EPM-driven document pipeline processing method and related equipment

Abstract

The invention relates to the technical field of intelligent processing of power generation documents, in particular to an EPM-driven document pipeline processing method and related equipment.The EPM-driven document pipeline processing device comprises four core modules: a data access module is connected with an EPM system in real time through a preset API interface, automatically obtains science and technology documents, carries out multi-format analysis and extracts text content and metadata; the intelligent classification module performs multi-dimensional automatic classification on the analyzed documents and generates labels based on a pre-training text classification model and a professional dictionary; the document segmentation module is used for identifying logic paragraph boundaries by adopting a semantic segmentation algorithm, carrying out structured analysis on complex elements and segmenting a document into semantically coherent text units; the vector storage module converts the text unit into a set dimension vector and stores the set dimension vector into a vector database. All the modules are connected in sequence, full-process intelligent processing from document access, classification and segmentation to vector storage is achieved, and efficient management and application of science and technology documents are supported.

Description

Technical Field

[0001] This invention relates to the field of intelligent document processing technology, specifically to an EPM-driven document pipeline processing method and related equipment. Background Technology

[0002] As the power generation industry transforms towards digitalization and intelligence, knowledge management has become a core element in improving scientific research innovation and operational efficiency. The massive amounts of scientific and technological documents (such as technical reports, experimental data, patent documents, and technical standards) generated by research institutions like the Power and Thermal Engineering Research Institute in their daily work are mostly stored in Project Management (EPM) systems. Currently, the core functions of EPM systems focus on project management processes, and the accompanying documents are usually stored in unstructured or semi-structured formats, creating a "data silo" phenomenon. To extract and utilize these knowledge assets, existing technologies mainly rely on manual retrieval, reading, and organization. For complex elements in scientific and technological documents (such as professional formulas, technical drawings, and data charts), traditional optical character recognition (OCR) or simple text parsing technologies struggle to accurately identify and extract their semantic information. In knowledge retrieval, existing systems are mostly based on keyword matching. Although vector database technology can improve recall through semantic similarity retrieval, the retrieval results are still isolated text fragments, lacking context and logical connections. In the field of artificial intelligence, large language models perform well in general scenarios, but when directly using original documents from EPM systems for domain adaptation, they face problems such as chaotic original document formats and inconsistent quality. Manual cleaning and annotation are costly, and in private environments with limited computing resources, general-purpose, large-parameter models are difficult to fine-tune and deploy efficiently.

[0003] However, the existing technologies mentioned above still have the following inherent shortcomings that urgently need to be addressed. First, the automation and intelligence of document processing are low, heavily reliant on manual operation, resulting in low efficiency. Specifically, long technical reports require manual segmentation into semantic fragments by chapter or technical topic, a time-consuming and labor-intensive process prone to errors due to differences in personnel expertise; the semantic recognition and extraction accuracy of complex elements is insufficient, limiting the depth and breadth of knowledge processing. Second, the knowledge presentation format is monotonous and lacks semantic connections, leading to difficulties in knowledge retrieval and application. Existing retrieval technologies cannot effectively connect relevant content scattered across different documents, and the retrieval results lack context and logical connections, making it difficult to support complex scenarios such as fault diagnosis and technical decision-making; the construction of domain knowledge graphs is insufficient, failing to achieve the leap from "document retrieval" to "knowledge reasoning." Finally, data assets have failed to effectively empower cutting-edge artificial intelligence technologies, and large language models have poor domain adaptability in the power generation field. The original documents are poorly formatted and of inconsistent quality, making them unsuitable as datasets for fine-tuning. Manual cleaning and labeling are extremely costly, and in private environments with limited computing resources, large, general-purpose models with numerous parameters are difficult to fine-tune and deploy efficiently, making it difficult for them to be implemented and generate practical value in the power generation field. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide an EPM-driven document pipeline processing method and related equipment to address the shortcomings of the prior art, thereby solving the technical problem that traditional document management systems only support keyword matching and cannot perform semantic similarity retrieval.

