Risk assessment method, device, equipment and medium

Abstract

This invention relates to the field of artificial intelligence and discloses a risk assessment method, apparatus, device, and medium, comprising: acquiring initial business interaction data of the assessment object and performing semantic parsing to determine a first entity set; using a dynamic knowledge graph to determine the first entity node of the first entity set, acquiring the associated nodes of the first entity node, generating a dynamic information collection sequence, and sequentially acquiring feedback data for the dynamic information collection sequence; extracting a second entity set from the feedback data and comparing the second entity set with the first entity set; using a dynamic knowledge graph to determine the second entity node corresponding to the second entity set and determining the transmission path between the first and second entity nodes; and performing chain reasoning based on the transmission path and preset risk nodes in the dynamic knowledge graph to generate risk assessment results and reasoning path data. This method can be applied to fintech and smart healthcare business scenarios, and can improve the accuracy and reliability of underwriting risk assessment.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a risk assessment method, apparatus, device, and medium. Background Technology

[0002] Currently, insurance underwriting primarily relies on manually maintained static rule systems to complete risk assessments. The industry generally uses standardized questionnaires to collect user health information and conduct preliminary risk identification. However, this collection model has inherent limitations. On the one hand, it can only obtain single-structured questionnaire data, failing to effectively integrate multimodal health data such as physical examination reports, medical test indicators, and monitoring data from smart wearable devices, resulting in extremely limited information collection dimensions. On the other hand, the fixed questions cannot dynamically change their content based on user responses, leading to fragmented and unrelated health information that is difficult to fully reconstruct the user's true health status and cannot provide comprehensive and coherent data support for complex medical risk reasoning.

[0003] Constrained by such fragmented and low-correlation data, the reasoning ability of existing underwriting algorithm models is severely limited. Current mainstream shallow machine learning algorithms, such as decision trees and logistic regression, are only suitable for single, independent risk indicator judgment scenarios. They cannot uncover the complex logical chains of inherent relationships and progressive evolution between different risk factors, making it difficult to complete systematic and interconnected risk assessments.

[0004] In summary, existing underwriting technologies suffer from fragmented and unrelated assessment data due to their single data collection dimension and rigid inquiry mechanisms. This results in a lack of ability to logically deduce complex risk chains within the overall assessment system, ultimately leading to low accuracy and reliability in underwriting risk assessment. To address these technical pain points, there is an urgent need for an intelligent assessment solution that can integrate multimodal data and support logical deduction based on complex risk chains. Summary of the Invention

[0005] This invention provides an artificial intelligence-based risk assessment method, apparatus, computer equipment, and medium, aiming to solve the problem that existing assessment systems cannot integrate multimodal data and support complex risk logic chain progressive reasoning, resulting in low accuracy and poor reliability of underwriting risk assessment.

[0006] Firstly, a risk assessment method is provided, including: Obtain the initial business interaction data of the evaluation object, perform semantic parsing on the initial business interaction data, and determine the first entity set; The first entity node corresponding to the first entity set is determined by using a pre-constructed dynamic knowledge graph, at least one associated node of the first entity node is obtained, a dynamic information collection sequence for the evaluation object is generated based on at least one associated node, and feedback data of the evaluation object for the dynamic information collection sequence is obtained in sequence. Extract the second entity set from the feedback data and perform a consistency comparison between the second entity set and the first entity set; If the comparison passes, a dynamic knowledge graph is used to determine the second entity node corresponding to the second entity set, and to determine the transmission path between the first entity node and the second entity node. Chain reasoning is performed based on the transmission path and preset risk nodes in the dynamic knowledge graph to generate risk assessment results for the assessed object and corresponding reasoning path data.

[0007] Secondly, a risk assessment device is provided, comprising: The data parsing module is used to obtain the initial business interaction data of the evaluation object, perform semantic parsing on the initial business interaction data, and determine the first entity set; The information acquisition module is used to determine the first entity node corresponding to the first entity set using a pre-built dynamic knowledge graph, obtain at least one associated node of the first entity node, generate a dynamic information acquisition sequence for the evaluation object based on at least one associated node, and obtain feedback data of the evaluation object for the dynamic information acquisition sequence in sequence. The entity comparison module is used to extract a second entity set from the feedback data and perform a consistency comparison between the second entity set and the first entity set. The path determination module is used to determine the second entity node corresponding to the second entity set using a dynamic knowledge graph if the comparison passes, and to determine the transmission path between the first entity node and the second entity node. The risk assessment module is used to perform chain reasoning based on the transmission path and preset risk nodes in the dynamic knowledge graph, and generate the risk assessment results of the assessed object and the corresponding reasoning path data.

