Automobile financial scheme matching method and system based on knowledge graph

Abstract

The application discloses a car financial scheme matching method and system based on a knowledge graph, constructs a domain knowledge graph containing user nodes, vehicle nodes and financial product nodes by performing entity extraction and relation extraction on multi-source heterogeneous data, aggregates neighbor node information by using a graph attention network to generate a user node vector fused with behavior time sequence features, then converts product rules into logical constraint paths, filters out a candidate financial product set in the knowledge graph, and finally calculates a feature matching component of the user and the product, combines a correlation risk coefficient based on a connected path length and a benefit incentive component, weights to obtain a final matching score and outputs a recommendation result. The application can accurately identify user demand and take into account risk control, and improves recommendation accuracy.

Description

Technical Field

[0001] This invention belongs to the fields of financial technology and artificial intelligence technology, and relates to a method and system for matching auto finance solutions based on knowledge graphs. Background Technology

[0002] With the rapid development of the auto finance market, financial institutions have launched a wide variety of financial products. Current auto finance matching primarily relies on manual experience or simple rule-based screening. Existing technologies typically only utilize users' credit reports or basic attribute data for matching. This approach ignores the rich information contained in users' dynamic behavioral data and vehicle lifecycle data.

[0003] Furthermore, auto finance product rules often exist in unstructured text form, making them difficult to directly translate into computer-executable logic. Traditional matching methods cannot effectively handle multi-source heterogeneous data, resulting in low accuracy of recommendation results. Existing technologies also struggle to adapt to frequent changes in financial product rules, lacking dynamic adjustment capabilities.

[0004] Therefore, there is an urgent need for a method that can fuse multimodal data and achieve dynamic matching. Summary of the Invention

[0005] To address the problems existing in the background technology, this invention proposes a knowledge graph-based method and system for matching automotive finance solutions.

[0006] A first aspect of the present invention provides a knowledge graph-based method for matching auto finance solutions, comprising: Acquire multi-source heterogeneous data, including user basic attribute data, user dynamic behavior data, vehicle life cycle data, and auto finance product rule data; Entity and relation extraction are performed on the multi-source heterogeneous data to construct a knowledge graph in the field of auto finance, which includes user nodes, vehicle nodes, financial product nodes and the associated edges connecting each node. The knowledge graph is aggregated using a graph attention network, and the neighbor node information of each node is mapped into a low-dimensional vector to generate a user node vector that integrates relational features and behavioral temporal features. The auto finance product rule data is transformed into logical constraint paths on the knowledge graph. Based on the logical constraint paths, the financial product nodes are traversed and filtered to obtain a set of candidate financial products. Calculate the matching score between the user node vector and each financial product node in the candidate financial product set, and generate a target auto finance solution recommendation instruction based on the matching score.

[0007] Optionally, the step of extracting entities and relationships from the multi-source heterogeneous data to construct a knowledge graph for the automotive finance domain, including user nodes, vehicle nodes, financial product nodes, and connecting edges linking each node, includes: Extract the timestamps and event types from the user dynamic behavior data, and construct a time-series behavior edge between user nodes and vehicle nodes; Extract the admission criteria from the auto finance product rule data and construct attribute constraint edges between the financial product node and the vehicle node or user node. By uniquely encoding and mapping nodes from different data sources that point to the same objective entity, a heterogeneous information network containing structured facts and unstructured semantic labels is formed.

[0008] Optionally, the step of using a graph attention network to aggregate features of the knowledge graph, mapping the neighbor node information of each node into a low-dimensional vector, and generating a user node vector that integrates relational features and behavioral temporal features includes: For a target user node in a knowledge graph, obtain its first-order neighbor node set and corresponding edge features; The relevance weights between the target user node and each of its neighboring nodes are calculated using a preset attention mechanism. The feature vectors of neighboring nodes are weighted and summed according to the relevance weights, and then mapped using a non-linear activation function to update the hidden layer state of the target user node, thereby obtaining the user node vector.

[0009] Optionally, the step of converting the auto finance product rule data into logical constraint paths on the knowledge graph, and traversing and filtering the financial product nodes based on the logical constraint paths, includes: Analyze the terms and conditions of auto finance products to extract hard admission rules, including applicable car models, down payment ratio thresholds, and geographical restrictions; The hard admission rules are mapped to path constraints in graph database query statements; Execute the graph database query statement in the knowledge graph to remove financial product nodes that do not meet the path constraints, and retain the nodes that meet the conditions as the candidate financial product set.