[0005] The objective of this invention is achieved through the following technical solutions: In a first aspect, the present invention provides an EPM-driven document pipeline processing system, comprising: The data access module is used to establish a connection with the EPM system through a preset API interface or message queue, and to obtain scientific and technological documents from the EPM system in real time, and to perform multi-format parsing on the scientific and technological documents to extract text content and metadata. The intelligent classification module, connected to the data access module, is used to automatically classify the text content and metadata in the parsed scientific and technological documents in multiple dimensions based on a pre-trained text classification model and a professional dictionary, and generate classification tags. The intelligent segmentation module, connected to the intelligent classification module, is used to identify the logical paragraph boundaries and classification tags in the scientific and technological document based on the semantic segmentation algorithm, and to perform structured parsing of complex element text in the scientific and technological document to segment the scientific and technological document and obtain semantically coherent text units. A vector storage module, connected to the intelligent segmentation module, is used to convert the text units into vectors of a set dimension and store the vectors in a vector database.

[0006] As a further improvement of the present invention, the data access module includes: The process involves acquiring scientific and technological documents, parsing various document formats within these documents, extracting plain text content and complex element text content, and using the plain text content as text metadata. The system locates and identifies complex element text content, converts mathematical formulas in the complex element text content into LaTeX format to obtain mathematical formula metadata, performs OCR text extraction and semantic content annotation on data charts in the complex element text content to obtain chart metadata, and establishes an association index between the chart metadata and the corresponding text units.

[0007] As a further improvement of the present invention, the pre-trained text classification model in the intelligent classification module adopts a model based on the BERT architecture and is fine-tuned under supervision on technical literature corpus in the field of power generation.

[0008] As a further improvement of the present invention, the intelligent segmentation module is specifically used for: By calculating the semantic similarity between adjacent sentences or paragraphs, and combining the semantic similarity with the chapter title structure analysis of scientific and technological documents, the optimal semantic segmentation point is determined. Semantic segmentation is performed based on the optimal semantic segmentation point. During the semantic segmentation process, the association between the segmented text unit and the complex elements contained in the segmented text unit is maintained.

[0009] As a further improvement of the present invention, it also includes a knowledge graph construction module connected to the intelligent segmentation module, the knowledge graph construction module comprising: Extract entities and relationships between entities from the text units segmented by the intelligent segmentation module; The entities are used as nodes, and the relationships between the entities are used as edges to construct a knowledge graph for the power generation field and store it in a graph database. The entities include at least equipment, fault types, and technical parameters, and the relationships between the entities include causal relationships, inclusion relationships, and application relationships.

[0010] As a further improvement of the present invention, the vector storage module converts the text unit into a vector of a set dimension based on a sentence embedding model; the sentence embedding model is the Qwen3-Embedding model.

[0011] Secondly, the present invention provides an EPM-driven document pipeline processing method for implementing the aforementioned EPM-driven document pipeline processing system, comprising: The original scientific and technological documents are obtained from the EPM system through API interface or message queue, and the documents are parsed in multiple formats to extract text content and metadata. Based on a pre-trained text classification model and a professional dictionary, the parsed scientific and technological documents are automatically classified in multiple dimensions and classification tags are generated. Based on the semantic segmentation algorithm, the logical paragraph boundaries in the scientific and technological document are identified, and the complex element text in the scientific and technological document is structured and parsed to segment the scientific and technological document and obtain semantically coherent text units. The text unit is converted into a vector of a set dimension, and the vector is stored in a vector database.

[0012] As a further improvement to the present invention, it also includes: Entities and relationships are extracted from text units to construct and store a knowledge graph in the power generation field; Monitor changes in technical documents within the EPM system, and after pipeline processing of newly added or updated technical documents, update them to the vector database and knowledge graph.

