A standard-based power grid infrastructure feasibility study review auxiliary method and system

Abstract

The present application relates to the field of artificial intelligence technology, and in particular, provides a standard-based power grid infrastructure feasibility study review auxiliary method and system, which first identifies the embedded semantics of power grid infrastructure description document examples, and these standard documents belong to the target project text knowledge graph. Using these description document examples, index entity mining is performed on the target project text knowledge graph to discover the existing feasibility review index entities. By performing semantic clustering operations on the description document embedded semantics, multiple document embedding semantic clustering sets are obtained. The description document embedded semantics corresponding to the topic embedding semantics of each cluster set is perturbed to generate new description document embedded semantic examples to debug the knowledge graph analysis network, thereby realizing standard index entity mining to determine whether there are standard feasibility review index entities in its project text knowledge graph, so as to provide powerful reference and information assistance for the power grid infrastructure feasibility study review, which helps to improve the intelligence level of the review.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a standard-based power grid infrastructure feasibility study review auxiliary method and system. Background Art

[0002] Electricity is the basic energy source for production in various industries and people's lives. How to ensure the safe, stable operation and efficient supply of the power grid that transports electricity is of paramount importance. In each link of the power grid, such as equipment technical specifications, infrastructure, dispatching, operation and inspection, there are corresponding series of standards. Specifically, in the power business scenario, there are corresponding standards and specifications for planning and design, procurement, supervision, arrival acceptance, testing and inspection, operation and maintenance, and decommissioning. Through these numerous standards and specifications, the stable operation and efficient supply of the power grid are ensured, and the safety of equipment and personnel is ensured.

[0003] In the field of power grid infrastructure, the construction and analysis of project text knowledge graphs are crucial for project feasibility studies and reviews. However, traditional knowledge graph analysis methods often face problems such as chaotic data organization, inaccurate semantic recognition, and low entity mining efficiency when processing a large number of standard power grid infrastructure description documents. These problems directly lead to the difficulty in accurately identifying and extracting key entities in the project text knowledge graph, which in turn affects the decision-making quality and efficiency of project review. Summary of the invention

[0004] In order to solve the above problems, the present application provides a standard-based power grid infrastructure feasibility study review auxiliary method and system.

[0005] The present application embodiment provides a standard-based power grid infrastructure feasibility study review auxiliary method, which is applied to a big data processing system. The method includes:

[0006] Identify description document embedding semantic examples of each of a number of power grid infrastructure description document examples; wherein the power grid infrastructure description document examples belong to a target project text knowledge graph, and when index entity mining is performed on the target project text knowledge graph based on the power grid infrastructure description document examples, the obtained index entity mining results reflect that the target project text knowledge graph has a feasibility study review index entity;

[0007] Performing a semantic clustering operation on a plurality of the description document embedding semantic examples to obtain at least one document embedding semantic clustering set;

[0008] The first description document embedding semantic examples corresponding to the topic embedding semantics of each of the document embedding semantic clusters are disturbed to obtain second description document embedding semantic examples corresponding to each of the first description document embedding semantic examples. The second description document embedding semantic examples are used for debugging the knowledge graph analysis network. The knowledge graph analysis network is used to perform index entity mining on whether there is a feasibility review index entity in the project text knowledge graph in the power grid infrastructure description document to be analyzed.

[0009] In some optional schemes, before perturbing the first description document embedding semantic examples corresponding to the topic embedding semantics of each of the document embedding semantic clusters to obtain the second description document embedding semantic examples corresponding to each of the first description document embedding semantic examples, the method further includes: constructing at least one first perturbation semantic feature, and there are differences between the first perturbation semantic features;

[0010] The perturbation processing is performed on the first description document embedding semantic examples corresponding to the topic embedding semantics of each of the document embedding semantic clusters to obtain the second description document embedding semantic examples corresponding to each of the first description document embedding semantic examples, including: implementing the following steps for each of the document embedding semantic clusters respectively: setting the first perturbation semantic features for the first description document embedding semantic examples corresponding to the topic embedding semantics of the document embedding semantic clusters respectively, to obtain at least one second description document embedding semantic example corresponding to the first description document embedding semantic example.

[0011] In some optional solutions, constructing at least one first disturbance semantic feature includes:

[0012] Performing feature encoding at least once on the text knowledge satisfying the first thermal condition to obtain at least one feature encoding result, and using each of the feature encoding results as each of the first disturbance semantic features, wherein there are differences between the feature encoding results;

[0013] Alternatively, text knowledge satisfying at least one second thermal condition is obtained, and for each of the second thermal conditions, the first disturbance semantic feature is obtained by feature encoding from the text knowledge satisfying the second thermal condition, and there are differences between the second thermal conditions.

[0014] In some optional schemes, before perturbing the first description document embedding semantic examples corresponding to the topic embedding semantics of each of the document embedding semantic clusters to obtain the second description document embedding semantic examples corresponding to each of the first description document embedding semantic examples, the method further includes: constructing second perturbation semantic features corresponding to each of the document embedding semantic clusters;

[0015] The method of performing perturbation processing on the first description document embedding semantic examples corresponding to the topic embedding semantics of each of the document embedding semantic clusters to obtain the second description document embedding semantic examples corresponding to each of the first description document embedding semantic examples comprises: implementing the following steps for each of the document embedding semantic clusters respectively: for the first description document embedding semantic examples corresponding to the topic embedding semantics of the document embedding semantic clusters, setting the second perturbation semantic features corresponding to the document embedding semantic clusters to obtain the second description document embedding semantic examples corresponding to the first description document embedding semantic examples.

[0016] In some optional solutions, constructing the second perturbation semantic features corresponding to each of the document embedding semantic cluster sets includes:

[0017] The following steps are respectively implemented for each document embedding semantic cluster set:

[0018] Determining the semantic thermal conditions of embedding the document into the semantic clustering set;

[0019] Constructing text knowledge that satisfies the semantic thermal conditions of the document embedding semantic clustering set;

[0020] From the text knowledge satisfying the semantic thermal conditions of the document embedding semantic cluster set, feature encoding is performed to obtain a second perturbation semantic feature corresponding to the document embedding semantic cluster set.

[0021] In some optional schemes, after perturbing the first description document embedding semantic examples corresponding to the topic embedding semantics of each of the document embedding semantic clusters to obtain the second description document embedding semantic examples corresponding to each of the first description document embedding semantic examples, the method further includes:

[0022] Obtaining a derived knowledge graph, wherein the derived knowledge graph does not include the target project text knowledge graph;

[0023] Extracting a derived knowledge graph semantic vector of the derived knowledge graph;

[0024] Do one of the following:

[0025] Integrate the second description document embedding semantic example and the derived knowledge graph semantic vector to obtain a third description document embedding semantic example;

[0026] The derived knowledge graph semantic vector is determined as the third description document embedding semantic example.

[0027] In some optional schemes, after perturbing the first description document embedding semantic examples corresponding to the topic embedding semantics of each of the document embedding semantic clusters to obtain the second description document embedding semantic examples corresponding to each of the first description document embedding semantic examples, the method further includes:

[0028] Extracting an associated description document embedding semantic example of an associated description document example, and constructing a third perturbation semantic feature; wherein the associated description document example belongs to a target project text knowledge graph, and when index entity mining is performed on the target project text knowledge graph based on the associated description document example, the obtained index entity mining result reflects that there is no feasibility review index entity in the target project text knowledge graph;

[0029] According to the associated description document embedding semantic example and the third perturbation semantic feature, calling the perturbation semantic construction network to construct a fourth description document embedding semantic example;

[0030] The second description document embedded semantic example and the fourth description document embedded semantic example are integrated to obtain a fifth description document embedded semantic example.

[0031] In some optional embodiments, the method further comprises:

[0032] Obtaining a basic perturbation semantic construction network and a first associated description document embedding semantic example of the first associated description document example, and constructing a fourth perturbation semantic feature;

[0033] According to the first associated description document embedding semantic example and the fourth perturbation semantic feature, calling the basic perturbation semantic construction network to construct a sixth description document embedding semantic example;

[0034] Calling a document semantic processing network to parse the sixth description document embedding semantic example to obtain a document semantic parsing viewpoint, wherein the document semantic parsing viewpoint indicates a confidence weight that the sixth description document embedding semantic example is a priori description document embedding semantic example;

[0035] Determining an index value of a training error of the basic perturbation semantic construction network based on a priori description document embedding semantic examples and the document semantic parsing viewpoint;

[0036] According to the indicator value of the training error, the basic perturbation semantic construction network is debugged to obtain the perturbation semantic construction network.

[0037] In some optional solutions, the step of integrating the second description document embedded semantic example and the fourth description document embedded semantic example to obtain a fifth description document embedded semantic example includes:

[0038] Obtaining a first feature enhancement coefficient of the second description document embedded semantic example, and obtaining a second feature enhancement coefficient of the fourth description document embedded semantic example;

[0039] According to the first feature enhancement coefficient and the second feature enhancement coefficient, the second description document embedded semantic example and the fourth description document embedded semantic example are semantically enhanced to obtain the fifth description document embedded semantic example.

[0040] In some optional schemes, after perturbing the first description document embedding semantic examples corresponding to the topic embedding semantics of each of the document embedding semantic clusters to obtain the second description document embedding semantic examples corresponding to each of the first description document embedding semantic examples, the method further includes:

[0041] Performing joint discrimination on a plurality of second description document embedded semantic examples respectively to obtain joint discrimination results of each second description document embedded semantic example, wherein the joint discrimination results represent the commonality measurement weight between the second description document embedded semantic example and the prior description document embedded semantic example;

[0042] Based on the joint discrimination results of each of the second description document embedded semantic examples, a target description document embedded semantic example whose joint discrimination result meets the joint discrimination requirement is selected from a plurality of the second description document embedded semantic examples.