[0008] Thirdly, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the aforementioned risk assessment method.

[0009] Fourthly, a computer-readable storage medium is provided, which stores a computer program that, when executed by a processor, implements the steps of the aforementioned risk assessment method.

[0010] The aforementioned risk assessment method, apparatus, computer equipment, and storage medium enable a solution that obtains initial business interaction data of the assessment object and performs semantic parsing on this data to obtain a first entity set. This solution is compatible with multimodal data and achieves structured purification, enabling the fusion and standardization of multi-source data. By employing a pre-constructed dynamic knowledge graph to determine the first entity node corresponding to the first entity set, at least one associated node of the first entity node is obtained, and a dynamic information collection sequence for the assessment object is generated based on this sequence. Feedback data from the assessment object to this dynamic information collection sequence is then sequentially acquired. This solution leverages the dynamic knowledge graph to generate personalized dynamic information collection sequences based on the initial entity nodes, enabling adaptive expansion of related queries based on user feedback data. This ensures that the collected user information has logical coherence, thoroughly improving the problem of data fragmentation and lack of... The system addresses the issue of interconnectedness; subsequently, by extracting a second entity set from the feedback data and comparing it with the first entity set for consistency verification, the system verifies the authenticity and accuracy of the collected information. A dynamic knowledge graph is used to determine the second entity node corresponding to the second entity set and the transmission path between the first and second entity nodes. Based on this transmission path and preset risk nodes in the dynamic knowledge graph, chain-like reasoning is performed to generate the risk assessment results for the assessed object and the corresponding reasoning path data. This system leverages the dynamic knowledge graph to organize risk transmission paths and combines preset risk nodes to complete chain-like progressive risk reasoning, overcoming the shortcomings of traditional isolated judgments of single risk indicators. It supports in-depth mining of the inherent connections and progressive evolution logic of health risks, accurately identifies potential associated risks, and effectively improves the accuracy and reliability of insurance underwriting risk assessment in complex scenarios. Attached Figure Description

[0011] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a schematic diagram of an application environment for the risk assessment method in one embodiment of the present invention; Figure 2 This is a flowchart illustrating a risk assessment method according to one embodiment of the present invention; Figure 3 This is a flowchart illustrating the process prior to step S20 of the risk assessment method in one embodiment of the present invention; Figure 4 This is a flowchart illustrating step S30 of the risk assessment method in one embodiment of the present invention; Figure 5This is a flowchart illustrating step S60 of the risk assessment method in one embodiment of the present invention; Figure 6 This is a schematic diagram of the risk assessment device in one embodiment of the present invention; Figure 7 This is a schematic diagram of the structure of a computer device according to an embodiment of the present invention; Figure 8 This is another structural schematic diagram of a computer device according to one embodiment of the present invention. Detailed Implementation

[0013] 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, not all, of the embodiments of the present invention. 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.

[0014] The risk assessment method provided in this invention can be applied to, for example... Figure 1In this application environment, one or more clients communicate with the server via a network. The server can obtain the initial business interaction data of the evaluation object, and perform semantic parsing on the initial business interaction data to obtain the first entity set. This allows for compatibility with multimodal data and completion of structured purification, achieving multi-source data fusion and regularization. By using a pre-built dynamic knowledge graph, the server determines the first entity node corresponding to the first entity set, obtains at least one associated node of the first entity node, and generates a dynamic information collection sequence for the evaluation object based on at least one associated node. The server then sequentially obtains the feedback data of the evaluation object for the dynamic information collection sequence. It can generate personalized dynamic information collection sequences based on the initial entity nodes using the dynamic knowledge graph, enabling adaptive expansion of related queries based on user feedback data. This ensures that the collected user information has logical correlation, thoroughly improving the problems of data fragmentation and lack of linkage and correlation. Subsequently, by... The feedback data extracts a second entity set and compares it with the first entity set for consistency verification, thus confirming the authenticity and accuracy of the collected information. A dynamic knowledge graph is used to determine the second entity node corresponding to the second entity set and the transmission path between the first and second entity nodes. Based on this transmission path and preset risk nodes in the dynamic knowledge graph, chain-like reasoning is performed to generate the risk assessment result of the assessed object and the corresponding reasoning path data. This approach leverages the dynamic knowledge graph to organize risk transmission paths and combines preset risk nodes to complete chain-like progressive risk reasoning, overcoming the shortcomings of traditional isolated judgments of single risk indicators. It supports in-depth mining of the inherent correlation and progressive evolution logic of health risks, accurately identifying potential related risks, and effectively improving the accuracy and reliability of insurance underwriting risk assessment in complex scenarios. The client can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. The server can be implemented using a standalone server or a server cluster consisting of multiple servers. The invention will be described in detail below through specific embodiments.