[0010] Optionally, calculating the matching score between the user node vector and each financial product node in the candidate financial product set includes: Calculate the cosine similarity between the user node vector and the financial product node vector to obtain the feature matching component; The association risk coefficient of a user is calculated based on the length of the connection path between the user node and the preset risk node set in the knowledge graph. Obtain the expected return parameters of financial institutions for the financial product nodes, and obtain the return incentive components; The final matching score is obtained by weighting and summing the feature matching components, the associated risk coefficient, and the benefit incentive components.

[0011] Optionally, after constructing the auto finance domain knowledge graph, which includes user nodes, vehicle nodes, financial product nodes, and connecting edges linking the nodes, the method further includes: Monitor changes in external data sources, including new user behavior records or updates to financial product rules; An incremental update subgraph is generated based on the change event, and the incremental update subgraph is integrated into the existing knowledge graph of the automotive finance field to maintain the real-time nature of the graph data.

[0012] Optionally, the user dynamic behavior data includes clickstream data and browsing duration data of users in the auto finance application, as well as test drive trajectory data of offline stores; the vehicle life cycle data includes vehicle mileage, battery health data, and maintenance records.

[0013] A second aspect of the present invention provides a knowledge graph-based auto finance solution matching system, comprising: The data acquisition module is used to acquire multi-source heterogeneous data, including user basic attribute data, user dynamic behavior data, vehicle life cycle data, and auto finance product rule data. The graph construction module is used to extract entities and relationships from the multi-source heterogeneous data and construct a knowledge graph in the field of auto finance that includes user nodes, vehicle nodes, financial product nodes and the associated edges connecting each node. The feature learning module is used to aggregate features of the knowledge graph using a graph attention network, mapping the neighbor node information of each node into a low-dimensional vector, and generating a user node vector that integrates relational features and behavioral temporal features. The rule reasoning module is used to transform the rule data of the auto finance products into logical constraint paths on the knowledge graph, and to traverse and filter the financial product nodes based on the logical constraint paths to obtain a set of candidate financial products. The matching decision module is used to calculate the matching score between the user node vector and each financial product node in the candidate financial product set, and generate a target auto finance solution recommendation instruction based on the matching score.

[0014] Optionally, the feature learning module is specifically used for: For a target user node in a knowledge graph, obtain its first-order neighbor node set and corresponding edge features; The relevance weights between the target user node and each of its neighboring nodes are calculated using a preset attention mechanism. The feature vectors of neighboring nodes are weighted and summed according to the relevance weights, and then mapped using a non-linear activation function to update the hidden layer state of the target user node.

[0015] Optionally, the rule reasoning module is used for: Analyze the terms and conditions of auto finance products to extract the hard admission rules; The hard admission rules are mapped to path constraints in graph database query statements; Execute the graph database query statement in the knowledge graph to remove financial product nodes that do not meet the path constraints, and obtain the candidate financial product set.

[0016] Compared with the prior art, the present invention has the following beneficial effects: This invention discloses a knowledge graph-based method and system for matching auto finance solutions. By constructing a knowledge graph for the auto finance domain, including user nodes, vehicle nodes, financial product nodes, and connecting edges between these nodes, multi-source heterogeneous data is transformed into a structured semantic network. This method breaks through the limitations of data silos in traditional auto finance operations, achieving deep integration of user dynamic behavior data, vehicle lifecycle data, and auto finance product rule data. Through the knowledge graph format, this invention can intuitively express the complex relationships between entities, providing an interpretable data foundation for subsequent intelligent recommendations.

[0017] This invention utilizes graph attention networks to aggregate features from knowledge graphs, enabling adaptive learning of the influence of different neighboring nodes on target user nodes. By calculating relevance weights and performing weighted summation, the generated user node vector not only integrates the user's static attributes but also includes dynamic behavioral temporal features and associated vehicle features. This feature representation method effectively distinguishes between key correlation information and noisy information, allowing the model to more accurately capture users' potential financial needs and significantly improve the accuracy of user profiling.