[0043] In some optional schemes, after performing semantic clustering operation on several examples of document embedding semantics to obtain at least one document embedding semantic cluster set, the method further includes: determining a first document embedding semantic cluster set from the at least one document embedding semantic cluster set, and determining feature differences between each second document embedding semantic cluster set and the first document embedding semantic cluster set, wherein the second document embedding semantic cluster set is a document embedding semantic cluster set in the at least one document embedding semantic cluster set that is different from the first document embedding semantic cluster set; determining a third document embedding semantic cluster set whose feature difference is higher than a feature difference threshold, and a fourth document embedding semantic cluster set whose feature difference is not higher than the feature difference threshold, from the at least one document embedding semantic cluster set;

[0044] The perturbation processing of the first description document embedding semantic examples corresponding to the topic embedding semantics of each of the document embedding semantic clusters to obtain the second description document embedding semantic examples corresponding to each of the first description document embedding semantic examples includes: perturbation processing of the first description document embedding semantic examples corresponding to the topic embedding semantics of each of the third document embedding semantic clusters to obtain the second description document embedding semantic examples corresponding to each of the third document embedding semantic clusters; perturbation processing of the first description document embedding semantic examples corresponding to the topic embedding semantics of each of the fourth document embedding semantic clusters to obtain the second description document embedding semantic examples corresponding to each of the fourth document embedding semantic clusters; wherein the number of the second description document embedding semantic examples corresponding to each of the third document embedding semantic clusters is less than the number of the second description document embedding semantic examples corresponding to each of the fourth document embedding semantic clusters.

[0045] An embodiment of the present application provides a big data processing system, comprising at least one processor and a memory; the memory stores computer-executable instructions; the at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor executes the above method.

[0046] An embodiment of the present application provides a computer-readable storage medium on which a computer program is stored. The computer program implements the above method when running.

[0047] By deeply identifying the embedded semantics in the power grid infrastructure description document, the embodiment of the present application can more accurately capture the key information in the document and provide a more accurate data basis for the subsequent knowledge graph analysis. At the same time, through semantic clustering operations and perturbation processing technology, the embodiment of the present application further improves the efficiency and accuracy of data processing, so that the knowledge graph analysis network can more efficiently identify and mine feasibility review index entities during debugging and optimization.

[0048] In detail, firstly, the embedded semantics of power grid infrastructure description document examples are identified. These standard documents belong to the target project text knowledge graph. Using these description document examples, the target project text knowledge graph is mined for index entities to discover the existing feasibility review index entities. By performing semantic clustering operations on the description document embedding semantics, multiple document embedding semantic clustering sets are obtained. The description document embedding semantics corresponding to the topic embedding semantics of each clustering set are perturbed to generate new description document embedding semantic examples to debug the knowledge graph analysis network, thereby realizing standard index entity mining to determine whether there are standard feasibility review index entities in its project text knowledge graph, so as to provide powerful reference and information assistance for the power grid infrastructure feasibility review, which will help improve the intelligence level of the review. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 A flow chart of a standard-based power grid infrastructure feasibility review auxiliary method provided in an embodiment of the present application.

[0050] Figure 2 A schematic diagram of the structure of a big data processing system 200 provided in an embodiment of the present application. DETAILED DESCRIPTION

[0051] In order to better understand the above technical scheme, the technical scheme of the present application is described in detail below through the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present application and the specific features in the embodiments are detailed descriptions of the technical scheme of the present application, rather than limitations on the technical scheme of the present application. In the absence of conflict, the embodiments of the present application and the technical features in the embodiments can be combined with each other.

[0052] Figure 1 A standard-based power grid infrastructure feasibility study review auxiliary method is shown, which is applied to a big data processing system. The method includes the following steps 100 to 130.

[0053] Step 110, identifying description document embedded semantic examples of respective several power grid infrastructure description document examples; wherein, the power grid infrastructure description document examples belong to the target project text knowledge graph, and when index entity mining is performed on the target project text knowledge graph based on the power grid infrastructure description document examples, the obtained index entity mining results reflect that the target project text knowledge graph has a feasibility study review index entity.

[0054] In an embodiment of the present application, the example of a power grid infrastructure description document is a detailed description of the construction of power grid infrastructure, which may include detailed information on each aspect of the project, such as planning, design, and implementation. They are not only a reference for project implementation, but also an important basis for project review, evaluation, and risk analysis. These example documents are usually used as input to a big data processing system to train models, extract key information, and establish knowledge graphs. In actual applications, standard power grid infrastructure documents are generally corresponding PDF documents, such as standard operating instructions, technical specifications, typical design documents, etc., and the corresponding detailed information data can be extracted through OCR technology.

[0055] The embedded semantic examples of description documents refer to the key semantic information extracted after in-depth analysis of power grid infrastructure description documents through natural language processing and machine learning technology. This information is converted into a numerical representation (i.e., an embedded vector) that can capture the theme, intent, and key concepts in the document, making it easier for computers to process and analyze it efficiently. The embedded semantic examples are the basis for big data processing systems to understand and mine document content.

[0056] In power grid infrastructure projects, the target project text knowledge graph can include various project-related entities, such as equipment, materials, processes, specifications, etc., as well as the relationships between them. This knowledge graph is an important basis for big data processing systems to conduct analysis and decision-making, and can help the system understand the project content more comprehensively and improve the accuracy of information mining.

[0057] Index entity mining refers to identifying and extracting key entities related to power grid infrastructure projects from a large amount of text data through specific algorithms and technical means. These entities can be equipment names, material specifications, technical standards, etc., which are crucial for project review and management. Index entity mining is a core function of big data processing systems and can significantly improve the efficiency of information retrieval and analysis.

[0058] Feasibility study index entities refer to key information closely related to project review in project feasibility study reports or related documents. This information can include the project's investment budget, technical route, market analysis, etc., and is a key factor in assessing project feasibility and risk. The big data processing system provides strong support for project decision-making and management by identifying and mining these feasibility study index entities.

[0059] In step 110, the big data processing system begins to identify the description document embedding semantic examples of several power grid infrastructure description document examples. These power grid infrastructure description documents are part of the target project text knowledge graph, and contain rich project information and key data. The system uses advanced natural language processing and machine learning technologies to conduct in-depth analysis of these documents and accurately extract the semantic information therein. The key to this step is that the system must not only understand the surface text of the document, but also dig out the deep meaning and contextual relationship behind the text. By converting these semantic information into a computer-understandable format (i.e., embedded semantic examples), the big data processing system lays a solid foundation for subsequent knowledge graph analysis and index entity mining. In addition, when the system performs index entity mining on the target project text knowledge graph based on these power grid infrastructure description document examples, the mining results obtained can clearly reflect the existence of feasibility study review index entities in the knowledge graph. These entities are key elements in project review and management, and may involve core information such as the project's investment scale, technology selection, and construction period. By identifying and extracting these entities, the big data processing system provides strong data support and intelligent analysis for project decision-making in the power industry.

[0060] Step 120: Perform semantic clustering operation on a plurality of the description document embedding semantic examples to obtain at least one document embedding semantic clustering set.

[0061] In the embodiment of the present application, semantic clustering is a data processing technology, which refers to the process of grouping or classifying a set of data with similar semantics using a specific algorithm (such as a clustering algorithm). In the context of power grid infrastructure description document processing, semantic clustering aims to group the description document embedded semantic examples according to their similarity in subject, content or meaning. Through this operation, the system can organize a large number of documents into clusters that are easier to manage and analyze, and each cluster represents a specific semantic field or topic.

[0062] Document embedding semantic clustering refers to a set of documents formed after semantic clustering. In this set, each document has high semantic similarity and they are clustered around one or more central themes or concepts. Each document embedding semantic clustering can be regarded as an independent subset of documents with a common theme, which facilitates subsequent more in-depth analysis, retrieval or mining operations. Such clustering helps to improve the efficiency and accuracy of information processing, especially when processing large-scale document collections.

[0063] In detail, step 120 is a key link in the big data processing process, which involves further organizing and classifying the description document embedding semantic examples extracted in the previous step. In this step, the system performs semantic clustering operations on several description document embedding semantic examples. For example, the big data processing system uses advanced clustering algorithms, such as K-means, hierarchical clustering or DBSCAN, to analyze the similarities and differences between each description document embedding semantic example. Through these algorithms, the system can identify which documents are semantically closer and classify them into the same cluster. The purpose of the above process is to transform the original, disorganized document collection into a structured and meaningful document embedding semantic cluster set. Each cluster set represents a unique semantic field, which can be about a specific power grid infrastructure project stage, technical details, management strategy, etc. This organization method not only facilitates subsequent information retrieval and mining, but also improves the efficiency and accuracy of the knowledge graph analysis network. Finally, after the semantic clustering operation, the big data processing system generates at least one (usually multiple) document embedding semantic cluster sets. These clusters provide a high-quality data foundation for subsequent steps (such as perturbation processing and knowledge graph analysis), thus ensuring the accuracy and effectiveness of the entire processing pipeline.

[0064] Step 130, perturb the first description document embedding semantic examples corresponding to the topic embedding semantics of each of the document embedding semantic clusters to obtain second description document embedding semantic examples corresponding to each of the first description document embedding semantic examples, wherein the second description document embedding semantic examples are used for debugging the knowledge graph analysis network, and the knowledge graph analysis network is used for index entity mining to determine whether there is a feasibility review index entity in the project text knowledge graph in the power grid infrastructure description document to be analyzed.

[0065] In the embodiments of the present application, topic embedding semantics refers to the core semantic information about a topic extracted from text data through natural language processing and machine learning techniques. This information is converted into a numerical vector that can capture the key content and contextual relationships about the topic in the document. In the processing of power grid infrastructure description documents, topic embedding semantics helps the system understand and analyze the content related to a specific topic in the document.