[0015] Please see Figure 2 As shown, Figure 2 A flowchart illustrating the risk assessment method provided in this embodiment of the invention includes the following steps S20-S60: S20: Obtain the initial business interaction data of the evaluation object, perform semantic parsing on the initial business interaction data, and determine the first entity set; The risk assessment method provided by this invention is applicable to intelligent risk assessment engines such as intelligent customer service and intelligent assistants in various application scenarios. This processing engine is implemented through a server-side component, which can acquire the initial business interaction data of the assessment object in real time. This initial business interaction data is collected by the front-end interaction layer and includes structured question-and-answer data filled in by the assessment object and text of the data to be assessed. This text can include multimodal interaction information such as medical examination reports, self-reported physical symptoms, and descriptions of health conditions. For the acquired initial business interaction data, a semantic reasoning engine is invoked to perform semantic parsing. Information extraction is performed on the structured question-and-answer data and the text of the data to be assessed using NLP technology, identifying and extracting key core entities such as symptoms, diseases, examination indicators, and lifestyle habits. All extracted entities are summarized and organized into a first entity set, where the entity types within the first entity set are consistent with the entity layer categories preset in the dynamic knowledge graph.

[0016] In some embodiments, such as Figure 3 As shown, before step S20 above, that is, before obtaining the initial business interaction data of the evaluation object, performing semantic parsing on the initial business interaction data, and determining the first entity set, the following steps are also included: S11: Acquire multi-source heterogeneous domain knowledge data and historical business interaction data; S12: Extract business entities and the relationships between business entities from domain knowledge data and historical business interaction data to construct an initial knowledge graph; S13: When new business data is detected, the new business data is represented by a graph neural network to update the embedding vectors of business entities and the weight attributes of relationships in the initial knowledge graph, thereby obtaining a dynamic knowledge graph.

[0017] For steps S11-S13, before obtaining the initial business interaction data of the evaluation object, the following steps are also included: Acquire multi-source heterogeneous domain knowledge data and historical business interaction data. The multi-source heterogeneous domain knowledge data includes professional domain data such as medical literature, underwriting rules and terms, disease diagnosis and treatment standards, and industry risk rating data. Historical business interaction data includes underwriting business data such as past customer health records, physical examination reports, wearable device monitoring data, and historical underwriting records and feedback data, covering both structured and unstructured text data.

[0018] An initial knowledge graph is constructed by extracting business entities and their relationships from domain knowledge data and historical business interaction data. Business entities are key elements identified from multi-source data, including categories such as diseases, symptoms, examination indicators, drugs, underwriting risk levels, and insurance rules. Relationships are business and medical connections between entities, including complication associations, disease evolution associations, indicator-to-disease associations, risk level mappings, and treatment contraindication associations. The initial knowledge graph is a basic network architecture built upon business entities and their relationships, comprising an entity layer, a relationship layer, and an attribute layer. The entity layer contains the various key elements identified, including entity categories such as diseases, symptoms, examination indicators, drugs, insurance rules, and risk levels. The relationship layer defines the medical and business association logic between entities, used to define entity linkages such as complication associations, treatment contraindications, disease evolution, risk level mappings, and indicator associations. The attribute layer contains feature parameters associated with each business entity and its relationships, including attributes such as medical evidence level, data source timeliness, risk weight, and association strength.