[0013] Thirdly, the present invention provides a computer-readable storage medium for storing one or more programs, the one or more programs including instructions that, when executed by a computing device, cause the computing device to perform the above-described EPM-driven document pipeline processing method.

[0014] Fourthly, the present invention provides a computing device, comprising: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs including steps for performing the above-described EPM-driven document pipeline processing method.

[0015] The beneficial effects of this invention are as follows: This invention provides an EPM-driven document pipeline processing system. Through a data access module, a real-time connection is established with the EPM system using a preset API interface or message queue, enabling automatic acquisition and multi-format parsing of scientific and technological documents. This directly extracts text content and metadata, improving data access efficiency and format compatibility. The intelligent classification module automatically classifies the parsed documents from multiple dimensions based on a pre-trained text classification model and a professional dictionary, generating classification labels to improve classification accuracy and adapt to multi-domain classification needs. The intelligent segmentation module uses semantic segmentation algorithms to identify logical paragraph boundaries and structurally parse complex text elements, segmenting the document into semantically coherent text units, enhancing the logical coherence and structure of text processing. The vector storage module converts text units into vectors of a set dimension and stores them in a vector database, achieving efficient vector storage and rapid similarity calculation. The synergistic effect of these modules forms a fully automated processing flow from real-time data access, multi-dimensional classification, structured segmentation to vector storage, significantly improving the efficiency, accuracy, and systematic storage capacity of scientific and technological document processing. Compared with existing technologies, it has a more efficient data processing flow and more accurate document management results. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of the structure of the EPM-driven document pipeline processing system in an embodiment of the present invention; Figure 2 This is an internal structural diagram of a computer device in an embodiment of the present invention. Detailed Implementation

[0018] To make the objectives and technical solutions of this invention clearer and easier to understand, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. The specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention.

[0019] The technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings and specific embodiments. The described embodiments are only some embodiments of the present invention, and not all embodiments.

[0020] Example 1 like Figure 1 As shown, this embodiment provides an EPM-driven document pipeline processing system, including: a data access module, used to establish a connection with the EPM system through a preset API interface or message queue, and to obtain scientific and technological documents from the EPM system in real time, and to perform multi-format parsing of the scientific and technological documents to extract text content and metadata; an intelligent classification module, connected to the data access module, used to automatically classify the parsed scientific and technological documents in multiple dimensions based on a pre-trained text classification model and a professional dictionary, and to generate classification tags; an intelligent segmentation module, connected to the intelligent classification module, used to identify the logical paragraph boundaries in the scientific and technological documents based on a semantic segmentation algorithm, and to perform structured parsing of complex element text in the scientific and technological documents, segmenting the scientific and technological documents to obtain semantically coherent text units; and a vector storage module, connected to the intelligent segmentation module, used to convert the text units into vectors of a set dimension and store the vectors in a vector database.

[0021] The working principle of this system is as follows: A data access module establishes a connection with the EPM system through a preset API interface or message queue, and acquires scientific and technological documents in real time for multi-format parsing. This constructs an automated data acquisition and preliminary structuring channel, overcoming the inefficiency of manually retrieving and extracting documents from the EPM system, and significantly improving the efficiency of initial data processing. The intelligent classification module automatically classifies documents in multiple dimensions based on a pre-trained text classification model and a domain-specific dictionary. It integrates general model capabilities with prior domain knowledge to understand and tag document content, improving the professionalism and accuracy of document classification, and providing a foundation for subsequent... The subsequent processing provides structured guidance; the intelligent segmentation module identifies logical paragraph boundaries based on semantic segmentation algorithms and performs structured parsing of complex elements in the document to obtain semantically coherent text units. This achieves collaborative understanding and segmentation of the document's deep semantic logic and multimodal elements, resulting in the technical effect of ensuring the integrity, logic, and element correlation of the semantic units after document segmentation. The vector storage module transforms the semantically coherent text units into high-dimensional vectors and stores them in a vector database, mapping unstructured text information into computable and comparable vector forms and establishing an index, transforming document content into a knowledge form that supports efficient semantic retrieval. These modules work together to form a complete technical closed loop from automatic access to original documents, intelligent understanding, structured segmentation to vectorized storage. Ultimately, this results in the overall technical effect of automating and intelligentizing the entire document processing process, fundamentally replacing inefficient manual processing modes, and forming a vectorized knowledge base that supports semantic retrieval.