[0066] The first description document embedding semantic examples refer to the description document embedding semantic examples selected from each document embedding semantic cluster set after the semantic clustering operation, which represent the topic embedding semantics of the cluster set. These example documents are highly similar in semantics to other documents in the cluster set they belong to, so they can be used as typical representatives of the cluster set for subsequent analysis and processing.

[0067] Perturbation processing is a data processing technique that tests the stability and robustness of a system by making small, controlled modifications or adding noise to the original data. In the power grid infrastructure description document processing, perturbation processing is applied to the first description document embedding semantic examples to generate new, slightly different second description document embedding semantic examples. These perturbed examples are used to debug and verify the performance of the knowledge graph analysis network.

[0068] The second description document embedding semantic examples are new examples obtained by perturbing the first description document embedding semantic examples. These new examples introduce some minor changes while retaining the original semantic information to simulate various variations and noises that may occur in actual situations. By inputting these second description document embedding semantic examples into the knowledge graph analysis network, the generalization ability and robustness to noise of the network can be tested.

[0069] The knowledge graph analysis network is a deep learning model designed specifically for processing and analyzing data in the project text knowledge graph. In the power grid infrastructure project, the network can perform deep mining on the power grid infrastructure description documents to be analyzed to determine whether there are feasibility review index entities in them. These entities are key information in project review and management, and are of great significance for assessing the feasibility and risk of the project. The knowledge graph analysis network supports project decision-making in the power industry by learning and identifying these entities.

[0070] Further, in step 130, the big data processing system perturbs the first description document embedding semantics examples corresponding to the topic embedding semantics of each document embedding semantic cluster set. The purpose of this step is to generate second description document embedding semantics examples, and these newly generated examples will be used for debugging the knowledge graph analysis network. For example, the system makes slight modifications or adds noise to the first description document embedding semantics examples in each document embedding semantic cluster set to simulate various changes that may occur in actual situations. This perturbation processing helps to test the stability and generalization ability of the knowledge graph analysis network, that is, whether the network can maintain consistent performance when facing slightly different inputs. After perturbation processing, the system generates second description document embedding semantics examples that are similar to the original examples but not exactly the same. These second examples are then input into the knowledge graph analysis network to verify whether the network can accurately identify and mine the feasibility review index entities therein. Through this step, the big data processing system can not only effectively debug and optimize the knowledge graph analysis network, but also ensure that the network has better robustness and accuracy in practical applications. This is of great significance for project management, risk control and decision support in the power industry.

[0071] In combination with the above content, the following is an introduction through a complete application scenario. In the power industry, the review and management of power grid infrastructure projects are crucial links. In order to more effectively identify and mine key project-related information from a large number of standard documents, an advanced big data processing system is introduced into this process. The following is a specific application scenario, which shows how to implement a technical solution through a big data processing system to identify feasibility review index entities in power grid infrastructure description documents. The big data processing system first identifies the description document embedding semantic examples of several power grid infrastructure description document examples. These power grid infrastructure description documents are part of the target project text knowledge graph. The knowledge graph is a graphical data structure used to represent the relationship between entities, which can help better understand and analyze data. In this step, the big data processing system deeply analyzes these documents through natural language processing and machine learning technologies to extract key semantic information. After being systematically processed, these power grid infrastructure description document examples are used to mine index entities in the target project text knowledge graph. The mining results clearly reflect the existence of feasibility review index entities in the target project text knowledge graph, which are the key to further analysis and decision-making. Next, the big data processing system performed semantic clustering on several description document embedding semantic examples. Through the clustering algorithm, the system divided these examples into different clusters according to their semantic similarity, namely, document embedding semantic cluster sets. Each cluster set represents a class of documents with similar topics or contents, which helps to further analyze and process specific topics or contents. After completing the semantic clustering, the big data processing system began to perturb the first description document embedding semantic examples corresponding to the topic embedding semantics of each document embedding semantic cluster set. Perturbation processing is a method of testing the stability of a system by introducing small changes. Here, the system fine-tuned the first description document embedding semantic example to generate the corresponding second description document embedding semantic example. These perturbed examples were then used to debug the knowledge graph analysis network. The knowledge graph analysis network is a specially designed deep learning model that can perform deep mining on the project text knowledge graph in the power grid infrastructure description document to be analyzed to determine whether there are feasibility review index entities.

[0072] Through this series of processing steps, the big data processing system not only successfully identified and mined key feasibility review index entities from the power grid infrastructure description documents, but also further improved the accuracy and stability of the knowledge graph analysis network through semantic clustering and perturbation processing. This is of great significance for project management, risk control and decision support in the power industry.

[0073] The embodiment of the present application can accurately capture the core semantic information in the document by deeply identifying the embedded semantics of the description document in the example of the power grid infrastructure description document, laying a solid foundation for subsequent semantic analysis and entity mining. This recognition process ensures that the index entities extracted from the target project text knowledge graph can truly reflect the key elements of the feasibility study review, thereby greatly improving the accuracy and pertinence of the index entity mining. By performing semantic clustering operations on the description document embedded semantic examples, the embodiment of the present application further realizes the efficient organization and classification of power grid infrastructure description documents. This clustering method not only simplifies the complexity of data processing, but also enables documents with similar semantics to be centrally managed, which facilitates subsequent data analysis and mining.

[0074] In addition, the embodiment of the present application also innovatively introduces perturbation processing technology to fine-tune the first description document embedding semantic examples corresponding to the topic embedding semantics in each document embedding semantic cluster. This step generates second description document embedding semantic examples, which not only retain the core features of the original data, but also introduce minor changes that may occur in actual scenarios, thereby effectively improving the debugging efficiency and accuracy of the knowledge graph analysis network. Finally, these processed second description document embedding semantic examples are used to debug the knowledge graph analysis network, which can accurately determine whether there is a feasibility review index entity in the project text knowledge graph in the power grid infrastructure description document to be analyzed. In this way, not only the performance of the knowledge graph analysis network is optimized, but also strong technical support is provided for the feasibility review of power grid infrastructure projects, significantly improving the review efficiency and accuracy.

[0075] In some possible embodiments, before perturbing the first description document embedding semantic examples corresponding to the topic embedding semantics of each of the document embedding semantic clusters to obtain the second description document embedding semantic examples corresponding to each of the first description document embedding semantic examples, the method further includes: constructing at least one first perturbation semantic feature, and there are differences between each of the first perturbation semantic features.

[0076] Based on this, the first description document embedding semantic examples corresponding to the topic embedding semantics of each of the document embedding semantic clusters are perturbed to obtain second description document embedding semantic examples corresponding to each of the first description document embedding semantic examples, including: implementing the following steps for each of the document embedding semantic clusters respectively: setting the first perturbation semantic features for the first description document embedding semantic examples corresponding to the topic embedding semantics of the document embedding semantic clusters respectively, to obtain at least one second description document embedding semantic example corresponding to the first description document embedding semantic example.

[0077] Based on this embodiment, the big data processing system will construct at least one first perturbation semantic feature before performing perturbation processing on the first description document embedding semantic examples corresponding to the topic embedding semantics of each document embedding semantic cluster set. These first perturbation semantic features are carefully designed to simulate various possible semantic changes to test the robustness and accuracy of the knowledge graph analysis network. There are differences between each first perturbation semantic feature, which ensures that the generated second description document embedding semantic examples are diverse, thereby more comprehensively testing the performance of the network.

[0078] For example, the big data processing system implements a series of steps for each document embedding semantic cluster. First, the system selects the first description document embedding semantic examples corresponding to the theme embedding semantics of the cluster. These examples are representative documents obtained after the document clustering operation, and their semantics are highly related to the theme of the cluster.

[0079] Next, the big data processing system sets the first perturbation semantic features for each of these first description document embedding semantic examples. This step is achieved by adding minor semantic changes to the original examples, such as replacing certain words, adjusting sentence structures, or introducing new information elements. These changes are intended to simulate semantic noise or variation that can occur in actual situations to test the performance of the knowledge graph analysis network when faced with inputs that are not completely consistent or slightly changed.

[0080] By setting these first perturbation semantic features, the big data processing system can generate second description document embedding semantic examples that are similar to but not identical to the original first description document embedding semantic examples. These second examples not only retain the core semantic information of the original data, but also introduce new variation factors, thereby more comprehensively reflecting various situations that may occur in actual scenarios.

[0081] For example, the numerical feature vector of a first description document embedding semantic example is [0.5, 0.3, 0.2], which represents the position of the document in a specific semantic space. The big data processing system can construct a first perturbation semantic feature whose numerical representation is [0.05, -0.03, 0.02], and then apply this perturbation feature to the original example to obtain a new second description document embedding semantic example with a numerical feature vector of [0.55, 0.27, 0.22]. This process is repeated for each first description document embedding semantic example to generate multiple second description document embedding semantic examples. The above perturbed second description document embedding semantic examples are then used to debug and optimize the knowledge graph analysis network. By inputting these examples into the network and observing the output of the network, the big data processing system can evaluate the performance of the network in the face of various semantic changes. This helps to discover potential problems in the network and make corresponding adjustments and optimizations.

[0082] In this way, by introducing the first disturbance semantic features and generating the second description document embedding semantic examples, the big data processing system can more comprehensively test and verify the performance of the knowledge graph analysis network. This method not only improves the robustness and accuracy of the network, but also provides more reliable technical support for the feasibility study review of power grid infrastructure projects. At the same time, the above method also enhances the adaptability and flexibility of the big data processing system in the face of complex and changeable actual scenarios.

[0083] In some preferred embodiments, constructing at least one first disturbance semantic feature includes: performing feature encoding on text knowledge that meets the first thermal condition at least once to obtain at least one feature encoding result, and using each of the feature encoding results as each of the first disturbance semantic features, and there are differences between each of the feature encoding results; or, obtaining text knowledge that meets at least one second thermal condition, and for each of the second thermal conditions, feature encoding the text knowledge that meets the second thermal condition to obtain the first disturbance semantic feature, and there are differences between each of the second thermal conditions.