[0019] The system monitors new business data in real time. This new business data can include newly released medical guidelines, updated underwriting policies, new customer underwriting feedback data, and new health monitoring data, among other iterative data. When new business data is detected, a graph neural network is used to mine its features, extracting various characteristics such as entity latent features, semantic features of associations, and risk evolution features. The business entities contained in the new business data are embedded into the network topology of an initial knowledge graph. Based on the various feature representations learned by the graph neural network, the embedding vectors of corresponding business entities within the initial knowledge graph are iteratively updated one by one. These embedding vectors represent the digital features of the business entities. Simultaneously, the weight attributes of the relationships between entities within the initial knowledge graph are updated in conjunction with the co-occurrence frequency of entity associations in the new business data, business update rules, and the strength of medical evidence. Through bidirectional incremental iterative updates of the business entity embedding vectors and the weight attributes of the relationships, the feature parameters and association strengths in the initial knowledge graph are adjusted, resulting in a dynamic knowledge graph.

[0020] As can be seen from the above scheme, the entire process, by collecting multi-source heterogeneous domain data and historical business data, can enrich the data source dimensions of the knowledge graph and improve the underwriting knowledge system. By extracting business entities and their relationships to build an initial knowledge graph, a standardized underwriting entity relationship network can be constructed. Through graph neural networks, representation learning of new business data and incremental updates to the graph parameters enable dynamic iterative optimization of the knowledge graph. By updating entity embedding vectors and relationship weights in real time, the knowledge graph adapts to the latest medical standards and underwriting business scenarios, ensuring the accuracy and timeliness of risk reasoning.

[0021] Furthermore, step S13 above, that is, when new business data is detected, performs representation learning on the new business data through a graph neural network, updates the embedding vectors of business entities and the weight attributes of relationships in the initial knowledge graph, and obtains a dynamic knowledge graph, including the following steps: S131: When new business data is detected, the new business data will be converted into incremental graph structure data; S132: Input the incremental graph structure data into a pre-trained graph neural network model, and use the aggregation function in the graph neural network model to aggregate the neighbor node features of each node in the incremental graph structure data; S133: Update the embedding vector of the corresponding business entity in the initial knowledge graph based on the aggregated neighbor node features, and update the weight attribute of the association relationship based on the co-occurrence frequency in the newly added business data to obtain the dynamic knowledge graph.

[0022] For steps S131-S133, when new business data is detected, the new business data is transformed into incremental graph structure data according to a preset input format. The incremental graph structure data is structured graph data adapted to the input format of a graph neural network. It is a new subgraph data obtained by transforming scattered and heterogeneous new business data into entity and relation structured data, and is used to participate in the incremental iterative learning of the knowledge graph.

[0023] Incremental graph structure data is input into a pre-trained graph neural network model. The model's built-in aggregation function is then invoked to perform neighbor node feature aggregation operations on all nodes within the incremental graph structure data. The aggregation function, the core computational unit of the graph neural network model, is used to traverse all adjacent nodes connected to a single target node in the incremental subgraph. Neighbor node features consist of basic feature information such as the entity attributes, initial feature vectors, and association attributes of the adjacent nodes in the incremental graph structure data. During the aggregation process, the features of all neighbor nodes corresponding to a single node are integrated, fused, and normalized to summarize the discrete local neighbor features into global node features.

[0024] The embedding vectors of corresponding business entities in the initial knowledge graph are updated based on the aggregated neighbor node features. Specifically, the aggregated global node features are used as the basis for updating, and the original embedding vectors of the matched business entities in the initial knowledge graph are iteratively replaced to complete the update of entity features. The weight attributes of association relationships are updated based on the co-occurrence frequency in the newly added business data. The co-occurrence frequency is the frequency of synchronous associations between two groups of business entities obtained from the statistics of the newly added business data, reflecting the tightness and probability of association between entities. Specifically, the original weight parameters of corresponding entity association relationships in the initial knowledge graph are iteratively corrected based on the co-occurrence frequency in the newly added business data. High-frequency associations correspond to increased weight values, while low-frequency or invalid associations correspond to decreased weight values, to achieve dynamic calibration of the graph association strength and obtain a dynamic knowledge graph.

[0025] As can be seen from the above scheme, the entire process, by transforming new business data into incremental graph structure data, can adapt to the training input requirements of graph neural networks. By aggregating neighbor node features through graph neural networks, comprehensive integration of surrounding relational feature information of entities can be achieved. Then, updating the entity embedding vector based on the aggregated features can optimize the feature representation accuracy of graph entities. By updating the association weight attributes according to the data co-occurrence frequency, dynamic and accurate iteration of knowledge graph relationships can be realized.