[0022] The data access module includes: acquiring scientific and technological documents, parsing various document formats in the documents to extract plain text content and complex element text content, and using the plain text content as text metadata; locating and recognizing complex element text content, converting mathematical formulas in the complex element text content into LaTeX format to obtain mathematical formula metadata; performing OCR text extraction and semantic content annotation on data charts in the complex element text content to obtain chart metadata, and establishing an association index between the chart metadata and the corresponding text units.

[0023] The pre-trained text classification model in the intelligent classification module is based on the BERT architecture and has been fine-tuned under supervision on technical literature corpus in the power generation field.

[0024] The intelligent segmentation module is specifically used to determine the optimal semantic segmentation point by calculating the semantic similarity between adjacent sentences or paragraphs, and then combining the semantic similarity with the chapter title structure analysis of the scientific document. Based on the optimal semantic segmentation point, semantic segmentation is performed, and during the semantic segmentation process, the association between the segmented text units and the complex elements contained in the segmented text units is maintained.

[0025] It also includes a knowledge graph construction module connected to the intelligent segmentation module. The knowledge graph construction module includes: extracting entities and relationships between entities from the text units segmented by the intelligent segmentation module; constructing a knowledge graph in the power generation field and storing it in a graph database, with entities as nodes and relationships between entities as edges; entities include at least equipment, fault types, and technical parameters, and relationships between entities include causal relationships, inclusion relationships, and application relationships.

[0026] The vector storage module transforms text units into vectors of a set dimension based on a sentence embedding model; the sentence embedding model is the Qwen3-Embedding model.

[0027] The specific implementation is as follows: The vector storage module is responsible for converting the semantically segmented text units into high-dimensional vector representations. Specifically, the Qwen3-Embedding model is used as the sentence embedding model. Each text unit is concatenated into a complete string and input into the Qwen3-Embedding model. The model outputs a 768-dimensional floating-point vector, which represents the semantic information of the text unit. These vectors, along with their corresponding text unit IDs and source document information, are stored in a dedicated vector database. This embodiment uses the Milvus vector database. In practical applications, other publicly available models such as BGE-M3 and text2vec can be used for the sentence embedding model, and other systems that support vector retrieval, such as Weaviate and Chroma, can also be used for the vector database. This embodiment does not impose any limitations. A deep learning model is used to compress and encode the semantic information of the text into fixed-dimensional vectors, and these vectors are stored in a database that supports fast similarity calculation, transforming the semantic search problem of text into a spatial distance calculation problem of vectors.

[0028] Example 2 Based on the EPM-driven document pipeline processing system proposed in Example 1, this example further illustrates a system integrating data access, intelligent processing, and asset construction. The core of this solution lies in an end-to-end automated pipeline driven by the EPM system.

[0029] At the data access and preprocessing layer, the system establishes real-time or timed data channels with the EPM system through standard API interfaces and message queues, automatically acquiring newly generated or updated scientific and technological documents (such as technical reports, patents, and experimental data). The system has a built-in multi-format parsing engine capable of processing various formats such as PDF, Word, and CAD drawings. It not only extracts plain text but also accurately identifies and locates complex elements in documents, such as mathematical formulas (and converts them to LaTeX standard format), data charts, and references, laying the foundation for subsequent in-depth structured processing.