[0084] Based on this embodiment, the big data processing system will adopt a more sophisticated and targeted method when constructing the first disturbance semantic features. For example, the system can generate these features in two ways.

[0085] The first way is that the big data processing system performs feature encoding on the text knowledge that meets the first thermal condition. The so-called "first thermal condition" can refer to the importance, frequency of occurrence, or relevance of certain text knowledge in the field of power grid infrastructure. The system performs feature encoding on these qualified text knowledge at least once, and each encoding will obtain a feature encoding result, which is used as the first disturbance semantic feature. Since the conditions or methods of each encoding can be slightly different, there will be differences between the feature encoding results obtained, so that the generated disturbance features can be diverse and representative. For example, there is a text description about a power grid infrastructure project: "This project will build a 220kV substation to meet the city's growing electricity demand." This text meets the first thermal condition because it involves the core content of power grid infrastructure. The big data processing system can perform feature encoding on it multiple times. For example, the first encoding focuses on power facilities and capacity information, and obtains feature vectors [0.8, 0.2, 0.0, ...], while the second encoding can focus more on project purpose and needs, and obtain different feature vectors [0.5, 0.3, 0.2, ...]. These two feature vectors can be used as two first perturbation semantic features.

[0086] The second way is that the big data processing system obtains text knowledge that satisfies at least one second thermal condition. Similar to the first thermal condition, the second thermal condition is also defined based on the importance of text knowledge in a specific field or other relevant criteria. However, here, the system encodes features from the text knowledge that satisfies each different second thermal condition to obtain the first disturbance semantic feature. Since there are differences between each second thermal condition, the disturbance features obtained in this way will also be different. Taking the environmental protection requirements in the power grid infrastructure project as an example, this can be regarded as a second thermal condition. The system encodes features from the text knowledge related to environmental protection requirements, such as: "This project will use low-noise transformers and advanced dust removal equipment to ensure compliance with environmental protection standards." This text can be encoded as a specific feature vector, such as [0.3, 0.5, 0.2, ...], where higher values ​​can correspond to the importance of environmental protection equipment and measures.

[0087] Through these two methods, the big data processing system can construct diverse and targeted first perturbation semantic features. These features are then used to perturb the first description document embedding semantic examples to generate second description document embedding semantic examples. These examples not only retain the key information of the original document, but also introduce new change factors, thereby more effectively testing and optimizing the performance of the knowledge graph analysis network. Therefore, by constructing the first perturbation semantic features, the big data processing system can more accurately simulate the semantic changes in actual scenarios, thereby improving the accuracy and robustness of the knowledge graph analysis network. The implementation of this method not only helps the feasibility study review of power grid infrastructure projects to be carried out more efficiently and accurately, but also demonstrates the powerful capabilities and flexibility of the big data processing system in complex text analysis tasks.

[0088] In other exemplary embodiments, before perturbing the first description document embedding semantic examples corresponding to the topic embedding semantics of each of the document embedding semantic clusters to obtain the second description document embedding semantic examples corresponding to each of the first description document embedding semantic examples, the method also includes: constructing a second perturbation semantic feature corresponding to each of the document embedding semantic clusters.

[0089] Then the first description document embedding semantic examples corresponding to the topic embedding semantics of each of the document embedding semantic clusters are perturbed to obtain second description document embedding semantic examples corresponding to each of the first description document embedding semantic examples, including: implementing the following steps for each of the document embedding semantic clusters respectively: for the first description document embedding semantic examples corresponding to the topic embedding semantics of the document embedding semantic clusters, setting the second perturbation semantic feature corresponding to the document embedding semantic cluster set, to obtain the second description document embedding semantic examples corresponding to the first description document embedding semantic examples.

[0090] Based on this embodiment, before performing the perturbation processing, the big data processing system constructs corresponding second perturbation semantic features for each document embedding semantic cluster in addition to constructing the general first perturbation semantic features. These second perturbation semantic features are customized for specific clusters and are intended to more accurately simulate the semantic changes that may occur in documents within the cluster.

[0091] For example, the big data processing system first analyzes the characteristics of each document embedding semantic cluster, including its topic, keyword distribution, sentence structure, etc. Based on these analyses, the system generates one or more second perturbation semantic features for each cluster. These features may include fine-tuning common words in the cluster, changes in sentence structure, or the introduction of new information closely related to the topic of the cluster.

[0092] Taking the "substation construction" cluster in the power grid infrastructure project as an example, its second perturbation semantic feature may involve fine-tuning of specific information such as substation capacity, equipment selection, and construction location. For example, a first description document embedding semantic example about "220kV substation construction" may have a numerical feature vector of [0.6, 0.2, 0.1, 0.1] (the numbers here represent the weights of different semantic dimensions, for illustration only). The second perturbation semantic feature constructed by the system for this cluster may include fine-tuning "220kV" to "110kV" or "330kV", or introducing a description of the specific location of the substation. When this second perturbation semantic feature is applied to the original first description document embedding semantic example, a new second description document embedding semantic example will be obtained, and its numerical feature vector may change due to perturbation, such as becoming [0.5, 0.25, 0.15, 0.1].

[0093] When implementing the perturbation process, the big data processing system applies the corresponding second perturbation semantic features to the first description document embedding semantic examples corresponding to the topic embedding semantics of each document embedding semantic cluster set. In this way, each cluster set will generate a series of second description document embedding semantic examples that are similar to the original examples but not exactly the same.

[0094] In this way, the differences between different clusters are taken into account, so that the generated second description document embedding semantic examples are closer to the semantic changes that may occur in actual scenarios. In this way, the big data processing system can more comprehensively test and verify the performance of the knowledge graph analysis network in the face of various specific scenarios. It can be seen that by constructing the corresponding second perturbation semantic features for each document embedding semantic cluster set and performing perturbation processing, the big data processing system can generate more targeted and diverse second description document embedding semantic examples. These examples not only help to improve the accuracy and robustness of the knowledge graph analysis network, but also make the feasibility review of power grid infrastructure projects more detailed and comprehensive. At the same time, this method also demonstrates the flexibility and accuracy of the big data processing system in processing complex text data.

[0095] In the next step, the second perturbation semantic feature corresponding to each of the document embedded semantic clusters is constructed, including: implementing the following steps for each of the document embedded semantic clusters: determining the semantic thermal conditions of the document embedded semantic clusters; constructing text knowledge that meets the semantic thermal conditions of the document embedded semantic clusters; and obtaining the second perturbation semantic feature corresponding to the document embedded semantic cluster by feature encoding from the text knowledge that meets the semantic thermal conditions of the document embedded semantic clusters.

[0096] Based on this embodiment, the big data processing system further refines the construction process of the perturbation semantic features, especially the second perturbation semantic features for each document embedding semantic cluster set. This step is to ensure that each cluster set can obtain perturbation features that conform to its specific semantic environment, thereby improving the accuracy and pertinence of subsequent analysis.

[0097] For example, for each document embedded semantic cluster, the big data processing system first determines the semantic thermal conditions of the cluster. The "semantic thermal conditions" here can be understood as the key semantic features or themes shared by the documents in the cluster, which are of high importance and representativeness within the cluster. For example, if a cluster mainly discusses "environmental impact assessment of power grid infrastructure projects", then "environmental impact", "assessment methods" and "mitigation measures" may be the semantic thermal conditions of the cluster.

[0098] Next, the system constructs textual knowledge that satisfies these semantic thermal conditions. This usually involves retrieving and screening textual content that is highly relevant to the semantic thermal conditions from existing knowledge bases, document sets, or network resources. Taking "Environmental Impact Assessment of Power Grid Infrastructure Projects" as an example, the system can collect textual materials including laws and regulations on environmental impact assessment, assessment reports, expert opinions, etc.

[0099] Then, the big data processing system performs feature encoding from the text knowledge that meets the semantic thermal conditions to extract the second perturbation semantic features that can represent the specific semantic environment of the cluster. Feature encoding is a complex process that may include a variety of technical means such as word embedding, TF-IDF weight calculation, and topic modeling, aiming to convert text knowledge into numerical feature vectors. These feature vectors can capture key information in the text, such as the frequency of use of words and the semantic structure of the document.

[0100] For example, for the cluster set of “environmental impact assessment of power grid infrastructure projects”, the system can obtain a second perturbation semantic feature vector through feature encoding, such as [0.7, 0.2, 0.1, ...] (the numbers here represent the weights of different semantic dimensions, for illustration only), where higher weights may correspond to core semantics such as “environmental impact” and “assessment standards”.

[0101] Finally, these second perturbation semantic features will be applied to the corresponding first description document embedding semantic examples to generate second description document embedding semantic examples. These examples not only retain the key information of the original document, but also simulate the semantic changes that may occur in actual scenarios by introducing specific perturbation features.

[0102] Through this method, the big data processing system can construct a more accurate and targeted second perturbation semantic feature for each document embedding semantic cluster set. This not only helps to improve the performance and accuracy of the subsequent knowledge graph analysis network, but also makes the feasibility review of power grid infrastructure projects more detailed, professional and comprehensive. At the same time, this method also demonstrates the flexibility and in-depth analysis capabilities of the big data processing system when processing complex text data.

[0103] Under a preferred design idea, after the first description document embedding semantic examples corresponding to the topic embedding semantics of each of the document embedding semantic clusters are perturbed to obtain the second description document embedding semantic examples corresponding to each of the first description document embedding semantic examples, the method also includes: obtaining a derived knowledge graph, which does not include the target project text knowledge graph; refining the derived knowledge graph semantic vector of the derived knowledge graph; implementing one of the following steps: integrating the second description document embedding semantic example and the derived knowledge graph semantic vector to obtain a third description document embedding semantic example; and determining the derived knowledge graph semantic vector as the third description document embedding semantic example.