[0026] S30: Use a pre-built dynamic knowledge graph to determine the first entity node corresponding to the first entity set, obtain at least one associated node of the first entity node, generate a dynamic information collection sequence for the evaluation object based on at least one associated node, and obtain the feedback data of the evaluation object for the dynamic information collection sequence in sequence. In this embodiment, after obtaining the first entity set, a pre-constructed dynamic knowledge graph is invoked to perform entity matching. The dynamic knowledge graph is an entity relationship network built by integrating multi-source medical data, insurance terms, and historical underwriting data, containing entity layers, relationship layers, and attribute layers. Each entity in the first entity set is mapped to the dynamic knowledge graph, retrieving the corresponding first entity node. Based on the preset association relationships within the dynamic knowledge graph, all associated nodes connected to the first entity node are traversed and retrieved. Based on the business and medical logic corresponding to the associated nodes, a dynamic information collection sequence adapted to the assessment object is generated. This dynamic information collection sequence is a progressive question sequence adaptively generated based on the entity association logic of the knowledge graph, possessing sequential association logic. The generated dynamic information collection sequence is pushed to the front-end interactive interface one by one, guiding the assessment object to answer sequentially. The assessment object's answers to the dynamic information collection sequence are collected and organized into corresponding feedback data.

[0027] In some embodiments, such as Figure 4As shown, step S30 above, which involves using a pre-constructed dynamic knowledge graph to determine the first entity node corresponding to the first entity set, obtaining at least one associated node of the first entity node, generating a dynamic information collection sequence for the evaluation object based on the at least one associated node, and sequentially obtaining the feedback data of the evaluation object for the dynamic information collection sequence, includes the following steps: S31: Determine the first entity node corresponding to the first entity set in the pre-constructed dynamic knowledge graph; S32: Obtain the first associated node that has a preset association relationship with the first entity node from the dynamic knowledge graph; S33: Based on the attribute information of the first associated node, generate a first-level information collection request and add it to the dynamic information collection sequence; S34: Receive feedback data from the evaluation object regarding the information collection request for the first level; S35: Based on feedback data, perform node jumping in the dynamic knowledge graph to determine the second associated node; S36: Generate a second-level information collection request based on the attribute information of the second associated node, and add the second-level information collection request to the dynamic information collection sequence in sequence. The granularity of the second-level information collection request is smaller than that of the first-level information collection request. S37: Repeat steps S34 to S36 until all the associated nodes of the feedback data in the dynamic knowledge graph have been traversed, or the number of iterations has reached the preset number of loop requests, then stop information collection and summarize the dynamic information collection sequence and feedback data.

[0028] For steps S31-S37, after obtaining the first entity set, the pre-constructed dynamic knowledge graph is first retrieved. The entities in the first entity set are semantically matched and category verified with the entity nodes in the dynamic knowledge graph. The graph nodes corresponding to the attributes and categories are filtered out, and the mapping relationship between each entity in the first entity set and the graph nodes is established to determine the first entity node.

[0029] Then, starting from the first entity node, the system traverses the association links of the dynamic knowledge graph relationship layer. Based on the preset association rules, it filters out downstream nodes that are related to the first entity node, thus obtaining the first associated node. The preset association rules are the association rules preset by the dynamic knowledge graph relationship layer, covering association logic such as disease evolution, complications, indicator tracing, and underwriting risks.

[0030] The system reads the attribute information corresponding to the first associated node in the dynamic knowledge graph attribute layer, extracts the query dimension, query topic, verification direction, etc., and generates query questions based on this information, which serve as the first-level information collection request. This first-level information collection request is then added to the dynamic information collection sequence to build the initial query sequence.

[0031] The first-level information collection request within the dynamic information collection sequence is pushed to the interactive terminal for the assessment subject to respond. The system receives the assessment subject's answer choices and text content, obtaining feedback data for the first-level information collection request. This feedback data consists of structured data and text information from the assessment subject's responses to each level of collection request, including health confirmation, symptom descriptions, and medical history records.