[0018] This invention transforms auto finance product rule data into logical constraint paths on a knowledge graph, achieving automated and standardized execution of financial product screening rules. By traversing and filtering through graph database queries, the system can quickly eliminate financial product nodes that do not meet the strict admission rules, retaining only those nodes that meet the criteria as a set of candidate financial products. This method avoids omissions and misjudgments that may occur during manual rule review, significantly improving screening efficiency and ensuring that the recommended results strictly comply with the risk control requirements of financial institutions.

[0019] This invention comprehensively considers feature matching components, association risk coefficients, and profit incentive components when calculating matching scores. By introducing the path length between user nodes and a preset set of risk nodes in a knowledge graph, this invention can effectively identify users' implicit association risks. Simultaneously, by incorporating the expected return parameters of financial institutions, this invention balances user experience, risk control, and commercial interests during the recommendation process. This multi-dimensional evaluation mechanism ensures that the ultimately recommended financial products not only meet user needs but also safeguard the asset security and profit goals of financial institutions. Attached Figure Description

[0020] Figure 1 This is a flowchart of a knowledge graph-based auto finance solution matching method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of a knowledge graph-based auto finance solution matching system according to an embodiment of the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] In one embodiment, such as Figure 1 As shown, a knowledge graph-based method for matching auto finance solutions is provided, and this method is applied to... Figure 1 Taking China as an example, the following specific steps will be used: S10: Obtain multi-source heterogeneous data, including user basic attribute data, user dynamic behavior data, vehicle life cycle data, and auto finance product rule data.

[0023] S20: Entity extraction and relation extraction are performed on the multi-source heterogeneous data to construct a knowledge graph in the field of auto finance, which includes user nodes, vehicle nodes, financial product nodes, and related edges connecting each node.

[0024] S30: Use a graph attention network to aggregate features of the knowledge graph, map the neighbor node information of each node into a low-dimensional vector, and generate a user node vector that integrates relational features and behavioral temporal features.

[0025] S40: Transform the auto finance product rule data into logical constraint paths on the knowledge graph, and traverse and filter the financial product nodes based on the logical constraint paths to obtain a set of candidate financial products.

[0026] S50: Calculate the matching score between the user node vector and each financial product node in the candidate financial product set, and generate a target auto finance solution recommendation instruction based on the matching score.

[0027] Specifically, the process of acquiring multi-source heterogeneous data includes obtaining basic user attribute data from internal databases of financial institutions, obtaining dynamic user behavior data from in-vehicle terminals or mobile internet applications, obtaining vehicle lifecycle data from vehicle sensors and maintenance record systems, and obtaining auto finance product rule data from financial product publishing platforms. The basic user attribute data specifically covers age, occupation, income level, and credit score; the dynamic user behavior data specifically includes driving habits, driving trajectory, and frequency of emergency braking; and the vehicle lifecycle data specifically includes battery health, mileage, and component wear.

[0028] When extracting entities and relationships from multi-source heterogeneous data, natural language processing techniques are used to identify entity names in the text and align the same entity from different sources. In constructing a knowledge graph for the automotive finance sector, users, vehicles, and financial products are mapped to nodes in the graph, and interconnected edges are used to connect nodes with interactive relationships, thus forming a network data structure. During feature aggregation of the knowledge graph using a graph attention network, the relevance weights between the target user node and its neighboring nodes are calculated. The features of the neighboring nodes are then weighted and summed based on these weights to generate a user node vector that integrates relational features and behavioral temporal features.

[0029] When transforming auto finance product rule data into logical constraint paths, the hard admission conditions in the product terms are parsed and mapped to path constraints in graph database query statements. Based on these logical constraint paths, financial product nodes are traversed and filtered, eliminating nodes that do not meet the path constraints and retaining those that do to form a candidate financial product set. When calculating the matching score between the user node vector and each node in the candidate financial product set, a vector similarity algorithm is used to calculate the feature matching component, which is then weighted and summed using the association risk coefficient and return incentive component. Finally, a target auto finance solution recommendation instruction is generated based on the matching score. This embodiment can deeply integrate multimodal data, improving the accuracy and timeliness of financial solution matching.

[0030] In one embodiment, step S10, after constructing the auto finance domain knowledge graph including user nodes, vehicle nodes, financial product nodes, and the associated edges connecting each node, further includes the following steps: S11: Listen for change events from external data sources, including user-added behavior records or financial product rule updates.

