowledge graph based on a formal knowledge base of AUTOSAR domain ontology, using multiple high-dimensional embedding vectors and multiple visual embedding vectors. The partitioning module is used to partition the document knowledge graph into semantic knowledge blocks using the semantic subgraph partitioning method.