[0030] At the core intelligent processing pipeline layer, the system executes three key operations. First, intelligent classification: the system utilizes a pre-trained language model fine-tuned on technical literature in the power generation field, combined with a constructed domain-specific dictionary, to automatically classify documents in a multi-dimensional and fine-grained manner, such as tagging them by technical field, document type, and equipment involved. Second, semantic segmentation: for long technical documents, a comprehensive strategy based on semantic similarity calculation and chapter structure analysis is adopted to automatically segment the document into semantically coherent and appropriately sized paragraphs (such as "Experimental Methods," "Data Results," and "Analysis Conclusions"), maintaining the association with the aforementioned complex elements. Third, vectorization and storage: using the domain-optimized sentence embedding model Qwen3-Embedding, the segmented text fragments are converted into high-dimensional vectors and stored in the dedicated vector database Milvus. Simultaneously, the extracted key entities and relationships are stored in a graph database, gradually constructing a knowledge graph for the power generation field.

[0031] At the data asset construction and application layer, this solution transforms the processed results into directly usable data assets. On one hand, vector fragments and knowledge graphs together form a powerful foundation for knowledge retrieval and question answering, supporting semantic similarity-based retrieval and graph-based reasoning question answering. On the other hand, the system utilizes templates and rules to automatically combine high-quality text fragments and summary / question-answer pairs into datasets for supervised fine-tuning of large models.

[0032] Example 3 Based on the EPM-driven document pipeline processing system in Embodiment 1, this embodiment proposes a processing method for an EPM-driven document pipeline processing system. The method includes: obtaining original scientific and technological documents from the EPM system via an API interface or message queue, and performing multi-format parsing on the documents to extract text content and metadata; automatically classifying the parsed documents in multiple dimensions based on a pre-trained text classification model and a professional dictionary, and generating classification labels; identifying logical paragraph boundaries in the documents based on a semantic segmentation algorithm, and performing structured parsing on complex element text in the documents to segment the documents into semantically coherent text units; converting the text units into vectors of a set dimension, and storing the vectors in a vector database.

[0033] Entities and relationships are extracted from text units to construct and store a knowledge graph in the power generation field; changes in technical documents in the EPM system are monitored, and newly added or updated technical documents are processed in a pipeline and then updated to the vector database and knowledge graph.

[0034] This method is executed by a computer program deployed on a server, and the specific steps are as follows: Data access and parsing. It calls the API provided by the EPM system or subscribes to its Kafka message topics for publishing document updates to obtain the original files of newly generated technical documents. It uses a pre-defined tool library to parse PDF, DOCX, and other format files, extracting the plain text content. Simultaneously, it uses an OCR engine to recognize mathematical formulas and data charts in the document, converting formulas to LaTeX format, extracting chart text and types, and generating structured metadata containing text, formulas, charts, and their positional relationships.

[0035] The document text content (e.g., title and abstract) is input into a pre-trained text classification model. This model is based on the BERT architecture and fine-tuned on a corpus of technical literature in the power generation field. The model outputs the probability that the document belongs to each predefined category (technical field, document type, device object), and the categories with probabilities greater than a threshold (e.g., 0.7) are used as the labels for the document.

[0036] For the categorized full-text documents, the Sentence-BERT model is used to calculate the semantic similarity between sentences, and the semantic segmentation points are determined by combining the document's chapter title structure (identified through regular expressions to identify title formats). The document is then segmented into multiple text units according to these segmentation points. During this process, based on the generated association index, each text unit is bound to the metadata of complex elements such as mathematical formulas and data charts falling within its text scope.

[0037] Each text unit (including its associated complex element description text) is input into the Qwen3-Embedding model to generate a 768-dimensional semantic vector. The vectors of all text units and their metadata (such as ID, source document, and category tags) are then batch-imported into the Milvus vector database to build an index. The working principle is as follows: through the automated pipeline execution of the above four core steps, the original unstructured documents in the EPM system are gradually transformed into a collection of knowledge fragments with rich tags, clear structure, and established semantic indexes.

