Knowledge graph-based evaluation index construction method and device, equipment and medium

Abstract

The embodiment of the invention provides an evaluation index construction method and device based on a knowledge graph, equipment and a medium, which can be applied to the field of financial science and technology, and the method comprises the following steps: carrying out sub-graph retrieval in the knowledge graph to obtain an initial graph; selecting a target type from relation types contained in the initial image as an image segmentation basis to obtain a segmented image; reconstructing the segmented image into a reconstructed image; and determining a target evaluation dimension according to the target type, generating context information according to the reconstructed graph and the target evaluation dimension, and generating a target evaluation index under the target evaluation dimension based on the context information through a large language model. According to the embodiment of the invention, the large language model focuses on analyzing key factors under the meaning of a single business, and more accurate evaluation indexes are generated.

Description

Technical Field

[0001] This application relates to the technical field of natural language processing, is applicable to the field of fintech, and particularly relates to a method and device, equipment, and medium for constructing evaluation indicators based on a knowledge graph. Background Art

[0002] Evaluation indicators are specific signs describing the state of a system. For example, when evaluating a certain solution, cost evaluation indicators are usually set up to measure the benefits of implementing the solution. In the field of fintech, especially in the financial risk control scenario, risk indicators need to be established for risk analysis.

[0003] In related methods, the construction of the evaluation indicator system mainly relies on expert discussions, with strong subjectivity and weak reliability of the evaluation indicators. Therefore, how to improve the accuracy of evaluation indicators has become an urgent technical problem to be solved. Summary of the Invention

[0004] The main purpose of the embodiments of this application is to propose a method and device, equipment, and medium for constructing evaluation indicators based on a knowledge graph, aiming to improve the accuracy of evaluation indicators.

[0005] To achieve the above purpose, the first aspect of the embodiments of this application proposes a method for constructing evaluation indicators based on a knowledge graph, and the method includes:

[0006] Obtain a target problem, where the target problem includes an evaluation object;

[0007] Perform subgraph retrieval in a preset knowledge graph according to the evaluation object to obtain an initial graph;

[0008] Obtain the relationship types of entity relationships in the initial graph, and select a target type from the relationship types;

[0009] Based on the target type, perform graph segmentation on the initial graph to obtain a segmented graph; the relationship types of entity relationships in the same segmented graph are the same;

[0010] Determine a target evaluation dimension according to the target type;

[0011] Perform graph data reconstruction on the segmented graph to obtain a reconstructed graph;

[0012] Generate context information according to the reconstructed graph and the target evaluation dimension;

[0013] Generate a reply to the target problem and the context information through a preset large language model to obtain a target answer; the target answer includes target evaluation indicators corresponding to the target evaluation dimension.

[0014] To achieve the above purpose, the second aspect of the embodiments of this application proposes a device for constructing evaluation indicators based on a knowledge graph, and the device includes:

[0015] A problem acquisition module, configured to acquire a target problem, where the target problem includes an evaluation object;

[0016] A sub-graph retrieval module, configured to perform sub-graph retrieval in a preset knowledge graph according to the evaluation object to obtain an initial graph;

[0017] A target type selection module, configured to obtain the relationship type of the entity relationship in the initial graph and select a target type from the relationship types;

[0018] A graph segmentation module, configured to segment the initial graph based on the target type to obtain a segmented graph; the relationship types of the entity relationships in the same segmented graph are the same;

[0019] A target evaluation dimension determination module, configured to determine a target evaluation dimension according to the target type;

[0020] A graph data reconstruction module, configured to reconstruct the graph data of the segmented graph to obtain a reconstructed graph;

[0021] A context information generation module, configured to generate context information according to the reconstructed graph and the target evaluation dimension;

[0022] A response generation module, configured to generate a response to the target problem and the context information through a preset large language model to obtain a target answer; the target answer includes target evaluation indicators corresponding to the target evaluation dimension.

[0023] To achieve the above object, a third aspect of the embodiments of the present application proposes an electronic device, which includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the method of the first aspect is implemented.

[0024] To achieve the above object, a fourth aspect of the embodiments of the present application proposes a computer-readable storage medium, which stores a computer program, and when the computer program is executed by a processor, the method of the first aspect is implemented.

[0025] The method, device, equipment, and medium for constructing evaluation indicators based on a knowledge graph proposed in this application retrieve entities or entity relationships related to the evaluation object in the knowledge graph through subgraph retrieval to obtain an initial graph; select a target type from the relationship types contained in the initial graph as the basis for graph segmentation, and segment the initial subgraph into segmented graphs that focus on a single business meaning (that is, the relationship types of entity relationships in the same segmented graph are the same); then reconstruct the segmented graphs into reconstructed graphs, which can effectively filter out invalid entity relationships; determine the target evaluation dimension according to the target type, associate the relationship type with the evaluation dimension, and each reconstructed graph is used as the context information for the corresponding evaluation dimension. The evaluation indicators under this evaluation dimension are generated by a large language model based on the context information, which is beneficial for the large language model to focus on analyzing the key factors under a single business meaning and generate more accurate evaluation indicators. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 is a flowchart of the method for constructing evaluation indicators based on a knowledge graph provided by an embodiment of this application;

[0027] Figure 2 is Figure 1 a flowchart of step S102 in

[0028] Figure 3 is Figure 1 a flowchart of step S103 in

[0029] Figure 4 is Figure 1 a flowchart of step S105 in

[0030] Figure 5 is Figure 1 a flowchart of step S106 in

[0031] Figure 6 is Figure 1 another flowchart of step S106 in

[0032] Figure 7 is a schematic structural diagram of the device for constructing evaluation indicators based on a knowledge graph provided by an embodiment of this application;

[0033] Figure 8 is a schematic hardware structure diagram of an electronic device provided by an embodiment of this application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0034] In order to make the objectives, technical solutions, and advantages of this application clearer, the following further elaborates on this application in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not used to limit this application.