[0032] The received feedback data undergoes semantic recognition and content verification to extract entity information such as health confirmation, symptoms, medical history, and indicator status. The system retrieves the entity nodes corresponding to the matching entity information within a dynamic knowledge graph. Using the first associated node as a base point, it traverses the links according to the graph's built-in association directions, jumping to the retrieved entity node to complete the node jump operation and determine the second associated node. Node jumping is a retrieval process that, based on the entity information parsed from the feedback data, matches the corresponding node in the graph and then jumps directionally along the association link from the current node. The second associated node is a subdivided entity node downstream of the first associated node, corresponding to more refined health and risk scenarios.

[0033] The system reads the attribute information corresponding to the second associated node in the dynamic knowledge graph attribute layer, extracts detailed query dimensions, query topics, and verification directions, and generates query questions as the second-level information collection request. The second-level information collection request focuses on dimensions such as symptom details, indicator values, duration of illness, and medical treatment, with a smaller granularity than the first-level information collection request. The second-level information collection requests are then sequentially arranged after the first-level information collection requests and appended to the dynamic information collection sequence to complete the query sequence iteration.

[0034] Then, steps S34 to S36 are executed repeatedly, namely, the operations of receiving feedback data, node jumping, generating fine-grained information collection requests, and appending dynamic information collection sequences. The query content and dynamic information collection sequence are updated iteratively based on each round of feedback data. Information collection stops when all related nodes in the dynamic knowledge graph have been traversed, or when the preset number of iterations is reached. The preset number of iterations is a maximum number of iterations used to constrain the scope of query iterations and improve collection efficiency. The iteratively generated information collection requests are integrated to form a dynamic information collection sequence, and the feedback data obtained from the evaluation object's responses is collected.

[0035] As can be seen from the above scheme, the entire process expands the data collection dimensions layer by layer based on the entity relationships of a dynamic knowledge graph, enabling adaptive information collection. Through a hierarchical collection approach from coarse-grained to fine-grained, it can accurately and deeply uncover detailed information about target health risks. By relying on feedback data to complete iterative collection of graph nodes, it avoids the limitations of fixed collection patterns. By setting dual termination conditions for graph traversal and loop count, it can effectively control the collection scope and cost, achieving refined and progressive intelligent information collection.

[0036] S40: Extract the second entity set from the feedback data and perform a consistency comparison between the second entity set and the first entity set; In this embodiment, after obtaining the feedback data from the evaluation object, a semantic reasoning engine is invoked to perform semantic parsing and entity extraction processing on the feedback data. Entity information is extracted from the answers to dynamic queries and summarized to obtain a second entity set. After entity extraction, based on the entity correspondence and medical logic constraints built into the dynamic knowledge graph, a consistency comparison and verification is performed between the second entity set and the first entity set to determine whether there are any anomalies such as factual conflicts, logical contradictions, or information inconsistencies in the entity information obtained from the two collections. If anomalies exist, the second entity set is determined to be inconsistent with the first entity set, and the comparison fails; if no anomalies exist, the second entity set is determined to be consistent with the first entity set, and the comparison passes.

[0037] In some embodiments, step S40 above, namely extracting the second entity set from the feedback data and performing a consistency comparison between the second entity set and the first entity set, includes the following steps: S41: Perform semantic parsing on the feedback data, extract semantic entities and entity relationships from the feedback data, and obtain the second entity set; S42: Perform a cross-comparison between the second entity set and the first entity set; S43: If there is an entity conflict between the second entity set and the first entity set, the consistency comparison is determined to have failed, and an early warning signal is generated based on the content of the entity conflict. S44: If there is no conflict between the second entity set and the first entity set, then the consistency comparison is deemed successful.

[0038] For steps S41-S44, after obtaining the feedback data of the evaluation object, the feedback data is first subjected to semantic parsing processing to identify and extract the semantic entities and entity relationships contained in the feedback data, forming a second entity set.

[0039] Then, the second entity set is cross-checked with the first entity set item by item to verify the consistency of the entity content, corresponding entity attributes, and entity association relationships within the two sets, and to screen for differences and contradictions between the two sets.

[0040] If, after cross-checking, there is an entity conflict between the second entity set and the first entity set, the consistency comparison between the entities is deemed to have failed. An entity conflict occurs when the two entity sets have contradictory and incompatible content differences in dimensions such as health information, medical history, indicator status, and behavioral habits. Based on the specific content of the entity conflict identified during screening, a matching warning signal is generated.