[0031] S12: Generate an incremental update subgraph based on the change event, and integrate the incremental update subgraph into the existing knowledge graph of the automotive finance field to maintain the real-time nature of the graph data.

[0032] Specifically, the monitoring of change events from external data sources is achieved by deploying a data capture connector. This connector continuously monitors the transaction logs of the business database, which record all write operations. When new driving behavior data is detected from a user or an update to the loan product's eligibility rules by a financial institution, the connector captures this operation record and generates a change event message. This change event message includes the operation type, timestamp, and the specific data content after the change.

[0033] The process of generating an incremental update subgraph based on the change event first involves parsing the change event message to identify the affected target entities. If the change event is a user adding a behavior record, the feature vector corresponding to the behavior is extracted, and a temporal behavior edge is constructed between the user node and the corresponding behavior node. If the change event is a financial product rule update, the new rule text is parsed, and the attribute constraint edges between the financial product node and the vehicle node or user node are updated. The newly generated nodes and edges are then combined to form the incremental update subgraph.

[0034] When integrating the incremental update subgraph into the existing automotive finance knowledge graph, a transaction processing mechanism for the graph database is employed. The system first locates the node to be updated in the existing automotive finance knowledge graph based on the entity's unique identifier, and locks the node using a distributed lock to prevent concurrent conflicts. Subsequently, the new edges in the incremental update subgraph are written to the graph database, and the attribute information of the relevant nodes is updated. After the writing is complete, the distributed lock is released, and the transaction is committed, thereby completing the real-time synchronization of the graph data.

[0035] This invention avoids the enormous computational overhead of fully rebuilding the knowledge graph by monitoring change events and generating incrementally updated subgraphs. This method can perceive changes in user behavior and adjustments to product rules in real time, ensuring that the information stored in the knowledge graph remains consistent with the current business situation. This significantly improves the timeliness of subsequent financial solution matching, enabling the system to output recommendation results based on the latest data.

[0036] It is important to note that the user dynamic behavior data specifically encompasses clickstream data, browsing duration data, and test drive trajectory data from offline stores within the auto finance application. The clickstream data records the sequence of user actions within the application interface, including the specific functional modules clicked, page navigation paths, and operation time intervals. The browsing duration data reflects the time a user spends on specific financial product details pages or vehicle information pages, representing the user's level of interest. The test drive trajectory data from offline stores is collected through in-vehicle terminals and includes vehicle location coordinates, driving speed, and acceleration changes during the test drive.

[0037] The vehicle's full lifecycle data specifically includes vehicle mileage, battery health data, and maintenance records. Vehicle mileage refers to the total distance the vehicle has traveled since its manufacture, and this data is periodically uploaded to the server via the on-board diagnostic system. Battery health data, specific to new energy vehicles, includes the battery pack's remaining capacity, charge / discharge cycle count, and internal resistance variation curve. Maintenance records cover the dates of all maintenance performed at authorized service centers, the names of replaced parts, and detailed descriptions of the repairs performed.

[0038] In one embodiment, step S20, namely, performing entity extraction and relation extraction on the multi-source heterogeneous data to construct a knowledge graph for the automotive finance domain that includes user nodes, vehicle nodes, financial product nodes, and associated edges connecting each node, further includes the following steps: S21: Extract the timestamps and event types from the user dynamic behavior data, and construct a time-series behavior edge between user nodes and vehicle nodes.

[0039] S22: Extract the admission conditions from the auto finance product rule data and construct attribute constraint edges between the financial product node and the vehicle node or user node.

[0040] S23: Uniquely encode and map nodes pointing to the same objective entity from different data sources to form a heterogeneous information network containing structured facts and unstructured semantic labels.

[0041] Specifically, the process of entity extraction and relation extraction from the multi-source heterogeneous data includes parsing the user dynamic behavior data. The system reads the timestamp and event type fields from the data, where the event type includes rapid acceleration, rapid deceleration, or prolonged idling. Based on the timestamps, the chronological order of the actions is determined, and a temporal behavior edge is established between the user node and the corresponding vehicle node. The direction of the temporal behavior edge is from the user node to the vehicle node, and the edge's attributes store the specific time and type label of the action.