[0104] Based on this design idea, after completing the perturbation processing of the first description document embedding semantic example corresponding to the topic embedding semantics of each document embedding semantic cluster, the big data processing system will further use the derived knowledge graph to enrich and expand the existing semantic information. This step aims to improve the generalization ability of the knowledge graph analysis network and its ability to process unknown information by introducing external knowledge.

[0105] First, the big data processing system obtains a derived knowledge graph. This derived knowledge graph is different from the target project text knowledge graph. It may contain knowledge points and relationships that are related to the target project but not exactly the same. For example, in the scenario of a power grid infrastructure project, the derived knowledge graph may cover relevant laws and regulations, technical standards, and lessons learned from similar projects in the power industry.

[0106] Next, the system extracts the semantic vector of the derived knowledge graph. This process involves converting the entities, relationships, and attributes in the graph into numerical feature vectors for subsequent mathematical operations and comparative analysis. The extracted semantic vector of the derived knowledge graph captures the key information in the graph and provides a basis for subsequent integration or determination operations.

[0107] The big data processing system then implements one of the following two steps:

[0108] (1) Integrate the second description document embedding semantic example and the derived knowledge graph semantic vector to obtain the third description document embedding semantic example. This step is to combine the two through some fusion technology (such as weighted average, splicing, linear transformation, etc.) to form a more comprehensive and rich semantic representation. Taking the power grid infrastructure project as an example, the second description document embedding semantic example is a vector about substation construction [0.6, 0.2, 0.15, 0.05], while the derived knowledge graph semantic vector provides information about power industry safety standards [0.2, 0.4, 0.3, 0.1]. By integrating these two vectors, a third description document embedding semantic example containing information on both substation construction and safety standards can be obtained;

[0109] (2) Directly determine the derived knowledge graph semantic vector as the third description document embedding semantic example. This method is suitable for those cases where the derived knowledge graph semantic vector itself is comprehensive and representative enough and can be directly used as a new description document embedding semantic example.

[0110] Regardless of the method, the final third description document embedding semantic examples will contain richer semantic information, which will help improve the accuracy and generalization ability of the knowledge graph analysis network. Especially in the face of complex and changeable actual scenarios, this semantic representation method that integrates internal and external knowledge can better cope with unknown and changing situations.

[0111] It can be seen that by introducing the derived knowledge graph and refining its semantic vector, and then integrating or determining the operation with the perturbed second description document embedding semantic example, the big data processing system can generate a more comprehensive and accurate third description document embedding semantic example. This not only helps to improve the performance of the knowledge graph analysis network, but also provides richer and deeper semantic support for application scenarios such as feasibility study review of power grid infrastructure projects.

[0112] In some other embodiments, after the first description document embedded semantic examples corresponding to the topic embedded semantics of each of the document embedded semantic clusters are perturbed to obtain the second description document embedded semantic examples corresponding to each of the first description document embedded semantic examples, the method further includes: refining the associated description document embedded semantic examples of the associated description document examples, and constructing a third perturbation semantic feature; wherein the associated description document examples belong to the target project text knowledge graph, and when index entity mining is performed on the target project text knowledge graph based on the associated description document examples, the obtained index entity mining results reflect that there is no feasibility review index entity in the target project text knowledge graph; based on the associated description document embedded semantic examples and the third perturbation semantic feature, calling the perturbation semantic construction network to construct a fourth description document embedded semantic example; integrating the second description document embedded semantic example and the fourth description document embedded semantic example to obtain a fifth description document embedded semantic example.

[0113] Based on this embodiment, after completing the perturbation processing of the first description document embedding semantic example corresponding to the topic embedding semantics of each document embedding semantic cluster, the big data processing system will perform further operations to deal with the situation where there is no feasibility review index entity in the target project text knowledge graph.

[0114] First, the system extracts the associated description document embedding semantic examples of the associated description document examples. These associated description document examples belong to the target project text knowledge graph, and when the index entity mining is performed on them, the results show that there is no feasibility review index entity in the target project text knowledge graph. In other words, these documents are related to the project to be reviewed, but effective review information cannot be mined directly from them.

[0115] Next, the big data processing system constructs a third perturbation semantic feature. This feature is designed to simulate and introduce some minor semantic changes in order to mine more hidden information from the semantic examples of the associated description document embedding. Similar to the previous perturbation semantic feature, the third perturbation semantic feature is also generated based on a deep understanding of the project text and context.

[0116] Then, the system calls the perturbation semantic construction network to construct the fourth description document embedding semantic example based on the associated description document embedding semantic example and the third perturbation semantic feature. This process can be seen as a "perturbation version" of the original associated description document, which aims to reveal information that may have been overlooked by introducing minor semantic changes.

[0117] Finally, the big data processing system integrates the second description document embedding semantics example (the example after preliminary perturbation) and the fourth description document embedding semantics example (the example after associated description document perturbation), thereby obtaining a more comprehensive and rich fifth description document embedding semantics example. This integrated example not only contains the information of the original document, but also incorporates the hidden information mined from the associated description documents, providing a more detailed data foundation for subsequent knowledge graph analysis and feasibility study review.

[0118] For example, a feasibility study review is being conducted on a power grid infrastructure project. After preliminary perturbation processing, the second description document embedding semantic example is obtained, which mainly reflects basic information such as the scale, location and expected benefits of the project. However, when trying to mine more information from the associated description documents, it is found that although these documents are related to the project, they do not directly mention the key information of the feasibility study review. At this time, by constructing the third perturbation semantic feature and calling the perturbation semantic construction network to generate the fourth description document embedding semantic example, it may be possible to reveal certain risk points or optimization suggestions that were originally hidden in the associated description documents. Finally, integrating this information into the fifth description document embedding semantic example can provide more comprehensive and in-depth support for the feasibility study review of the project.

[0119] In this way, through this method, the big data processing system can still construct a semantic example of a description document with rich information when there is no direct feasibility review index entity in the target project text knowledge graph, providing strong support for subsequent analysis and review work. This not only improves the flexibility and adaptability of the system, but also further enhances its practical value in complex text data processing.

[0120] In some optional embodiments, the method also includes: obtaining a basic perturbation semantic construction network and a first associated description document embedded semantic example of the first associated description document example, and constructing a fourth perturbation semantic feature; based on the first associated description document embedded semantic example and the fourth perturbation semantic feature, calling the basic perturbation semantic construction network to construct a sixth description document embedded semantic example; calling a document semantic processing network to parse the sixth description document embedded semantic example to obtain a document semantic parsing viewpoint, wherein the document semantic parsing viewpoint indicates that the sixth description document embedded semantic example is a confidence weight of a priori description document embedded semantic example; based on the priori description document embedded semantic example and the document semantic parsing viewpoint, determining an index value of the training error of the basic perturbation semantic construction network; based on the index value of the training error, debugging the basic perturbation semantic construction network to obtain the perturbation semantic construction network.

[0121] In this embodiment, the big data processing system takes a series of steps to train and optimize the perturbation semantic construction network to improve its accuracy and effectiveness in generating examples that describe document embedding semantics.

[0122] First, the system obtains a basic perturbation semantic construction network, which is an untrained initial model for subsequent debugging and optimization. At the same time, the system also obtains the first associated description document embedding semantic example of the first associated description document example and constructs a fourth perturbation semantic feature. This fourth perturbation semantic feature is the result of a slight perturbation of the first associated description document embedding semantic example, and is used to simulate the semantic changes that may be encountered in actual applications.

[0123] Next, the system calls the basic perturbation semantic construction network to construct the sixth description document embedding semantic example based on the first associated description document embedding semantic example and the fourth perturbation semantic feature. This process can be regarded as a forward propagation of the network, and by inputting specific semantic features and examples, the description document embedding semantic example output by the network is obtained.

[0124] Then, the big data processing system calls a document semantic processing network to parse the sixth description document embedding semantic example. This document semantic processing network is a pre-trained model that can deeply parse the input document embedding semantic example and output a document semantic parsing opinion. This opinion actually represents the confidence weight between the sixth description document embedding semantic example and the prior description document embedding semantic example, that is, the similarity or matching degree between the two.

[0125] After obtaining the viewpoint of document semantic parsing, the system determines the index value of the training error of the basic perturbation semantic construction network based on this viewpoint and the prior description document embedding semantic examples. This index value reflects the accuracy and performance of the network in generating description document embedding semantic examples, and is an important basis for subsequent network debugging and optimization.

[0126] Finally, according to the index value of the training error, the big data processing system debugs the basic perturbation semantic construction network. This process may include adjusting the network parameters, optimizing the network structure, improving the training algorithm, etc., with the aim of reducing the training error and improving the performance and accuracy of the network. After multiple iterations and optimizations, the system obtains a trained perturbation semantic construction network with better performance.

[0127] Let's take an example to illustrate this process: there is a description document about a power grid infrastructure project. First, a sixth description document embedding semantic example is generated through the basic perturbation semantic construction network. Then, the document semantic processing network is used to parse this example to obtain a document semantic parsing view that represents the confidence weight between it and the prior description document embedding semantic example. If it is found that this confidence weight is low, it means that the generated example is very different from the prior example, and it is necessary to adjust the parameters and structure of the basic perturbation semantic construction network to reduce this difference. After multiple debugging and optimization, a perturbation semantic construction network that can more accurately generate description document embedding semantic examples can be obtained.

[0128] In this way, through this method, the big data processing system can continuously optimize and improve the perturbation semantic construction network, so that it can more accurately generate examples of document embedding semantics in practical applications. This not only improves the performance and accuracy of the system, but also provides a more reliable data foundation for subsequent knowledge graph analysis and feasibility study review.