[0035] It should be noted that although functional modules are divided in the device schematic diagram and the logical sequence is shown in the flowchart, in some cases, the steps shown or described can be executed in a different module division from that in the device or a different sequence from that in the flowchart. Terms such as "first" and "second" in the specification, claims and the above-mentioned drawings are used to distinguish similar objects and do not necessarily describe a specific order or sequence.

[0036] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0037] First, several terms involved in this application are parsed as follows:

[0038] Artificial intelligence (AI): It is a new technical science that studies and develops theories, methods, technologies and application systems for simulating, extending and expanding human intelligence; artificial intelligence is a branch of computer science. Artificial intelligence attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a way similar to human intelligence. The research in this field includes robots, speech recognition, image recognition, natural language processing and expert systems, etc. Artificial intelligence can simulate the information process of human consciousness and thinking. Artificial intelligence also refers to the theory, method, technology and application system that uses a digital computer or a machine controlled by a digital computer to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0039] Natural language processing (NLP): NLP uses a computer to process, understand and apply human languages (such as Chinese, English, etc.). NLP is a branch of artificial intelligence and an interdisciplinary subject of computer science and linguistics, and is often referred to as computational linguistics. Natural language processing includes syntactic analysis, semantic analysis, discourse understanding, etc. Natural language processing is commonly used in technical fields such as machine translation, handwritten and printed character recognition, speech recognition and text-to-speech conversion, information intention recognition, information extraction and filtering, text classification and clustering, public opinion analysis and opinion mining, etc., and involves data mining, machine learning, knowledge acquisition, knowledge engineering, artificial intelligence research related to language processing and linguistic research related to language computing, etc.

[0040] Subgraph retrieval (SR): SR is the process of locating and extracting relevant subgraphs from a graph database according to query conditions (such as keywords, semantic similarity or path matching).

[0041] Knowledge Graph (KG): A KG is a way of organizing and representing knowledge in a graph structure, used to describe concepts in the real world and their relationships. A KG connects entities and entity relationships to form a knowledge network that can express real-world information. The knowledge graph belongs to the field of artificial intelligence, is a further development of semantic networks and knowledge engineering, and is often referred to as a semantic network. Its main tasks include knowledge extraction, knowledge fusion, knowledge processing, etc. Knowledge graphs are commonly used in technical fields such as intelligent question answering, recommendation systems, and semantic search, and involve research in knowledge acquisition, knowledge representation, machine learning, and knowledge engineering.

[0042] Graph Partitioning (GP): Graph partitioning refers to dividing a graph into two or more subgraphs such that each subgraph maintains the integrity of the graph structure under specific conditions. Graph partitioning belongs to the fields of data structures and algorithms, computer vision, etc., and is an interdisciplinary subject of graph theory and computer vision. Its main task is to divide a graph into several subgraphs, and common goals include minimizing the number of edges between subgraphs and maximizing the connectivity within subgraphs. Graph partitioning is commonly used in technical fields such as social network analysis, image segmentation, and circuit layout, and involves research directions such as graph theory, data structures, and algorithm design.

[0043] Graph Data Reconstruction (GDR): Graph data reconstruction refers to the process of reorganizing, adjusting, or optimizing graph data. Graph data reconstruction belongs to the fields of data processing, data structures and algorithms, etc., and is an interdisciplinary subject of graph theory and data mining. Its main task is to convert the original graph data into a more suitable form for better storage, processing, and analysis. Graph data reconstruction is commonly used in technical fields such as social network analysis, bioinformatics, and traffic network optimization, and involves research directions such as data structures, algorithm design, and data mining.

[0044] Graph Convolution Layer (GCL): The graph convolution layer is the basic building block of a graph neural network (GNN). It aggregates node features to its neighbor nodes and updates the node representation using a non-linear transformation. This process is similar to the convolution operation in a traditional convolutional neural network (CNN), but the graph convolution layer is applicable to graph-structured data and can handle the irregular connection relationships between nodes. The graph convolution layer is commonly used in deep learning tasks on graph data, such as node classification, graph classification, and link prediction, and involves research directions such as graph neural networks, deep learning, and graph signal processing.

[0045] In the related art, due to the complex dependency relationships between entities in the knowledge graph, the accuracy of the method for constructing metrics based on the knowledge graph is not good. For example, in the field of fintech, since the professional knowledge graph is constructed based on complex and chaotic multi-source heterogeneous financial data, the entities in the professional knowledge graph have dependency relationships with multiple business meanings. If the data obtained from the financial knowledge graph is directly used as context information, it is difficult for the large language model to focus on the key factors under each business meaning, and the accuracy of the obtained evaluation metrics is not good.

[0046] Based on this, the embodiments of the present application provide a method, apparatus, device and medium for constructing evaluation metrics based on a knowledge graph, aiming to improve the accuracy of the evaluation metrics.

[0047] The method, apparatus, device and medium for constructing evaluation metrics based on a knowledge graph provided by the embodiments of the present application are specifically described through the following embodiments. First, a method for constructing evaluation metrics based on a knowledge graph in the embodiments of the present application is described.