[0042] When extracting the admission criteria from the auto finance product rule data, natural language processing technology is used to identify key constraint information in the text. These admission criteria include applicable vehicle models, geographical restrictions, or age restrictions. Based on the identified constraint information, attribute constraint edges are constructed between the finance product node and eligible vehicle or user nodes. These attribute constraint edges characterize the finance product's admission requirements for specific entity nodes, thereby transforming unstructured text rules into structured connections in a graph.

[0043] When uniquely encoding and mapping nodes pointing to the same objective entity from different data sources, an entity alignment algorithm is used. The system calculates the similarity of node attributes in different data sources and determines nodes with similarity higher than a preset threshold as the same objective entity. A unified, globally unique identifier is assigned to these nodes, and structured facts and unstructured semantic tags from different data sources are merged into this identifier, thereby forming a heterogeneous information network containing multimodal information.

[0044] The technical advantage of this embodiment lies in that, by constructing temporal behavioral edges, the knowledge graph can accurately record the temporal evolution characteristics of user-vehicle interactions. By constructing attribute constraint edges, a direct association between financial product rules and graph entities is achieved, facilitating efficient logical reasoning and filtering in the future. The heterogeneous information network formed by the unique encoding mapping effectively solves the problem of multi-source data silos, improving the integrity and consistency of data in the knowledge graph.

[0045] In one embodiment, step S30, which involves using a graph attention network to aggregate features of the knowledge graph, mapping the neighbor node information of each node into a low-dimensional vector, and generating a user node vector that integrates relational features and behavioral temporal features, further includes the following steps: S31: For a target user node in the knowledge graph, obtain its first-order neighbor node set and corresponding edge features.

[0046] S32: Calculate the correlation weights between the target user node and each neighbor node using a preset attention mechanism.

[0047] S33: The feature vectors of neighboring nodes are weighted and summed according to the relevance weights, and mapped through a non-linear activation function to update the hidden layer state of the target user node, thereby obtaining the user node vector.

[0048] Specifically, the process of using a graph attention network to aggregate features in the knowledge graph is as follows: First, for a target user node in the knowledge graph, the system traverses its directly connected edges to obtain a set of first-order neighbor nodes and their corresponding edge features. The set of first-order neighbor nodes includes vehicle nodes associated with the user, nodes of financial products viewed, and nodes representing the user's region. The edge features include the time interval in temporal behavior edges and the constraint type in attribute constraint edges.

[0049] When calculating the relevance weights between the target user node and its neighboring nodes using a pre-defined attention mechanism, the system concatenates the current feature vector of the target user node with the feature vectors of its neighboring nodes. A pre-defined linear transformation matrix is ​​used to map the concatenated vectors, and an attention coefficient is calculated using an activation function. Subsequently, the attention coefficients are normalized using the Softmax function to obtain the relevance weight of each neighboring node relative to the target user node. This weight reflects the importance of different neighboring information in assessing the user's financial creditworthiness.

[0050] When weighting and summing the feature vectors of neighboring nodes according to the relevance weights, the feature vector of each neighboring node is multiplied by its corresponding relevance weight, and all products are summed to obtain the aggregated feature. The aggregated feature is then mapped using a non-linear activation function to update the hidden layer state of the target user node, ultimately obtaining the user node vector. The user node vector integrates the user's static attributes, dynamic behavioral time series, and associated vehicle and product features.

[0051] The technical advantage of this embodiment lies in its ability to adaptively learn the influence of different neighboring nodes on the target user node using a graph attention network. This method effectively distinguishes between key association information and noise information, enabling the generated user node vector to more accurately represent the user's overall credit status and potential needs, thereby improving the accuracy of subsequent matching calculations.

[0052] In one embodiment, step S40, which involves converting the auto finance product rule data into logical constraint paths on the knowledge graph and traversing and filtering the financial product nodes based on the logical constraint paths, further includes the following steps: S41: Analyze the text terms of auto finance products and extract hard access rules, including applicable vehicle models, down payment ratio thresholds, and geographical restrictions.

[0053] S42: Map the hard admission rules to path constraints in the graph database query statement.

[0054] S43: Execute the graph database query statement in the knowledge graph, remove financial product nodes that do not meet the path constraint conditions, and retain the nodes that meet the conditions as the candidate financial product set.