[0129] It is worth mentioning that, based on the prior description document embedding semantics example and the document semantic parsing viewpoint, the indicator value of the training error of the basic perturbation semantic construction network is determined, including:

[0130] Calculate the semantic difference between the prior description document embedded semantic example and the sixth description document embedded semantic example; the semantic difference can be obtained by calculating the cosine distance, Euclidean distance or other similarity measurement method between the two embedded semantic examples, and the difference reflects the degree of dissimilarity between the generated description document embedded semantic example and the prior example;

[0131] The document semantic analysis viewpoint is converted into a weight coefficient, where the weight coefficient represents the confidence between the sixth description document embedding semantic example and the prior example; the weight coefficient can be mapped or converted according to the specific value of the document semantic analysis viewpoint, for example, the analysis viewpoint is converted into a probability distribution through a softmax function, and the corresponding probability value is selected as the weight coefficient;

[0132] The semantic difference is weighted by the weight coefficient to obtain a weighted semantic difference; by multiplying the weight coefficient, more confident description document embedded semantic examples can occupy a larger proportion in the training error calculation, thereby improving the accuracy and pertinence of the training error calculation;

[0133] The index value of the training error is determined based on the weighted semantic difference; the weighted semantic difference can be directly used as the index value of the training error, or the final index value is obtained through further processing (such as averaging, normalization, etc.), which is used to evaluate the performance of the basic perturbation semantic construction network and guide the network debugging and optimization process.

[0134] Through the above technical solution, the training error of the basic perturbation semantic construction network can be quantified more accurately, providing strong support for network optimization. At the same time, the introduction of the concepts of weight coefficient and weighted semantic difference makes the calculation of training error more in line with the needs of actual application scenarios, improving the performance of big data processing systems in semantic analysis and knowledge graph construction.

[0135] In some possible embodiments, the integrating the second description document embedded semantic example and the fourth description document embedded semantic example to obtain the fifth description document embedded semantic example includes: obtaining a first feature enhancement coefficient of the second description document embedded semantic example, and obtaining a second feature enhancement coefficient of the fourth description document embedded semantic example; and semantically enhancing the second description document embedded semantic example and the fourth description document embedded semantic example based on the first feature enhancement coefficient and the second feature enhancement coefficient to obtain the fifth description document embedded semantic example.

[0136] Based on this embodiment, when the big data processing system integrates the second description document embedded semantic example and the fourth description document embedded semantic example to obtain the fifth description document embedded semantic example, it adopts a specific method, namely the feature enhancement coefficient method, to ensure that the integrated embedded semantic example not only retains the original information but also is effectively enhanced.

[0137] First, the system obtains the first feature reinforcement coefficient of the second description document embedded semantic example. This coefficient is a numerical value that represents the importance or weight of the second description document embedded semantic example in the integration process. Similarly, the system also obtains a second feature reinforcement coefficient for the fourth description document embedded semantic example. These two coefficients may be calculated by an algorithm or manually set based on experience or actual needs.

[0138] Next, the big data processing system semantically enhances the second description document embedding semantic example and the fourth description document embedding semantic example according to the two feature enhancement coefficients. Specifically, the system may multiply the two description document embedding semantic examples with the corresponding feature enhancement coefficients to adjust their numerical values ​​to achieve semantic enhancement. This enhancement operation can make important features more prominent while suppressing unimportant features.

[0139] Finally, the system integrates the semantically enhanced second description document embedding semantic example and the fourth description document embedding semantic example to obtain the fifth description document embedding semantic example. This integration process may be a simple weighted average or a more complex fusion algorithm, depending on the system design and actual needs.

[0140] For example, there is a second description document embedded semantic example A and a fourth description document embedded semantic example B of a power grid infrastructure project, whose feature vectors are [0.5, 0.3, 0.2] and [0.4, 0.4, 0.2] respectively, and the corresponding first feature enhancement coefficient is 1.2 and the second feature enhancement coefficient is 0.8. When performing semantic enhancement, multiply A by 1.2 to obtain the enhanced A' [0.6, 0.36, 0.24]; multiply B by 0.8 to obtain the enhanced B' [0.32, 0.32, 0.16]. Then, A' and B' are integrated (for example, taking the average value) to obtain the feature vector of the fifth description document embedded semantic example C [0.46, 0.34, 0.2].

[0141] In this way, the big data processing system can more effectively integrate the embedded semantic examples of description documents from different sources, while ensuring that important features are properly strengthened during the integration process. This not only improves the quality of the embedded semantic examples after integration, but also provides a more accurate and comprehensive data basis for subsequent knowledge graph analysis and feasibility study review. In this way, by introducing the method of feature reinforcement coefficient, the big data processing system can more finely control the weights and contributions of different description document embedded semantic examples in the integration process, thereby obtaining a more representative and discriminating fifth description document embedded semantic example. The implementation of this method not only improves the flexibility and accuracy of the system, but also further enhances its practical value in complex text data processing.

[0142] In other possible embodiments, after the first description document embedded semantic examples corresponding to the topic embedded semantics of each of the document embedded semantic clusters are perturbed to obtain the second description document embedded semantic examples corresponding to each of the first description document embedded semantic examples, the method further includes: performing joint discrimination on several of the second description document embedded semantic examples respectively to obtain a joint discrimination result of each of the second description document embedded semantic examples, wherein the joint discrimination result represents a commonality measurement weight between the second description document embedded semantic example and the prior description document embedded semantic example; based on the joint discrimination result of each of the second description document embedded semantic examples, selecting a target description document embedded semantic example whose joint discrimination result meets the joint discrimination requirements from several of the second description document embedded semantic examples.

[0143] In other possible embodiments, after the perturbation processing, the big data processing system will further perform joint discrimination on each second description document embedding semantic example to ensure that the most representative target description document embedding semantic example is selected.

[0144] Specifically, the system first performs joint discrimination on several second description document embedding semantic examples. This joint discrimination process is implemented through a specific algorithm that can comprehensively consider the similarities and differences between multiple description document embedding semantic examples to give a joint discrimination result. This joint discrimination result actually represents the commonality measurement weight between the second description document embedding semantic example and the prior description document embedding semantic example, that is, it measures the degree of similarity between these examples in terms of subject or semantics.

[0145] For example, there are three second description document embedding semantic examples, denoted as S1, S2, and S3. Through the joint discriminant algorithm, the commonality measurement weights between them and the prior description document embedding semantic examples can be obtained, such as 0.8, 0.6, and 0.9 respectively. These weight values ​​reflect the semantic similarity between each example and the prior example.

[0146] Next, based on these joint discrimination results, the big data processing system will select the target description document embedding semantic examples that meet the joint discrimination requirements from the multiple second description document embedding semantic examples. This selection process can be sorting according to the weight value and then selecting the example with the highest weight as the target example; or it can be setting a threshold and selecting the example with a weight exceeding the threshold as the target example.

[0147] Continuing with the above example, if the example with the highest weight is selected as the target example, then S3 (with a weight of 0.9) will be selected as the target description document embedding semantic example.

[0148] By introducing the joint discrimination step, the big data processing system can more accurately evaluate the similarity between each second description document embedded semantic example and the prior example, and select the most representative target example. This helps to improve the accuracy and efficiency of subsequent processing, especially in scenarios that require precise semantic understanding, such as building knowledge graphs and conducting feasibility studies. In this way, the joint discrimination and selection method enables the big data processing system to more accurately identify key information when processing large amounts of text data, improving the overall performance and practical value of the system.

[0149] In an alternative embodiment, after performing a semantic clustering operation on several of the descriptive document embedding semantic examples to obtain at least one document embedding semantic cluster set, the method further includes: determining a first document embedding semantic cluster set from the at least one document embedding semantic cluster set, and determining feature differences between each second document embedding semantic cluster set and the first document embedding semantic cluster set, wherein the second document embedding semantic cluster set is a document embedding semantic cluster set in the at least one document embedding semantic cluster set that is different from the first document embedding semantic cluster set; determining a third document embedding semantic cluster set whose feature difference is higher than a feature difference threshold, and a fourth document embedding semantic cluster set whose feature difference is not higher than the feature difference threshold, from the at least one document embedding semantic cluster set.

[0150] Then the perturbation processing is performed on the first description document embedding semantic examples corresponding to the topic embedding semantics of each of the document embedding semantic clusters to obtain the second description document embedding semantic examples corresponding to each of the first description document embedding semantic examples, including: perturbation processing is performed on the first description document embedding semantic examples corresponding to the topic embedding semantics of each of the third document embedding semantic clusters to obtain the second description document embedding semantic examples corresponding to each of the third document embedding semantic clusters; perturbation processing is performed on the first description document embedding semantic examples corresponding to the topic embedding semantics of each of the fourth document embedding semantic clusters to obtain the second description document embedding semantic examples corresponding to each of the fourth document embedding semantic clusters; wherein the number of the second description document embedding semantic examples corresponding to each of the third document embedding semantic clusters is less than the number of the second description document embedding semantic examples corresponding to each of the fourth document embedding semantic clusters.

[0151] In an alternative embodiment, after executing the semantic clustering operation and obtaining at least one document embedding semantic clustering set, the big data processing system will further perform detailed analysis and processing on these clustering sets.

[0152] First, the system selects one of these document embedding semantic clusters as the first document embedding semantic cluster. This selection can be based on a specific criterion, such as the size of the cluster, internal consistency, or difference from other clusters. Once the first document embedding semantic cluster is determined, the system calculates the feature differences between the other document embedding semantic clusters (here referred to as the second document embedding semantic cluster) and the first document embedding semantic cluster.

[0153] Feature difference is a quantitative indicator used to measure the difference in semantic features between two clusters. This difference can be calculated by comparing the average embedding vectors, word frequency distribution, topic model output, etc. of two clusters.

[0154] Next, the system will classify all the second document embedding semantic clusters into two categories based on the calculated feature differences: third document embedding semantic clusters and fourth document embedding semantic clusters. The third document embedding semantic clusters are clusters whose feature differences are higher than the preset feature difference threshold, while the fourth document embedding semantic clusters are clusters whose feature differences are not higher than this threshold. The above feature difference threshold is an important parameter, which can be set according to specific application scenarios and requirements. For example, in a scenario that requires a high degree of distinction between different semantic groups, this threshold may be set relatively low so that even clusters with only slight differences can be distinguished.