[0041] If no entity conflicts are found between the second entity set and the first entity set after cross-comparison, it indicates that the entity information in the two sets is logically consistent and the content is consistent. Therefore, the consistency comparison between the entities is deemed to have passed, confirming that the collected feedback data is authentic and compliant, and can be used for subsequent risk reasoning calculations.

[0042] As can be seen from the above scheme, the entire process transforms unstructured feedback information into a comparable, standardized set of entities by semantically parsing the feedback data and extracting entities and relationships. Based on the cross-comparison of the two sets of entity data, contradictions and discrepancies between the preceding and following information can be accurately identified. By generating corresponding early warning signals for entity conflicts, timely risk alerts for abnormal information can be provided. Automatic verification of the comparison without conflicts enables rapid validation of the collected information's validity, ensuring the accuracy of subsequent risk reasoning.

[0043] Furthermore, step S43 above, which is to generate an early warning signal based on the content of the entity conflict, includes the following steps: S431: Identify the target entities that conflict with the first entity set in the second entity set, and determine the conflicting nodes corresponding to the target entities in the dynamic knowledge graph; S432: Obtain the first risk level attribute of the conflict node; S433: If the first risk level attribute is lower than the first preset risk threshold, a first warning signal is generated, and a review process is triggered based on the first warning signal; S434: If the first risk level attribute is higher than or equal to the first preset risk threshold, a second warning signal is generated, a supplementary supporting material request is generated based on the second warning signal, and the supplementary supporting material request is sent to the assessment object.

[0044] For steps S431-S434, when the cross-match between the second entity set and the first entity set fails, entities with content differences and information contradictions between the second and first entity sets are first identified and marked as target entities. The graph nodes corresponding to the target entities are retrieved from the dynamic knowledge graph and identified as conflict nodes. Here, the target entity is the entity content in the two entity sets that is inconsistent in information and has content contradictions; the conflict node is the entity node in the dynamic knowledge graph that precisely corresponds to the target entity, carrying conflict information and risk attributes.

[0045] The attribute data of the conflicting nodes within the attribute layer of the dynamic knowledge graph is read, and the risk level parameters corresponding to the nodes are extracted to obtain the first risk level attribute. The first risk level attribute is a risk quantification parameter inherent to the conflicting node, used to measure the severity of the risk corresponding to the entity information when an anomaly occurs.

[0046] The first risk level attribute is compared with the first preset risk threshold. If the first risk level attribute is lower than the first preset risk threshold, the current entity conflict risk level is determined to be low, a first warning signal is generated, and a manual review process is initiated based on the first warning signal. The first preset risk threshold is a pre-set critical value for risk level, used to distinguish between low-risk and high-risk conflicts; the first warning signal is the warning indicator corresponding to a low-level information conflict.

[0047] If the first risk level attribute is higher than or equal to the first preset risk threshold, the current entity conflict risk level is determined to be high, and a second warning signal is generated. Based on the second warning signal, a corresponding request for supplementary supporting materials is generated and pushed to the assessed entity so that it can supplement the corresponding supporting materials to complete the verification and evidence collection for the high-risk conflict. The second warning signal is a warning indicator corresponding to a high-level information conflict; the request for supplementary supporting materials is a material collection instruction generated for high-risk entity conflicts, used to verify the authenticity of the conflict information.

[0048] As can be seen from the above scheme, the entire process, by locating the graph conflict nodes corresponding to entity conflicts and reading the risk level attributes, generates early warning signals based on the differences in risk thresholds and risk level attributes, triggering corresponding processing procedures, and thus achieving hierarchical control of underwriting anomalies. By issuing supplementary material requests for high-risk conflicts, high-risk information contradictions can be accurately verified, realizing an upgrade from passive anti-fraud to proactive risk verification.

[0049] S50: If the comparison passes, the dynamic knowledge graph is used to determine the second entity node corresponding to the second entity set, and the transmission path between the first entity node and the second entity node is determined. In this embodiment, if the comparison passes, meaning the consistency comparison result is without conflict or anomaly, a dynamic knowledge graph is used to perform node matching on the second entity set to find the second entity node corresponding to the second entity set. Subsequently, the association links between each entity node in the dynamic knowledge graph are traversed to locate the transmission path connecting the first entity node and the second entity node, so as to reconstruct the evolution and association logic between different risk elements.