[0038] Example 4 In another embodiment of the present invention, a computer-readable storage medium is provided as a storage component within a terminal device, the function of which is to store programs and data. It should be noted that the computer-readable storage medium here encompasses not only the built-in storage components of the terminal device but also extended storage components supported by the device. Essentially, it is a tangible medium capable of containing or storing programs that can be invoked by or in conjunction with an instruction execution system, device, or apparatus. This storage medium provides storage areas for the terminal's operating system and stores one or more instructions suitable for processor loading and execution, which can constitute one or more computer programs containing program code.

[0039] Specifically, examples of computer-readable storage media (a non-exclusive list) include: electrical connections with one or more wires, portable disks, hard disks, random access memory, read-only memory, erasable programmable read-only memory, optical fibers, portable optical disc read-only memory, optical storage devices, magnetic storage devices, or any reasonable combination of the above types.

[0040] The storage medium may also include data signals propagated as part of a baseband portion or a carrier wave, carrying readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any reasonable combination of both. Furthermore, computer-readable storage medium may also refer to other readable media besides conventional readable storage media, capable of sending, propagating, or transmitting programs for use or operation by an instruction execution system, apparatus, or device. Program code on the storage medium can be transmitted via any suitable medium, including but not limited to wireless, wired, optical fiber, or any reasonable combination thereof.

[0041] The program code used to implement the operations of this invention can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, as well as conventional procedural programming languages ​​such as C. The execution modes of the program code include: running entirely on the user's computing device, running partially on the user's device as a standalone software package, running partially in a distributed manner on both the user's device and a remote computing device, or running entirely on a remote computing device or server. When a remote computing device is involved, the device can be connected to the user's computing device via any type of network such as a local area network (LAN) or a wide area network (WAN), or connected to an external computing device via the Internet through an Internet service provider.

[0042] The processor is capable of loading and executing one or more instructions stored in a computer-readable storage medium to implement the corresponding steps of the EPM-driven document pipeline processing system described in Example 1.

[0043] Example 5 Figure 2This is a schematic diagram of a computer device provided according to an embodiment of the present invention.

[0044] Please see Figure 2 The terminal device is a computer device. In this embodiment, the computer device 60 includes a processor 61, a memory 62, and a computer program 63 stored in the memory 62 and executable on the processor 61. When executed by the processor 61, the computer program 63 implements the EPM-driven document pipeline processing method of this embodiment. To avoid repetition, details are omitted here. Alternatively, when executed by the processor 61, the computer program 63 implements the functions of each model / unit in the computing system that constitutes the EPM-driven document pipeline processing method of this embodiment. To avoid repetition, details are omitted here.

[0045] Computer device 60 can be a desktop computer, laptop, handheld computer, cloud server, or other computing device. Computer device 60 may include, but is not limited to, a processor 61 and a memory 62. Those skilled in the art will understand that... Figure 2 This is merely an example of computer device 60 and does not constitute a limitation on computer device 60. It may include more or fewer components than shown, or combine certain components, or different components. For example, computer device may also include input / output devices, network access devices, buses, etc.

[0046] The processor 61 may be a central processing unit (CPU), or other general-purpose processors, CPUs, graphics processing units (GPUs), digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, quantum computing-based data processing logic units, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0047] The memory 62 can be an internal storage unit of the computer device 60, such as a hard disk or RAM of the computer device 60. The memory 62 can also be an external storage device of the computer device 60, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. equipped on the computer device 60.

[0048] Furthermore, the memory 62 may include both internal storage units and external storage devices of the computer device 60. The memory 62 is used to store computer programs and other programs and data required by the computer device. The memory 62 can also be used to temporarily store data that has been output or will be output.