[0048] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence (AI) is a theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0049] Artificial intelligence basic technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, and mechatronics. Artificial intelligence software technologies mainly include several major directions such as computer vision technology, robotics technology, biometric technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0050] A method for constructing evaluation metrics based on a knowledge graph provided by the embodiments of the present application relates to the field of natural language processing technology. A method provided by the embodiments of the present application can be applied to a terminal, or to a server side, or can also be software running on a terminal or a server side. In some embodiments, the terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, etc.; the server side can be configured as an independent physical server, or can be configured as a server cluster or distributed system composed of multiple physical servers, or can also be configured as a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application implementing a method, etc., but is not limited to the above forms.

[0051] This application can be used in numerous general-purpose or special-purpose computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multi-processor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, and so on. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. This application can also be practiced in a distributed computing environment, where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0052] It should be noted that in each specific implementation manner of this application, when it comes to relevant processing that needs to be carried out based on data related to the user's identity or characteristics, such as user information, user behavior data, user historical data, and user location information, the user's permission or consent will be obtained first. Moreover, the collection, use, and processing of these data will comply with relevant laws, regulations, and standards. In addition, when this application embodiment needs to obtain the user's sensitive personal information, the user's separate permission or separate consent will be obtained through methods such as pop-up windows or jumping to a confirmation page. After clearly obtaining the user's separate permission or separate consent, the necessary user-related data for this application embodiment to operate normally will be obtained. The non-company software tools or components that appear in this application embodiment are only for illustrative purposes and do not represent actual use.

[0053] Figure 1 is an optional flowchart of a method for constructing evaluation indicators based on a knowledge graph provided by an embodiment of this application. Figure 1 The method in may include but is not limited to steps S101 to S108.

[0054] Step S101, obtain a target problem, where the target problem includes an evaluation object;

[0055] Step S102, perform subgraph retrieval in a preset knowledge graph according to the evaluation object to obtain an initial graph;

[0056] Step S103, obtain the relationship types of the entity relationships in the initial graph, and select a target type from the relationship types;

[0057] Step S104, based on the target type, perform graph segmentation on the initial graph to obtain a segmented graph; the relationship types of the entity relationships in the same segmented graph are the same;

[0058] Step S105: Determine the target evaluation dimension according to the target type.

[0059] Step S106: Reconstruct the graph data of the segmented graph to obtain a reconstructed graph.

[0060] Step S107: Generate context information based on the reconstructed graph and the target evaluation dimension.

[0061] Step S108: Generate a target answer by using a pre-set large language model to answer the target question and the context information; the target answer includes the target evaluation indicators corresponding to the target evaluation dimension.

[0062] Steps S101 to S108 shown in the embodiments of the present application retrieve entities or entity relationships related to the evaluation object in the knowledge graph through sub-graph retrieval to obtain an initial graph; select the target type from the relationship types contained in the initial graph as the basis for graph segmentation, and segment the initial sub-graph into segmented graphs that focus on a single business meaning (that is, the relationship types of entity relationships in the same segmented graph are the same); then reconstruct the segmented graph into a reconstructed graph, which can effectively filter out invalid entity relationships; determine the target evaluation dimension according to the target type, so that the relationship type is associated with the evaluation dimension, and each reconstructed graph is used as the context information for the corresponding evaluation dimension. The evaluation indicators under this evaluation dimension are generated through a large language model based on the context information, which is conducive to the large language model focusing on analyzing the key factors under a single business meaning and generating more accurate evaluation indicators.

[0063] In step S101 of some embodiments, the evaluation object may be an individual, an organization, etc., which is not limited thereto. For example, in the fintech scenario, the evaluation object may be an enterprise, a bank, a financial product, or a supply chain network. When the target question is input in natural language form, the evaluation object in the target question can be identified through methods such as named entity recognition and semantic parsing. For example, when the target question is obtained as evaluating the supply chain stability of enterprise A under the background of rising raw material prices, the supply chain of enterprise A in the target question is identified as the evaluation object.

[0064] In step S102 of some embodiments, the pre-set knowledge graph may be a domain knowledge graph constructed based on professional knowledge, such as a financial knowledge graph or a medical knowledge graph; it may also be a general knowledge graph; it is not limited thereto.

[0065] Please refer to Figure 2 , in some embodiments, step S102 may include but is not limited to steps S201 to S202:

[0066] Step S201: Perform entity retrieval in the knowledge graph according to the evaluation object to obtain the target entity.

[0067] Step S202: Extract the sub-graph containing the target entity in the knowledge graph to obtain the initial graph.

[0068] In step S201 of some embodiments, retrieval keywords can be constructed according to the evaluation object, and relevant entities can be retrieved as target entities through structured query statements; alternatively, the evaluation object and the entities in the knowledge graph can be vectorized, and the most similar entity can be selected as the target entity based on vector similarity.

[0069] In step S202 of some embodiments, the neighbor nodes within a certain range of the target entity can be extracted as the initial graph with the target entity as the center, and the range of the initial graph can be controlled by setting the number of layers or distance threshold of the neighbor nodes; alternatively, multiple nodes related to the target entity can be extracted in the knowledge graph according to predefined path patterns; the scope is not limited thereto.

[0070] It is easy to understand that the knowledge graph is graph data, and the entity is the node in the graph data. The entity feature refers to the attribute or feature vector possessed by each node in the graph data, which is used to describe the internal attribute or state of the node. The entity feature can be numerical, categorical, text-based, etc. The entity relationship refers to the edge between nodes in the graph data, which is used to represent the association between entities. The entity relationship can also be provided with relationship features, and the relationship feature refers to the attribute or feature vector possessed by the entity relationship. For example, the relationship feature can represent the relationship type.