[0155] After completing the above classification, the big data processing system will perform perturbation processing on the first description document embedding semantic examples corresponding to the topic embedding semantics of the third document embedding semantic cluster set and the fourth document embedding semantic cluster set. The purpose of perturbation processing is to increase the diversity and generalization ability of the data, which usually includes adding noise, performing minor transformations, and other operations.

[0156] Importantly, since the features of the third document embedding semantic clusters are quite different from those of the first document embedding semantic clusters, the system generates relatively few examples when generating second description document embedding semantic examples for these clusters. This is because these clusters are relatively independent in semantics and do not require too much perturbation to enhance their distinguishability.

[0157] On the contrary, for the fourth document embedding semantic cluster set, since the feature differences between them and the first document embedding semantic cluster set are small, the system needs to generate more second description document embedding semantic examples to increase the diversity and coverage of the data. This helps the model better learn and understand these similar but slightly different semantic cluster sets.

[0158] In this way, through this feature difference-based classification and perturbation processing method, the big data processing system can more effectively process and analyze a large amount of document embedding semantic clustering data. This not only improves the efficiency and accuracy of data processing, but also provides a higher quality data foundation for subsequent knowledge graph construction, semantic understanding and other tasks.

[0159] In some independent embodiments, after mining the feasibility study review index entities of the project text knowledge graph in the power grid infrastructure description document to be analyzed through the knowledge graph analysis network, the method also includes: generating a query structure tree for the power grid infrastructure description document to be analyzed based on the feasibility study review index entities of the project text knowledge graph in the power grid infrastructure description document to be analyzed.

[0160] In detail, after successfully mining the feasibility study review index entities of the project text knowledge graph in the power grid infrastructure description document to be analyzed through the knowledge graph analysis network, the big data processing system will further use these entities to construct a query structure tree. This query structure tree is specially customized for the power grid infrastructure description document to be analyzed, which can greatly improve the efficiency of information retrieval and data analysis.

[0161] Specifically, the generation process of the query structure tree is as follows: First, the big data processing system will identify and extract all the feasibility study review index entities that have been mined. These entities may include key information such as project name, project location, construction scale, investment budget, construction period, etc. Each entity will become a node in the query structure tree. Then, the system will analyze the relationship between these entities, such as which entities are parent-child relationships and which entities are sibling relationships. These relationships will determine the hierarchical structure of the query structure tree and the connection method between nodes. Taking the specific power grid infrastructure project as an example, the feasibility study review index entities such as "XX substation expansion project", "XX city", "200 million investment budget", and "2-year construction period" have been mined. Then, when generating the query structure tree, "XX substation expansion project" may be set as the root node, and "XX city", "200 million investment budget" and "2-year construction period" are respectively used as its child nodes, forming a hierarchical tree structure. This query structure tree not only intuitively displays the key elements and relationships of the power grid infrastructure project, but also provides users with a convenient query and analysis tool. Users can quickly obtain detailed information and other associated data related to different nodes in the tree by clicking on the node.

[0162] In addition, the big data processing system can also dynamically adjust and optimize the query structure tree according to user needs. For example, when a user is interested in a specific aspect of information, the system can promote the entity in that aspect to a higher level so that the user can access the data more quickly.

[0163] In this way, by generating a query structure tree for the power grid infrastructure description document to be analyzed, the big data processing system not only improves the efficiency of information retrieval, but also provides users with a more intuitive and flexible data analysis tool. This helps users better understand and analyze various aspects of power grid infrastructure projects, thereby making more informed decisions.

[0164] It is worth mentioning that based on the feasibility study review index entity of the project text knowledge graph in the power grid infrastructure description document to be analyzed, a query structure tree for the power grid infrastructure description document to be analyzed is generated, further including:

[0165] Entity-relationship mapping steps: First, the power grid infrastructure description document to be analyzed is deeply parsed to extract the feasibility study review index entities in the project text knowledge graph, and these entities are mapped with their relationships in the document to form entity-relationship pairs;

[0166] Hierarchical structure determination step: Next, the hierarchical position of each entity in the query structure tree is determined according to the relationship attributes in the entity relationship pair, wherein the relationship attributes include parent-child relationship, parallel relationship, attribute relationship, etc., to ensure that the generated query structure tree can clearly reflect the logical relationship between entities;

[0167] Node weight assignment step: After determining the hierarchical structure of the query structure tree, assign weights to each node. The weight assignment is based on factors such as the importance of the entity in the document, the frequency of occurrence, and the degree of association with other entities, so that in the subsequent query process, the query results can be sorted and optimized according to the weights;

[0168] Dynamic adjustment and optimization steps: Dynamically adjust and optimize the query structure tree according to user query requirements, including but not limited to adding and removing nodes, adjusting hierarchical positions, and redistributing weights, etc., to ensure that the query structure tree always matches the actual needs of users;

[0169] Visual display step: Finally, the generated query structure tree is displayed to the user in a visual form, including but not limited to tree diagrams, network diagrams, etc., so that the user can intuitively understand the overall structure of the power grid infrastructure project and the relationship between various entities, thereby conducting more efficient and accurate query and analysis.

[0170] Through the above technical solution, a query structure tree can be generated that not only conforms to the actual characteristics of the power grid infrastructure project, but also meets the personalized query needs of users, further enhancing the application value of the big data processing system in the field of power grid infrastructure.

[0171] In the big data processing system, the system performs a series of refined processing steps for the power grid infrastructure description documents to be analyzed to generate a highly customized query structure tree. This process integrates cutting-edge technologies in multiple fields such as natural language processing, knowledge graph technology, and data visualization, providing strong support for information query and analysis of power grid infrastructure projects.

[0172] First, in the entity relationship mapping step, the big data processing system conducts an in-depth analysis of the power grid infrastructure description document to be analyzed. Through natural language processing technology, the system can accurately identify and extract the feasibility review index entities of the project text knowledge graph in the document. These entities may include key information such as project name, geographical location, project scale, investment budget, construction time, etc. The system then maps these entities to their relationships in the document to form a clear entity relationship pair. For example, the system may recognize that a "substation expansion project" is a power grid infrastructure project in a "certain city", thereby establishing a geographical location relationship between the two entities.

[0173] Next, in the hierarchical structure determination step, the big data processing system determines the hierarchical position of each entity in the query structure tree based on the relationship attributes in the entity relationship pair, such as parent-child relationship, parallel relationship, attribute relationship, etc. This step ensures that the generated query structure tree can clearly reflect the logical relationship between entities. Taking the power grid infrastructure project as an example, "a substation expansion project" may be positioned as the root node of the query structure tree, while entities such as "a city", "project scale", and "investment budget" are respectively its child nodes, forming a hierarchical tree structure.

[0174] In the node weight assignment step, the system assigns weights to each node based on factors such as the entity's importance in the document, frequency of occurrence, and degree of association with other entities. This step is intended to provide a basis for sorting and optimizing query results based on weights in subsequent query processes. For example, the entity "investment budget" may be assigned a higher weight due to its importance in project decision-making.

[0175] The dynamic adjustment and optimization steps are the embodiment of the flexibility and intelligence of the big data processing system. According to the user's query requirements, the system can dynamically adjust and optimize the query structure tree. This includes but is not limited to the addition and removal of nodes, the adjustment of hierarchical positions, and the redistribution of weights. For example, when a user is interested in the geographic location information of a project, the system can promote the nodes related to the geographic location to a higher level so that the user can access this data more quickly.

[0176] Finally, in the visualization step, the big data processing system displays the generated query structure tree to the user in a visualized form. This includes but is not limited to intuitive and easy-to-understand chart forms such as tree diagrams and network diagrams. Through these visualization tools, users can understand the overall structure of the power grid infrastructure project and the relationship between various entities at a glance, so as to conduct more efficient and accurate query and analysis.

[0177] In this way, the big data processing system can generate a query structure tree that not only meets the actual characteristics of power grid infrastructure projects, but also meets the personalized query needs of users. This not only improves the efficiency and accuracy of information query, but also provides users with an intuitive and flexible data analysis tool, further enhancing the application value of the big data processing system in the field of power grid infrastructure.

[0178] Further, Figure 2 Schematic diagram of a big data processing system 200 provided in an embodiment of the present application. Figure 2 The big data processing system 200 shown includes a processor 210, which can call and run a computer program from a memory to implement the method in the embodiment of the present application.

[0179] Alternatively, if Figure 2 As shown, the big data processing system 200 may further include a memory 230. The processor 210 may call and run a computer program from the memory 230 to implement the method in the embodiment of the present application.

[0180] The memory 230 may be a separate device independent of the processor 210 , or may be integrated into the processor 210 .

[0181] Alternatively, if Figure 2 As shown, the big data processing system 200 may further include a transceiver 220, and the processor 210 may control the transceiver 220 to interact with other devices, specifically, may send information or data to other devices, or receive information or data sent by other devices.

[0182] Optionally, the big data processing system 200 can implement the corresponding processes corresponding to the storage engine or components in the storage engine (such as a processing module) or a device deployed with a storage engine in each method of the embodiments of the present application. For the sake of brevity, they are not repeated here.

[0183] It should be understood that the processor of the embodiment of the present application may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method embodiment can be completed by an integrated logic circuit of hardware in the processor or an instruction in the form of software. The above processor may be a general-purpose processor, a digital signal processor (Digital Signal Processor, DSP), an application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), a field programmable gate array (Field Programmable Gate Array, FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present application can be implemented or executed. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The steps of the method disclosed in the embodiment of the present application can be directly embodied as a hardware decoding processor to be executed, or the hardware and software modules in the decoding processor are combined and executed. The software module may be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.