[0055] Specifically, the process of transforming the auto finance product rule data into logical constraint paths on the knowledge graph first involves parsing the textual terms of the auto finance products. The system uses natural language processing technology to perform semantic analysis on the unstructured product manuals, identifying and extracting hard access rules. These hard access rules specifically include applicable vehicle models, down payment thresholds, and geographical restrictions. For example, for a specific loan product, the system extracts that its applicable vehicle model is a new energy sedan, its down payment threshold is 20%, and its geographical restriction is limited to first-tier cities.

[0056] When mapping the hard admission rules to path constraints in graph database query statements, the system establishes a mapping relationship between rule elements and graph patterns. The applicable vehicle models are mapped to type matching constraints between financial product nodes and vehicle nodes; the down payment ratio threshold is mapped to a numerical range filtering condition for user node attribute values; and the geographical restrictions are mapped to an existence check of associations between user nodes and geographical nodes. The system uses graph query language to construct logical expressions containing the above constraints, forming the logical constraint paths.

[0057] When executing the graph database query statement in the knowledge graph, the system starts with all financial product nodes and traverses along the logical constraint path. During the traversal, the system checks layer by layer whether the node attributes and edge relationships satisfy the path constraint conditions. For financial product nodes that do not satisfy the path constraint conditions, the system removes them from the result set. For nodes that pass all constraint checks, the system retains them and includes them in the candidate financial product set.

[0058] The technical advantage of this embodiment lies in its automation and standardization of financial product screening rules by transforming textual terms into executable logical constraint paths. This method avoids omissions and misjudgments that may occur during manual rule review, significantly improving screening efficiency. Furthermore, the knowledge graph-based traversal screening fully utilizes the relationships between nodes, ensuring that the selected set of candidate financial products accurately matches the user's hard criteria, laying a solid foundation for subsequent personalized recommendations.

[0059] In one embodiment, step S50, namely calculating the matching score between the user node vector and each financial product node in the candidate financial product set, further includes the following steps: S51: Calculate the cosine similarity between the user node vector and the financial product node vector to obtain the feature matching component.

[0060] S52: Calculate the association risk coefficient of the user based on the length of the connection path between the user node and the preset risk node set in the knowledge graph.

[0061] S53: Obtain the expected return parameters of the financial institution for the financial product node, and obtain the return incentive component.

[0062] S54: The feature matching component, the associated risk coefficient, and the benefit incentive component are weighted and summed to obtain the final matching score.

[0063] Specifically, the process of calculating the matching score between the user node vector and each financial product node in the candidate financial product set first involves calculating the cosine similarity between the user node vector and the financial product node vector. The system extracts the user node vector and the financial product node vector from the vector space and calculates the cosine value of the angle between the two vectors in the multidimensional space. The cosine value of the angle serves as a feature matching component, representing the degree of semantic fit between the user's potential needs and the attributes of the financial product. When the cosine value of the angle approaches 1, it indicates that the user features and product features are highly similar.

[0064] Based on the length of the connection path between the user node and a preset set of risk nodes in the knowledge graph, the system calculates the user's association risk coefficient. The preset set of risk nodes includes nodes with historical default records, nodes representing high-risk industries, and nodes related to legal proceedings. The system searches the knowledge graph for the shortest path from the user node to any node in the preset set of risk nodes. If the connection path length is short, it indicates a strong association between the user and the risk entity, and the system assigns a higher association risk coefficient. If the connection path length is long or there is no connection path, it indicates a lower user risk, and the system assigns a lower association risk coefficient.

[0065] The system obtains the expected return parameters of financial institutions for the aforementioned financial product nodes, deriving a return incentive component. These expected return parameters include the product interest rate, service fee rate, and estimated cross-selling probability. The system determines the profitability of different financial product nodes based on historical business data, and normalizes this profitability to obtain the return incentive component. This component reflects the commercial value that recommending specific financial products brings to financial institutions.

[0066] When performing a weighted summation of the feature matching component, the association risk coefficient, and the benefit incentive component, the system invokes a preset weight configuration table. This weight configuration table defines the importance ratio of each component in the final decision. The system multiplies the feature matching component by a first weight, the reciprocal of the association risk coefficient by a second weight, and the benefit incentive component by a third weight. The three products are then added together to obtain the final matching score.