[0071] For example, in the fintech scenario, a financial knowledge graph is constructed. The financial knowledge graph contains entities: leasing company, battery factory, chip factory, lithium miner, and entity relationships: (leasing company, battery factory), (leasing company, chip factory), (battery factory, lithium miner). The entity features of the leasing company include revenue and products, the entity features of the battery factory include battery production capacity and battery sales amount; the entity features of the chip factory include chip production capacity and chip sales amount; the entity features of the lithium miner include lithium production capacity and lithium price. The relationship features of (leasing company, battery factory) include purchase amount and account period; the relationship features of (leasing company, chip factory) include purchase amount and account period; the relationship feature of (battery factory, lithium miner) includes annual purchase volume. The target problem is obtained as evaluating the financial risk of the leasing company, and the financial risk of the leasing company in the target problem is identified as the evaluation object. The leasing company is used as the retrieval keyword, and the entity with the same name is retrieved in the financial knowledge graph as the target entity. With the target entity as the center, the entities and entity relationships within one hop range are extracted as the initial graph. The initial graph includes three entities: the leasing company, the battery factory, and the chip factory, the entity relationships between these three entities, as well as the entity features and relationship features of these three entities.

[0072] The steps S201 to S202 illustrated in the embodiments of the present application retrieve entity in the knowledge graph according to the evaluation object to obtain the target entity, extract the initial graph including the target entity, and integrate the evaluation object and its related entities and relationships into the initial graph, providing rich context information for subsequent steps.

[0073] Please refer to Figure 3 , in some embodiments, step S103 may include but is not limited to steps S301 to S302:

[0074] Step S301, obtain the relationship type of the entity relationships in the initial graph, calculate the number of entity relationships belonging to the same relationship type, and obtain the relationship quantity;

[0075] Step S302, based on the relationship quantity, select a target type from the relationship types.

[0076] Understandably, there are multiple relationship types for entity relationships. For example, in the fintech scenario, the relationship types can be shareholding relationship, holding relationship, guarantee relationship, supplier relationship, stock pledge.

[0077] In step S302 of some embodiments, the relationship types can be sorted using the relationship quantity, and the K relationship types with the largest relationship quantity can be selected as the target types, where k is greater than or equal to 1; or a quantity threshold can be set, and the relationship types with a relationship quantity greater than the quantity threshold can be selected as the target types; it is not limited to this.

[0078] Steps S301 to S302 illustrated in the embodiments of the present application obtain the relationship quantity by calculating the number of entity relationships belonging to the same relationship type in the initial graph; according to the relationship quantity, a target type is selected from the relationship types, filtering out unimportant relationship types, thereby reducing the number of segmented graphs, and thus reducing the subsequent calculation amount.

[0079] In step S103 of some embodiments, a target type can also be selected from the relationship types based on the correlation between the relationship type and the evaluation object. For example, if the evaluation object is enterprise financial risk, then relationship types such as guarantee relationship, related party transactions, etc. are selected as the target types; or if the evaluation object is enterprise investment situation, then relationship types such as patent authorization, R & D cooperation, etc. are selected as the target types; or if the evaluation object is drug safety, then relationship types such as drug - drug interaction, allergy history, etc. are selected as the target types.

[0080] In step S104 of some embodiments, understandably, when there are multiple relationship types for entity relationships, multiple corresponding segmented graphs are obtained after graph segmentation.

[0081] In the fintech scenario, the evaluation object is enterprise A. The entities included in the initial graph obtained after subgraph retrieval can be: enterprise A, group B, supplier C, bank D, leasing company E, and guarantee company F; the entity relationships include: equity shareholding, supply chain procurement, financial credit, guarantee relationship, and patent licensing. Selecting equity shareholding, guarantee relationship, and financial credit as the target types, the corresponding equity shareholding relationship graph, guarantee relationship graph, and financial credit graph are obtained as the segmentation graphs after graph segmentation. Among them, the equity shareholding relationship graph includes entities: enterprise A, group B, and leasing company E, and the entity relationships: (group B, enterprise A), (enterprise A, leasing company E); the guarantee relationship graph includes entities: enterprise A, group B, and guarantee company F, and the entity relationships: (group B, enterprise A), (enterprise A, guarantee company F); the financial credit graph includes entities: enterprise A, bank D, and leasing company E, and the entity relationships: (bank D, enterprise A), (leasing company E, bank D).

[0082] In step S105 of some embodiments, the target type can be converted into the target evaluation dimension through a large language model; or a mapping relationship between the relationship type and the evaluation dimension can be established in advance based on factors such as semantic similarity, so that based on the mapping relationship, the target type is mapped into the target evaluation dimension; not limited to this. It should be particularly noted that the mapping relationship between the relationship type and the evaluation dimension can be one-to-one or many-to-one.

[0083] In the fintech scenario, if the target type is equity shareholding, then the target evaluation dimension is determined to be the stability of control rights; if the target type is the guarantee relationship, then the target evaluation dimension is determined to be the debt risk; if the target type is supply chain procurement, then the target evaluation dimension is determined to be the business continuity risk.

[0084] Please refer to Figure 4 , in some embodiments, step S105 may include but is not limited to steps S401 to S402:

[0085] Step S401, calculate the semantic similarity between the target type and the preset evaluation dimension to obtain similarity data;

[0086] Step S402, based on the similarity data, screen the preset evaluation dimension to obtain the target evaluation dimension.

[0087] Understandably, in some application scenarios, such as the field of financial risk assessment, multiple important evaluation dimensions are usually preset. Specifically, when the evaluation object is the credit business of a commercial bank, the preset evaluation dimensions are: capital adequacy, asset quality, profitability, liquidity risk, and management effectiveness; when the evaluation object is supply chain financial risk, the preset evaluation dimensions are: technology risk, moral risk, inventory pledge valuation fluctuation, and enterprise operation stability. For example, if the target types obtained are shareholding relationship, controlling relationship, and guarantee relationship, and the calculated semantic similarity between these three target types and capital adequacy exceeds the threshold, then capital adequacy is selected as the target evaluation dimension. Subsequently, the context information generated by the corresponding reconstructed graphs of these three target types is input into the large language model to obtain the target evaluation indicators for capital adequacy.