[0049] Any references to memory, databases, or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (Read-Only Memory). Memory includes ROM, magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0050] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

Claims

1. An EPM-driven document pipeline processing system, characterized in that, include: The data access module is used to establish a connection with the EPM system through a preset API interface or message queue, and to obtain scientific and technological documents from the EPM system in real time, and to perform multi-format parsing on the scientific and technological documents to extract text content and metadata. The intelligent classification module, connected to the data access module, is used to automatically classify the text content and metadata in the parsed scientific and technological documents in multiple dimensions based on a pre-trained text classification model and a professional dictionary, and generate classification tags. The intelligent segmentation module, connected to the intelligent classification module, is used to identify the logical paragraph boundaries and classification tags in the scientific and technological document based on the semantic segmentation algorithm, and to perform structured parsing of complex element text in the scientific and technological document to segment the scientific and technological document and obtain semantically coherent text units. A vector storage module, connected to the intelligent segmentation module, is used to convert the text units into vectors of a set dimension and store the vectors in a vector database.

2. The EPM-driven document pipeline processing system according to claim 1, characterized in that, The data access module includes: The process involves acquiring scientific and technological documents, parsing various document formats within these documents, extracting plain text content and complex element text content, and using the plain text content as text metadata. The system locates and identifies complex element text content, converts mathematical formulas in the complex element text content into LaTeX format to obtain mathematical formula metadata; it also performs OCR text extraction and semantic content annotation on data charts in the complex element text content to obtain chart metadata, and establishes an association index between the chart metadata and the corresponding text units.

3. The EPM-driven document pipeline processing system according to claim 1, characterized in that, The pre-trained text classification model in the intelligent classification module is based on the BERT architecture and is fine-tuned under supervision on technical literature corpus in the power generation field.

4. The EPM-driven document pipeline processing system according to any one of claims 1 to 3, characterized in that, The intelligent segmentation module is specifically used for: By calculating the semantic similarity between adjacent sentences or paragraphs, and combining the semantic similarity with the chapter title structure analysis of scientific and technological documents, the optimal semantic segmentation point is determined. Semantic segmentation is performed based on the optimal semantic segmentation point. During the semantic segmentation process, the association between the segmented text unit and the complex elements contained in the segmented text unit is maintained.

5. The EPM-driven document pipeline processing system according to claim 4, characterized in that, It also includes a knowledge graph construction module connected to the intelligent segmentation module, the knowledge graph construction module comprising: Extract entities and relationships between entities from the text units segmented by the intelligent segmentation module; The entities are used as nodes, and the relationships between the entities are used as edges to construct a knowledge graph for the power generation field and store it in a graph database. The entities include at least equipment, fault types, and technical parameters, and the relationships between the entities include causal relationships, inclusion relationships, and application relationships.

6. The EPM-driven document pipeline processing system according to claim 1, characterized in that, The vector storage module converts the text units into vectors of a set dimension based on a sentence embedding model; the sentence embedding model is the Qwen3-Embedding model.

7. An EPM-driven document pipeline processing method, used to implement the EPM-driven document pipeline processing system according to any one of claims 1 to 6, characterized in that, include: The original scientific and technological documents are obtained from the EPM system through API interface or message queue, and the documents are parsed in multiple formats to extract text content and metadata. Based on a pre-trained text classification model and a professional dictionary, the parsed scientific and technological documents are automatically classified in multiple dimensions and classification tags are generated. Based on the semantic segmentation algorithm, the logical paragraph boundaries in the scientific and technological document are identified, and the complex element text in the scientific and technological document is structured and parsed to segment the scientific and technological document and obtain semantically coherent text units. The text unit is converted into a vector of a set dimension, and the vector is stored in a vector database.

8. The EPM-driven document pipeline processing method according to claim 7, characterized in that, Also includes: Entities and relationships are extracted from text units to construct and store a knowledge graph in the power generation field; Monitor changes in technical documents within the EPM system, and after pipeline processing of newly added or updated technical documents, update them to the vector database and knowledge graph.

9. A computer-readable storage medium for storing one or more programs, characterized in that, The one or more programs include instructions that, when executed by a computing device, cause the computing device to perform the EPM-driven document pipeline processing method of claim 7 or 8.

10. A computing device, characterized in that, include: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs including steps for performing the EPM-driven document pipeline processing method of claim 7 or 8.

Images