[0067] The technical advantage of this embodiment lies in its ability to achieve a comprehensive evaluation from three dimensions: user experience, risk control, and business benefits, by fusing feature matching components, association risk coefficients, and profit incentive components. This method not only considers the semantic similarity between users and products but also incorporates multi-hop association risk information from knowledge graphs, effectively identifying users with latent risks. Simultaneously, the introduction of profit incentive components ensures that the recommendation results align with the financial institution's operational objectives, thereby maximizing business revenue while maintaining controllable risk.

[0068] In one embodiment, such as Figure 2 As shown, a knowledge graph-based auto finance solution matching system is provided. This system corresponds one-to-one with the knowledge graph-based auto finance solution matching method in the above embodiments. The knowledge graph-based auto finance solution matching system includes: a data acquisition module, a graph construction module, a feature learning module, a rule reasoning module, and a matching decision module. Detailed descriptions of each functional module are as follows: The data acquisition module is used to acquire multi-source heterogeneous data, including user basic attribute data, user dynamic behavior data, vehicle life cycle data, and auto finance product rule data. The graph construction module is used to extract entities and relationships from the multi-source heterogeneous data and construct a knowledge graph in the field of auto finance that includes user nodes, vehicle nodes, financial product nodes and the associated edges connecting each node. The feature learning module is used to aggregate features of the knowledge graph using a graph attention network, mapping the neighbor node information of each node into a low-dimensional vector, and generating a user node vector that integrates relational features and behavioral temporal features. The rule reasoning module is used to transform the rule data of the auto finance products into logical constraint paths on the knowledge graph, and to traverse and filter the financial product nodes based on the logical constraint paths to obtain a set of candidate financial products. The matching decision module is used to calculate the matching score between the user node vector and each financial product node in the candidate financial product set, and generate a target auto finance solution recommendation instruction based on the matching score.

[0069] Furthermore, the feature learning module is specifically used for: For a target user node in a knowledge graph, obtain its first-order neighbor node set and corresponding edge features; The relevance weights between the target user node and each of its neighboring nodes are calculated using a preset attention mechanism. The feature vectors of neighboring nodes are weighted and summed according to the relevance weights, and then mapped using a non-linear activation function to update the hidden layer state of the target user node.

[0070] Furthermore, the rule reasoning module is used for: Analyze the terms and conditions of auto finance products to extract the hard admission rules; The hard admission rules are mapped to path constraints in graph database query statements; Execute the graph database query statement in the knowledge graph to remove financial product nodes that do not meet the path constraints, and obtain the candidate financial product set.

[0071] Specific limitations regarding the knowledge graph-based auto finance solution matching system can be found in the limitations of the knowledge graph-based auto finance solution matching method described above, and will not be repeated here. Each module in the aforementioned knowledge graph-based auto finance solution matching system can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0072] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above.

[0073] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A knowledge graph-based method for matching auto finance solutions, characterized in that, include: Acquire multi-source heterogeneous data, including user basic attribute data, user dynamic behavior data, vehicle life cycle data, and auto finance product rule data; Entity and relation extraction are performed on the multi-source heterogeneous data to construct a knowledge graph in the field of auto finance, which includes user nodes, vehicle nodes, financial product nodes and the associated edges connecting each node. The knowledge graph is aggregated using a graph attention network, and the neighbor node information of each node is mapped into a low-dimensional vector to generate a user node vector that integrates relational features and behavioral temporal features. The auto finance product rule data is transformed into logical constraint paths on the knowledge graph. Based on the logical constraint paths, the financial product nodes are traversed and filtered to obtain a set of candidate financial products. Calculate the matching score between the user node vector and each financial product node in the candidate financial product set, and generate a target auto finance solution recommendation instruction based on the matching score.

2. The knowledge graph-based auto finance solution matching method according to claim 1, characterized in that, The step of extracting entities and relationships from the multi-source heterogeneous data to construct a knowledge graph for the automotive finance domain, including user nodes, vehicle nodes, financial product nodes, and connecting edges between these nodes, includes: Extract the timestamps and event types from the user dynamic behavior data, and construct a time-series behavior edge between user nodes and vehicle nodes; Extract the admission criteria from the auto finance product rule data and construct attribute constraint edges between the financial product node and the vehicle node or user node. By uniquely encoding and mapping nodes from different data sources that point to the same objective entity, a heterogeneous information network containing structured facts and unstructured semantic labels is formed.