[0088] In steps S401 to S402 illustrated in the embodiments of the present application, by semantic similarity, the target evaluation dimension is selected from the preset evaluation dimensions, and the association relationship between the relationship type and the evaluation dimension is automatically constructed, which is beneficial to improving the objectivity of the evaluation indicators.

[0089] In step S106 of some embodiments, deep learning models such as graph neural networks and autoencoders can be used to reconstruct the graph data of the segmented graph, or the graph data of the segmented graph can be reconstructed by merging entities, migrating entity relationships, etc., and the like.

[0090] Please refer to Figure 5 , in some embodiments, step S106 may include but is not limited to steps S501 to S506:

[0091] Step S501, obtain the entity features of the initial graph to obtain the initial graph features;

[0092] Step S502, obtain the entity relationships of the segmented graph to obtain the segmented graph relationships;

[0093] Step S503, through the first graph convolutional layer, perform graph convolutional processing on the initial graph features and the segmented graph relationships to obtain the first segmented graph features;

[0094] Step S504, perform feature aggregation processing on the first segmented graph features to obtain the first aggregated features;

[0095] Step S505, through the second graph convolutional layer, perform graph convolutional processing on the first aggregated features and the segmented graph relationships to obtain the second segmented graph features;

[0096] Step S506, perform decoding processing on the second segmented graph features through a decoder to obtain a reconstructed graph.

[0097] Understandably, the segmentation graph relationship can be represented by an adjacency matrix, which is a two-dimensional matrix used to represent the associations between nodes in graph data. The rows and columns of the matrix correspond to the nodes in the graph data respectively, and the values in the matrix represent the association strength between nodes or the existence of an association.

[0098] In step S504 of some embodiments, the first segmentation graph feature can be subjected to feature aggregation processing, such as feature addition, feature concatenation, feature weighted average, etc., to obtain a first aggregated feature, and it is not limited thereto.

[0099] In some embodiments, after step S505, the method for constructing an evaluation index based on a knowledge graph may further include:

[0100] Perform feature aggregation processing on the second segmentation graph feature to obtain a second aggregated feature;

[0101] Through a third graph convolutional layer, perform graph convolutional processing on the second aggregated feature and the segmentation graph relationship to obtain a third segmentation graph feature;

[0102] Replace the input data of the decoder with the third segmentation graph feature, that is, input the third segmentation graph feature into the decoder, and perform decoding processing on the third segmentation graph feature through the decoder to obtain a reconstructed graph.

[0103] Understandably, a fourth graph convolutional layer, a fifth graph convolutional layer, etc. can also be added in the embodiments of the present application to extract deeper feature representations.

[0104] Please refer to Figure 6 , in some embodiments, after step S505, the method for generating an evaluation index based on a knowledge graph further includes updating the second segmentation graph feature, which may include but is not limited to steps S601 to S602:

[0105] Step S601, perform feature aggregation processing on the second segmentation graph feature to obtain a second aggregated feature;

[0106] Step S602, perform concatenation processing on the second aggregated feature and the second segmentation graph feature to obtain a concatenated graph feature, and use the concatenated graph feature as the second segmentation graph feature.

[0107] In step S601 of some embodiments, the first segmentation graph feature can be subjected to feature aggregation processing, such as feature addition, feature concatenation, feature weighted average, etc., to obtain a second aggregated feature, and it is not limited thereto.

[0108] Steps S601 to S602 illustrated in the embodiments of the present application obtain a second aggregated feature by aggregating the features of the second segmentation graph; and splice the second segmentation graph feature representing the segmentation graph node feature and the second aggregated feature representing the initial graph node feature, thereby updating the second segmentation graph feature; which can extract important graph structure information from the global and local perspectives associated with the evaluation object, and further improve the representation ability of the second segmentation graph feature.

[0109] In the fintech scenario, an encoder including three parallel graph convolutional branches is constructed. The first graph convolutional branch includes a first graph convolutional layer and a second graph convolutional layer; the second graph convolutional branch includes a third graph convolutional layer and a fourth graph convolutional layer; the third graph convolutional branch includes a fifth graph convolutional layer and a sixth graph convolutional layer. The initial graph is segmented into an equity holding relationship graph, a guarantee relationship graph, and a financial credit graph, serving as the first segmentation graph, the second segmentation graph, and the third segmentation graph. The entity features of each entity in the initial graph are obtained to get the initial graph feature. The adjacency matrices of the above three segmentation graphs are respectively obtained as the first segmentation graph relationship, the second segmentation graph relationship, and the third segmentation graph relationship.

[0110] The initial graph feature and the first segmentation graph relationship are input into the first graph convolutional branch. Through the first graph convolutional layer, graph convolutional processing is performed on the initial graph feature and the first segmentation graph relationship to obtain a segmentation graph feature a. The initial graph feature and the second segmentation graph relationship are input into the second graph convolutional branch. Through the third graph convolutional layer, graph convolutional processing is performed on the initial graph feature and the second segmentation graph relationship to obtain a segmentation graph feature b. The initial graph feature and the third segmentation graph relationship are input into the third graph convolutional branch. Through the fifth graph convolutional layer, graph convolutional processing is performed on the initial graph feature and the third segmentation graph relationship to obtain a segmentation graph feature c. The segmentation graph features a, b, and c are added together to obtain the above-mentioned first aggregated feature.