[0184] It can be understood that the memory in the embodiments of the present application can be a volatile memory or a non-volatile memory, or can include both volatile and non-volatile memories. Among them, the non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory can be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct RAM bus random access memory (DR RAM). It should be noted that the memory of the systems and methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0185] It should be understood that the above-mentioned memory is exemplary but not restrictive. For example, the memory in the embodiments of the present application may also be static random access memory (static RAM, SRAM), dynamic random access memory (dynamic RAM, DRAM), synchronous dynamic random access memory (synchronous DRAM, SDRAM), double data rate synchronous dynamic random access memory (double data rate SDRAM, DDR SDRAM), enhanced synchronous dynamic random access memory (enhanced SDRAM, ESDRAM), synchronous link dynamic random access memory (synch link DRAM, SLDRAM) and direct memory bus random access memory (Direct Rambus RAM, DR RAM), etc. That is to say, the memory in the embodiments of the present application is intended to include but not limited to these and any other suitable types of memory.

[0186] Based on the above, a computer-readable storage medium is provided, on which a computer program is stored, and the computer program implements the above method when running.

[0187] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations.

Claims

1. A standard-based power grid infrastructure feasibility study review auxiliary method, characterized in that: Applied to a big data processing system, the method comprises: Identify description document embedding semantic examples of each of a number of power grid infrastructure description document examples; wherein the power grid infrastructure description document examples belong to a target project text knowledge graph, and when index entity mining is performed on the target project text knowledge graph based on the power grid infrastructure description document examples, the obtained index entity mining results reflect that the target project text knowledge graph has a feasibility study review index entity; Performing a semantic clustering operation on a plurality of the description document embedding semantic examples to obtain at least one document embedding semantic clustering set; Performing perturbation processing on the first description document embedding semantic examples corresponding to the topic embedding semantics of each of the document embedding semantic clusters to obtain second description document embedding semantic examples corresponding to each of the first description document embedding semantic examples, wherein the second description document embedding semantic examples are used for debugging the knowledge graph analysis network, and the knowledge graph analysis network is used for index entity mining to determine whether there is a feasibility study review index entity in the project text knowledge graph in the power grid infrastructure description document to be analyzed; Before perturbing the first description document embedding semantic examples corresponding to the topic embedding semantics of each of the document embedding semantic clusters to obtain the second description document embedding semantic examples corresponding to each of the first description document embedding semantic examples, the method further includes: constructing at least one first perturbation semantic feature, and there are differences between the first perturbation semantic features; The perturbation processing is performed on the first description document embedding semantic examples corresponding to the topic embedding semantics of each of the document embedding semantic clusters to obtain the second description document embedding semantic examples corresponding to each of the first description document embedding semantic examples, including: implementing the following steps for each of the document embedding semantic clusters respectively: setting the first perturbation semantic features for the first description document embedding semantic examples corresponding to the topic embedding semantics of the document embedding semantic clusters respectively, to obtain at least one second description document embedding semantic example corresponding to the first description document embedding semantic example.

2. The method according to claim 1, characterized in that The constructing at least one first disturbance semantic feature comprises: Performing feature encoding at least once on the text knowledge satisfying the first thermal condition to obtain at least one feature encoding result, and using each of the feature encoding results as each of the first disturbance semantic features, wherein there are differences between the feature encoding results; Alternatively, text knowledge satisfying at least one second thermal condition is obtained, and for each of the second thermal conditions, the first disturbance semantic feature is obtained by feature encoding from the text knowledge satisfying the second thermal condition, and there are differences between the second thermal conditions.

3. The method according to claim 1, characterized in that Before perturbing the first description document embedding semantic examples corresponding to the topic embedding semantics of each of the document embedding semantic clusters to obtain the second description document embedding semantic examples corresponding to each of the first description document embedding semantic examples, the method further includes: constructing second perturbation semantic features corresponding to each of the document embedding semantic clusters; The perturbation processing is performed on the first description document embedding semantic examples corresponding to the topic embedding semantics of each of the document embedding semantic clusters to obtain the second description document embedding semantic examples corresponding to each of the first description document embedding semantic examples, including: implementing the following steps for each of the document embedding semantic clusters respectively: for the first description document embedding semantic examples corresponding to the topic embedding semantics of the document embedding semantic clusters, setting the second perturbation semantic features corresponding to the document embedding semantic clusters to obtain the second description document embedding semantic examples corresponding to the first description document embedding semantic examples; Among them, the construction of the second perturbation semantic feature corresponding to each of the document embedding semantic clusters includes: implementing the following steps for each of the document embedding semantic clusters: determining the semantic thermal conditions of the document embedding semantic clusters; constructing text knowledge that meets the semantic thermal conditions of the document embedding semantic clusters; and obtaining the second perturbation semantic feature corresponding to the document embedding semantic cluster by feature encoding from the text knowledge that meets the semantic thermal conditions of the document embedding semantic clusters.

4. The method according to claim 1, characterized in that After perturbing the first description document embedding semantic examples corresponding to the topic embedding semantics of each of the document embedding semantic clusters to obtain the second description document embedding semantic examples corresponding to each of the first description document embedding semantic examples, the method further comprises: Obtaining a derived knowledge graph, wherein the derived knowledge graph does not include the target project text knowledge graph; Extracting a derived knowledge graph semantic vector of the derived knowledge graph; Do one of the following: Integrate the second description document embedding semantic example and the derived knowledge graph semantic vector to obtain a third description document embedding semantic example; The derived knowledge graph semantic vector is determined as the third description document embedding semantic example.

5. The method according to claim 1, characterized in that After perturbing the first description document embedding semantic examples corresponding to the topic embedding semantics of each of the document embedding semantic clusters to obtain the second description document embedding semantic examples corresponding to each of the first description document embedding semantic examples, the method further comprises: Extracting an associated description document embedding semantic example of an associated description document example, and constructing a third perturbation semantic feature; wherein the associated description document example belongs to a target project text knowledge graph, and when index entity mining is performed on the target project text knowledge graph based on the associated description document example, the obtained index entity mining result reflects that there is no feasibility review index entity in the target project text knowledge graph; According to the associated description document embedding semantic example and the third perturbation semantic feature, calling the perturbation semantic construction network to construct a fourth description document embedding semantic example; Integrate the second description document embedded semantic example and the fourth description document embedded semantic example to obtain a fifth description document embedded semantic example; Among them, the integrating the second description document embedded semantic example and the fourth description document embedded semantic example to obtain the fifth description document embedded semantic example includes: obtaining the first feature enhancement coefficient of the second description document embedded semantic example, and obtaining the second feature enhancement coefficient of the fourth description document embedded semantic example; according to the first feature enhancement coefficient and the second feature enhancement coefficient, semantically enhancing the second description document embedded semantic example and the fourth description document embedded semantic example to obtain the fifth description document embedded semantic example.

6. The method according to claim 5, characterized in that The method further comprises: Obtaining a basic perturbation semantic construction network and a first associated description document embedding semantic example of the first associated description document example, and constructing a fourth perturbation semantic feature; According to the first associated description document embedding semantic example and the fourth perturbation semantic feature, calling the basic perturbation semantic construction network to construct a sixth description document embedding semantic example; Calling a document semantic processing network to parse the sixth description document embedding semantic example to obtain a document semantic parsing viewpoint, wherein the document semantic parsing viewpoint indicates a confidence weight that the sixth description document embedding semantic example is a priori description document embedding semantic example; Determining an index value of a training error of the basic perturbation semantic construction network based on a priori description document embedding semantic examples and the document semantic parsing viewpoint; According to the indicator value of the training error, the basic perturbation semantic construction network is debugged to obtain the perturbation semantic construction network.

7. The method according to claim 1, characterized in that After perturbing the first description document embedding semantic examples corresponding to the topic embedding semantics of each of the document embedding semantic clusters to obtain the second description document embedding semantic examples corresponding to each of the first description document embedding semantic examples, the method further comprises: Performing joint discrimination on a plurality of second description document embedded semantic examples respectively to obtain joint discrimination results of each second description document embedded semantic example, wherein the joint discrimination results represent the commonality measurement weight between the second description document embedded semantic example and the prior description document embedded semantic example; Based on the joint discrimination results of each of the second description document embedded semantic examples, a target description document embedded semantic example whose joint discrimination result meets the joint discrimination requirement is selected from a plurality of the second description document embedded semantic examples.

8. The method according to claim 1, characterized in that After performing semantic clustering operation on several of the description document embedding semantic examples to obtain at least one document embedding semantic cluster set, the method further includes: determining a first document embedding semantic cluster set from the at least one document embedding semantic cluster set, and determining feature differences between each second document embedding semantic cluster set and the first document embedding semantic cluster set, wherein the second document embedding semantic cluster set is a document embedding semantic cluster set in the at least one document embedding semantic cluster set that is different from the first document embedding semantic cluster set; determining a third document embedding semantic cluster set whose feature difference is higher than a feature difference threshold, and a fourth document embedding semantic cluster set whose feature difference is not higher than the feature difference threshold, from the at least one document embedding semantic cluster set; The perturbation processing of the first description document embedding semantic examples corresponding to the topic embedding semantics of each of the document embedding semantic clusters to obtain the second description document embedding semantic examples corresponding to each of the first description document embedding semantic examples includes: perturbation processing of the first description document embedding semantic examples corresponding to the topic embedding semantics of each of the third document embedding semantic clusters to obtain the second description document embedding semantic examples corresponding to each of the third document embedding semantic clusters; perturbation processing of the first description document embedding semantic examples corresponding to the topic embedding semantics of each of the fourth document embedding semantic clusters to obtain the second description document embedding semantic examples corresponding to each of the fourth document embedding semantic clusters; wherein the number of the second description document embedding semantic examples corresponding to each of the third document embedding semantic clusters is less than the number of the second description document embedding semantic examples corresponding to each of the fourth document embedding semantic clusters.

9. A big data processing system, characterized in that: The method comprises at least one processor and a memory; the memory stores computer-executable instructions; the at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor executes the method according to any one of claims 1 to 8.

Images