3. The knowledge graph-based auto finance solution matching method according to claim 1, characterized in that, The step of using a graph attention network to aggregate features of the knowledge graph, mapping the neighbor node information of each node into a low-dimensional vector, and generating a user node vector that integrates relational features and behavioral temporal features includes: For a target user node in a knowledge graph, obtain its first-order neighbor node set and corresponding edge features; The relevance weights between the target user node and each neighbor node are calculated using a preset attention mechanism. The feature vectors of neighboring nodes are weighted and summed according to the relevance weights, and then mapped using a non-linear activation function to update the hidden layer state of the target user node, thereby obtaining the user node vector.

4. The knowledge graph-based auto finance solution matching method according to claim 1, characterized in that, The step of converting the auto finance product rule data into logical constraint paths on the knowledge graph, and traversing and filtering the financial product nodes based on the logical constraint paths, includes: Analyze the terms and conditions of auto finance products to extract hard admission rules, including applicable car models, down payment ratio thresholds, and geographical restrictions; The hard admission rules are mapped to path constraints in graph database query statements; Execute the graph database query statement in the knowledge graph to remove financial product nodes that do not meet the path constraints, and retain the nodes that meet the conditions as the candidate financial product set.

5. The knowledge graph-based auto finance solution matching method according to claim 1, characterized in that, The calculation of the matching score between the user node vector and each financial product node in the candidate financial product set includes: Calculate the cosine similarity between the user node vector and the financial product node vector to obtain the feature matching component; The association risk coefficient of a user is calculated based on the length of the connection path between the user node and the preset risk node set in the knowledge graph. Obtain the expected return parameters of financial institutions for the financial product nodes, and obtain the return incentive components; The final matching score is obtained by weighting and summing the feature matching components, the associated risk coefficient, and the benefit incentive components.

6. The knowledge graph-based auto finance solution matching method according to claim 1, characterized in that, Following the construction of the auto finance domain knowledge graph, which includes user nodes, vehicle nodes, financial product nodes, and connecting edges linking these nodes, the following is also included: Monitor changes in external data sources, including new user behavior records or updates to financial product rules; An incremental update subgraph is generated based on the change event, and the incremental update subgraph is integrated into the existing knowledge graph of the automotive finance field to maintain the real-time nature of the graph data.

7. The knowledge graph-based auto finance solution matching method according to claim 1, characterized in that, The user dynamic behavior data includes clickstream data and browsing time data in the auto finance application, as well as test drive trajectory data at offline stores; the vehicle life cycle data includes vehicle mileage, battery health data, and maintenance records.

8. A knowledge graph-based auto finance solution matching system, used to implement the method described in any one of claims 1 to 7, characterized in that, include: The data acquisition module is used to acquire multi-source heterogeneous data, including user basic attribute data, user dynamic behavior data, vehicle life cycle data, and auto finance product rule data. The graph construction module is used to extract entities and relationships from the multi-source heterogeneous data and construct a knowledge graph in the field of auto finance that includes user nodes, vehicle nodes, financial product nodes and the associated edges connecting each node. The feature learning module is used to aggregate features of the knowledge graph using a graph attention network, mapping the neighbor node information of each node into a low-dimensional vector, and generating a user node vector that integrates relational features and behavioral temporal features. The rule reasoning module is used to transform the rule data of the auto finance products into logical constraint paths on the knowledge graph, and to traverse and filter the financial product nodes based on the logical constraint paths to obtain a set of candidate financial products. The matching decision module is used to calculate the matching score between the user node vector and each financial product node in the candidate financial product set, and generate a target auto finance solution recommendation instruction based on the matching score.

9. The knowledge graph-based auto finance solution matching system according to claim 8, characterized in that, The feature learning module is specifically used for: For a target user node in a knowledge graph, obtain its first-order neighbor node set and corresponding edge features; The relevance weights between the target user node and each neighbor node are calculated using a preset attention mechanism. The feature vectors of neighboring nodes are weighted and summed according to the relevance weights, and then mapped using a non-linear activation function to update the hidden layer state of the target user node.

10. The knowledge graph-based auto finance solution matching system according to claim 8, characterized in that, The rule reasoning module is used for: Analyze the terms and conditions of auto finance products to extract the hard admission rules; The hard admission rules are mapped to path constraints in graph database query statements; Execute the graph database query statement in the knowledge graph to remove financial product nodes that do not meet the path constraints, and obtain the candidate financial product set.

Images