[0111] Through the second graph convolutional layer, graph convolutional processing is performed on the first aggregated feature and the first segmentation graph relationship to obtain a segmentation graph feature e. Through the fourth graph convolutional layer, graph convolutional processing is performed on the first aggregated feature and the second segmentation graph relationship to obtain a segmentation graph feature f. Through the sixth graph convolutional layer, graph convolutional processing is performed on the first aggregated feature and the third segmentation graph relationship to obtain a segmentation graph feature g. The segmentation graph features e, f, and g are added together to obtain the above-mentioned second aggregated feature.

[0112] The second aggregated feature and the segmentation graph feature e are spliced to obtain a segmentation graph feature e'; and the decoder decodes the segmentation graph feature e' to obtain a reconstructed equity holding relationship graph.

[0113] The second aggregated feature and the segmentation graph feature f are concatenated to obtain a segmentation graph feature f'; the decoder decodes the segmentation graph feature f' to obtain a reconstructed guarantee relationship graph.

[0114] The second aggregated feature and the segmentation graph feature g are concatenated to obtain a segmentation graph feature g'; the decoder decodes the segmentation graph feature g' to obtain a reconstructed financial credit graph.

[0115] In step S506 of some embodiments, the second segmentation graph feature can be decoded by a decoder based on a graph neural network (GNN) to obtain a reconstructed graph; or the second segmentation graph feature can be decoded by a graph autoencoder (GAE) to obtain a reconstructed graph; it is not limited to this.

[0116] In steps S501 to S506 illustrated in the embodiments of the present application, by using the initial graph feature containing the entity features in the initial graph and the segmentation graph relationship containing the entity relationships in the segmentation graph as the input data of the graph convolutional layer, the first graph convolutional layer updates the representation of the current node by aggregating the information of neighbor nodes. This process uses the initial graph feature to learn the global representation of entities and uses the segmentation graph relationship to control the information propagation path as the same type of entity relationship, reducing the interference of different types of entity relationships on feature extraction; the first segmentation graph feature output by the first graph convolutional layer is feature-aggregated to obtain a first aggregated feature; based on the first aggregated feature, the second graph convolutional layer aggregates the information of neighbor nodes in a wider range, fusing the information of more different-level neighbor nodes, improving the representation ability of the second segmentation graph feature, and thus improving the data quality of the reconstructed graph obtained based on the second segmentation graph feature.

[0117] In step S107 of some embodiments, the reconstructed graph can be converted into a natural language sequence, the target evaluation dimension is combined with the natural language sequence, and structured markers (such as delimiters) are added to enhance the structured semantics to obtain context information; or the entities and entity relationships in the reconstructed graph can be filled into the placeholders in a preset text template to obtain context information; it is not limited to this.

[0118] In some embodiments, after step S107, the method for generating evaluation indicators based on a knowledge graph further includes updating the context information, specifically including:

[0119] Step B1, obtaining a preset knowledge vector library; the preset knowledge vector library includes text blocks and text vectors of the text blocks;

[0120] Step B2, vectorizing the target problem to obtain a problem vector;

[0121] Step B3, calculating the distance between the problem vector and the text vector to obtain a vector distance;

[0122] Step B4, select target vectors from the text vectors based on the vector distance;

[0123] Step B5, screen out target blocks from the text blocks based on the target vectors;

[0124] Step B6, fill the target blocks into the context information.

[0125] Understandably, the knowledge vector library is a pre-constructed database that contains a large number of text blocks and their corresponding text vectors, and is used to store and index the vector representations of text data for quick retrieval and matching. A text block is the basic unit in the knowledge vector library and is usually a text segment with a certain semantic integrity. A text block can be a sentence, a paragraph, or a part of a document. A text vector is the numerical representation of a text block, and usually converts the text into a fixed-length vector through an embedding method (such as Word2Vec, BERT, GPT, etc.). The text vector can capture the semantic information of the text, facilitating similarity calculation and retrieval.

[0126] In the field of fintech, a knowledge vector library based on financial knowledge is constructed. The knowledge vector library can include the following text blocks: Text block 1: "A large technology company releases a new product, and the market reaction is positive", Text block 2: "The central bank announces to maintain the interest rate unchanged, and the market expectation is stable", Text block 3: "An industry faces rising raw material prices and increasing cost pressure".

[0127] Steps B1 to B6 shown in the embodiments of the present application can provide richer background information for generating evaluation indicators subsequently by retrieving text vectors similar to the question vector in the preset knowledge vector library and obtaining target blocks to fill into the context information, thereby improving the accuracy of the evaluation indicators.

[0128] In step S108 of some embodiments, a prompt can also be constructed, and the target question, the prompt, and the context information are input into the preset large language model together for reply generation to obtain a target answer. In some application scenarios, the target answer can include the monitoring threshold, calculation method, etc. of the evaluation indicator.

[0129] In the field of fintech, obtain the target question: Evaluate the supply chain financial risk of a cross-border e-commerce enterprise and generate key monitoring indicators; construct a prompt: As a supply chain financial risk control expert, please design quantitative evaluation indicators and thresholds according to the provided evaluation dimensions.

[0130] In summary, relevant entities or entity relationships of the evaluation object are retrieved in the knowledge graph through subgraph retrieval to obtain an initial graph; a target type is selected from the relationship types contained in the initial graph as the basis for graph segmentation, and the initial subgraph is segmented into segmented graphs that focus on a single business meaning (that is, the relationship types of entity relationships in the same segmented graph are the same); then the segmented graphs are reconstructed into reconstructed graphs. During the reconstruction process, the relationship control information of the segmented graphs is used to control the propagation path of information as the same type of entity relationship, reducing the interference of different types of entity relationships on feature extraction. At the same time, feature aggregation is performed on the features of different segmented graphs to improve the representation ability of the features of the segmented graphs, and invalid entity relationships can be effectively filtered out; the target evaluation dimension is determined according to the target type, and the relationship type is associated with the evaluation dimension. Then each reconstructed graph is used as the context information for the corresponding evaluation dimension, and the evaluation indicators under this evaluation dimension are generated by the large language model based on the context information, which is conducive to the large language model focusing on analyzing the key factors under a single business meaning and generating more accurate evaluation indicators

[0131] Please refer to Figure 7 , this embodiment of the present application also provides an evaluation index construction device based on a knowledge graph, which can implement the above-mentioned evaluation index construction method based on a knowledge graph. The device includes:

[0132] A problem acquisition module, configured to acquire a target problem, where the target problem includes an evaluation object;

[0133] A subgraph retrieval module, configured to perform subgraph retrieval in a preset knowledge graph according to the evaluation object to obtain an initial graph;

[0134] A target type selection module, configured to obtain the relationship types of entity relationships in the initial graph and select a target type from the relationship types;

[0135] A graph segmentation module, configured to perform graph segmentation on the initial graph based on the target type to obtain segmented graphs; the relationship types of entity relationships in the same segmented graph are the same;

[0136] A target evaluation dimension determination module, configured to determine a target evaluation dimension according to the target type;

[0137] A graph data reconstruction module, configured to perform graph data reconstruction on the segmented graphs to obtain reconstructed graphs;

[0138] A context information generation module, configured to generate context information according to the reconstructed graphs and the target evaluation dimension;

[0139] A response generation module, configured to generate a response to the target problem and the context information through a preset large language model to obtain a target answer; the target answer includes the target evaluation indicators corresponding to the target evaluation dimension.

[0140] The specific implementation manner of this device is basically the same as the specific embodiment of the above-mentioned knowledge graph-based evaluation index generation method, and will not be elaborated here.

[0141] An embodiment of this application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the above-mentioned knowledge graph-based evaluation index generation method. This electronic device can be any intelligent terminal including a tablet computer, an in-vehicle computer, etc.

[0142] Please refer to Figure 8 , Figure 8 which shows the hardware structure of the electronic device of another embodiment. The electronic device includes:

[0143] A processor 901, which can be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, etc., and is used to execute relevant programs to implement the technical solutions provided by the embodiments of this application;

[0144] A memory 902, which can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM), etc. The memory 902 can store an operating system and other application programs. When implementing the technical solutions provided by the embodiments of this specification through software or firmware, the relevant program codes are stored in the memory 902, and the processor 901 is called to execute the knowledge graph-based evaluation index generation method of the embodiments of this application;

[0145] An input / output interface 903, which is used to implement information input and output;

[0146] A communication interface 904, which is used to implement communication interaction between this device and other devices, and can achieve communication through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.);

[0147] A bus 905, which transmits information between various components of the device (such as the processor 901, the memory 902, the input / output interface 903, and the communication interface 904);

[0148] Among them, the processor 901, the memory 902, the input / output interface 903, and the communication interface 904 are communicatively connected to each other inside the device through the bus 905.

[0149] An embodiment of the present application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above method for generating evaluation indicators based on a knowledge graph is implemented.

[0150] As a non-transitory computer-readable storage medium, a memory can be used to store non-transitory software programs and non-transitory computer-executable programs. In addition, the memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, a flash memory device, or other non-transitory solid-state storage devices. In some embodiments, the memory may optionally include a memory remotely disposed relative to the processor, and these remote memories can be connected to the processor through a network. Examples of the above network include but are not limited to the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof.

[0151] The method for constructing evaluation indicators based on a knowledge graph, the device for constructing evaluation indicators based on a knowledge graph, an electronic device, and a storage medium provided by the embodiments of the present application retrieve entities or entity relationships related to an evaluation object in the knowledge graph through subgraph retrieval to obtain an initial graph; select a target type from the relationship types contained in the initial graph as the basis for graph segmentation, and segment the initial subgraph into segmented graphs that focus on a single business meaning (that is, the relationship types of entity relationships in the same segmented graph are the same); then reconstruct the segmented graphs into reconstructed graphs, which can effectively filter out invalid entity relationships; determine a target evaluation dimension according to the target type, associate the relationship type with the evaluation dimension, and each reconstructed graph is used as context information for the corresponding evaluation dimension. The evaluation indicators under this evaluation dimension are generated by a large language model based on the context information, which is beneficial for the large language model to focus on analyzing key factors under a single business meaning and generate more accurate evaluation indicators.

[0152] The embodiments described in the embodiments of the present application are for more clearly illustrating the technical solutions of the embodiments of the present application, and do not constitute a limitation on the technical solutions provided by the embodiments of the present application. Those skilled in the art know that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of the present application are equally applicable to similar technical problems.

[0153] Those skilled in the art can understand that the technical solutions shown in the figures do not limit the embodiments of the present application, and may include more or fewer steps than shown in the figures, or combine certain steps, or different steps.

[0154] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, that is, they may be located in one place, or may be distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0155] Those of ordinary skill in the art can understand that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, and their appropriate combinations.

[0156] In the description of this application and the above-mentioned accompanying drawings, the terms "first", "second", "third", "fourth", etc. (if any) are used to distinguish similar objects, and do not necessarily have to be used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of this application described here can be implemented in an order other than those illustrated or described here. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products, or devices.

[0157] It should be understood that in this application, "at least one (item)" means one or more, and "a plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships can exist. For example, "A and / or B" can mean: only A exists, only B exists, and both A and B exist at the same time. Among them, A and B can be singular or plural. The character " / " generally means that the associated objects before and after are in an "or" relationship. "At least one (one) of the following" or its similar expression means any combination of these items, including any combination of single item (one) or plural items (ones). For example, at least one (one) of a, b, or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0158] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the above-mentioned unit division is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection to each other can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be in an electrical, mechanical or other form.

[0159] The units described above as separate components may or may not be physically separated. The components shown as units may or may not be physical units, that is, they may be located in one place or distributed over multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0160] In addition, the functional units in various embodiments of the present application may be integrated into a processing unit, or each unit may exist physically alone, or two or more units may be integrated into one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.

[0161] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes multiple instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods in various embodiments of the present application. The foregoing storage medium includes: USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs and other various media that can store programs.

[0162] The preferred embodiments of the embodiments of the present application have been described above with reference to the accompanying drawings. However, this does not limit the scope of the rights of the embodiments of the present application. Any modifications, equivalent replacements, and improvements made by those skilled in the art without departing from the scope and essence of the embodiments of the present application shall be within the scope of the rights of the embodiments of the present application.

Claims

1. A method for generating evaluation indicators based on a knowledge graph, characterized in that The method includes: Obtain a target problem, where the target problem includes an evaluation object; According to the evaluation object, perform subgraph retrieval in a preset knowledge graph to obtain an initial graph; Obtain the relationship types of entity relationships in the initial graph, and select a target type from the relationship types; Based on the target type, perform graph segmentation on the initial graph to obtain a segmented graph; the relationship types of entity relationships in the same segmented graph are the same; Determine a target evaluation dimension according to the target type; Perform graph data reconstruction on the segmented graph to obtain a reconstructed graph; Generate context information according to the reconstructed graph and the target evaluation dimension; Use a preset large language model to generate a reply for the target problem and the context information to obtain a target answer; the target answer includes target evaluation indicators corresponding to the target evaluation dimension.

2. The method according to claim 1, wherein The performing graph data reconstruction on the segmented graph to obtain a reconstructed graph includes: Obtain the entity features of the initial graph to obtain initial graph features; Obtain the entity relationships of the segmented graph to obtain segmented graph relationships; Through a first graph convolutional layer, perform graph convolutional processing on the initial graph features and the segmented graph relationships to obtain first segmented graph features; Perform feature aggregation processing on the first segmented graph features to obtain first aggregated features; Through a second graph convolutional layer, perform graph convolutional processing on the first aggregated features and the segmented graph relationships to obtain second segmented graph features; Use a decoder to perform decoding processing on the second segmented graph features to obtain the reconstructed graph.

3. The method according to claim 2, wherein The method further includes updating the second segmented graph features, specifically including: Perform feature aggregation processing on the second segmented graph features to obtain second aggregated features; Perform splicing processing on the second aggregated features and the second segmented graph features to obtain spliced graph features, and use the spliced graph features as the second segmented graph features.

4. The method according to claim 1, wherein The determining a target evaluation dimension according to the target type includes: Calculate the semantic similarity between the target type and preset evaluation dimensions to obtain similarity data; Based on the similarity data, filter the preset evaluation dimensions to obtain the target evaluation dimension.

5. The method according to claim 1, characterized in that, The obtaining the relationship types of entity relationships in the initial graph and selecting a target type from the relationship types includes: Obtain the relationship types of entity relationships in the initial graph, calculate the number of entity relationships belonging to the same relationship type to obtain a relationship number; Based on the relationship number, select the target type from the relationship types.

6. The method according to any one of claims 1 to 5, characterized in that The according to the evaluation object, performing subgraph retrieval in a preset knowledge graph to obtain an initial graph includes: According to the evaluation object, perform entity retrieval in the knowledge graph to obtain target entities; Extract a subgraph containing the target entities in the knowledge graph to obtain the initial graph.

7. The method according to any one of claims 1 to 5, characterized in that, The method further includes updating the context information, specifically including: Obtain a preset knowledge vector library; the preset knowledge vector library includes text blocks and text vectors of the text blocks; Vectorize the target problem to obtain a problem vector; Calculate the distance between the problem vector and the text vectors to obtain a vector distance; Select a target vector from the text vectors based on the vector distance; Filter out target blocks from the text blocks based on the target vector; Fill the target blocks into the context information.

8. An apparatus for constructing an evaluation index based on a knowledge graph, characterized in that, The device includes: A problem acquisition module, configured to acquire a target problem, where the target problem includes an evaluation object; A sub-graph retrieval module, configured to perform sub-graph retrieval in a preset knowledge graph according to the evaluation object to obtain an initial graph; A target type selection module, configured to obtain the relationship types of entity relationships in the initial graph and select a target type from the relationship types; A graph segmentation module, configured to perform graph segmentation on the initial graph based on the target type to obtain a segmented graph; the relationship types of entity relationships in the same segmented graph are the same; A target evaluation dimension determination module, configured to determine a target evaluation dimension according to the target type; A graph data reconstruction module, configured to perform graph data reconstruction on the segmented graph to obtain a reconstructed graph; A context information generation module, configured to generate context information according to the reconstructed graph and the target evaluation dimension; A response generation module, configured to generate a target response to the target problem and the context information through a preset large language model to obtain a target answer; the target answer includes target evaluation indicators corresponding to the target evaluation dimension.

9. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory stores a computer program, and when the processor executes the computer program, it implements the knowledge graph-based evaluation index construction method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the knowledge graph-based evaluation index construction method according to any one of claims 1 to 7.

Images