[0050] S60: Based on the transmission path and the preset risk nodes in the dynamic knowledge graph, perform chain reasoning to generate the risk assessment results of the assessment object and the corresponding reasoning path data.

[0051] In this embodiment, the dynamic knowledge graph is pre-configured with several preset risk nodes. Based on the identified transmission paths and preset risk nodes, full-link chain reasoning is performed within the dynamic knowledge graph, deducing the risk associations with the preset risk nodes layer by layer along the transmission paths. After completing the chain reasoning operation, the risk assessment result for the assessment object is output, and the entity nodes, relationships, and transmission links traversed during the risk reasoning process are recorded, generating corresponding reasoning path data.

[0052] In some embodiments, such as Figure 5 As shown, step S60 above, which involves chain reasoning based on the transmission path and preset risk nodes in the dynamic knowledge graph to generate the risk assessment result of the assessment object and the corresponding reasoning path data, includes the following steps: S61: Obtain the preset risk nodes in the dynamic knowledge graph, and determine all candidate paths to the risk nodes from the transmission path; S62: Obtain the second risk level attribute of each node on each candidate path and the transmission probability of the relationship between each node; S63: Calculate the comprehensive risk score for each candidate path based on the second risk level attribute and the transmission probability; S64: Identify candidate paths with a comprehensive risk score higher than the second preset risk threshold as target risk paths, and generate risk assessment results based on the comprehensive risk score of the target risk paths; S65: Transform the node sequences and relationships of each node in the target risk path into inference path data.

[0053] For steps S61-S65, pre-configured risk nodes are retrieved within the dynamic knowledge graph. Combining the transmission path connecting the first entity node and the second entity node, the topological links of the transmission path are traversed to filter out all links that can connect to the risk nodes, resulting in multiple candidate paths. The preset risk nodes are entity nodes in the dynamic knowledge graph that have been pre-marked as having underwriting risks.

[0054] For each candidate path, the attribute information of all entity nodes covered by the candidate path is read sequentially, and the second risk level attribute corresponding to each node is extracted. At the same time, the propagation probability parameter corresponding to the association relationship between adjacent nodes is obtained. Among them, the second risk level attribute is the risk quantification score inherent to the entity node, which is used to characterize the degree of independent risk of a single node; the propagation probability is the quantification parameter corresponding to the node association relationship, which is used to characterize the probability of risk being transmitted along the corresponding link.

[0055] Then, the comprehensive risk score of the candidate paths is calculated. For each candidate path, based on the second risk level attribute of each node and the transmission probability between nodes, and combined with the preset weighted fusion calculation rules, the risk of a single node and the transmission risk of the link are integrated and calculated to obtain the comprehensive risk score corresponding to each candidate path. Among them, the comprehensive risk score is an overall quantitative score obtained by integrating the inherent risk of the node and the transmission characteristics of the link, which is used to determine the risk level of a single transmission path.

[0056] The comprehensive risk score of each candidate path is compared with a second preset risk threshold. Paths whose comprehensive risk scores exceed the preset risk threshold are selected and marked as target risk paths. Combining the target risk paths' scores, risk transmission levels, and risk correlation characteristics, corresponding risk judgment criteria are matched, and risk assessment results are generated for the assessed objects based on these criteria. The second preset risk threshold is a pre-set critical standard for risk scores, used to screen high-risk transmission paths.

[0057] Next, the topological structure data of the entity node sequence and the relationship between nodes arranged in sequence in the target risk path are extracted and converted into text-based, traceable reasoning path data to fully record the link basis and deduction process of this risk reasoning.

[0058] As can be seen from the above scheme, the entire process, based on the preset risk nodes and transmission paths of a dynamic knowledge graph, can accurately identify candidate paths with potential risk transmission. By combining the node risk level attributes with the transmission probability of the link for weighted calculation, the overall risk level of a single path can be quantified. By screening target risk paths through threshold comparison, high-risk transmission links can be accurately identified, achieving precise risk positioning. By generating corresponding risk assessment results based on the target risk paths, quantitative risk conclusions that fit the actual scenario can be output. By structurally transforming the node sequence and relationships of the target risk paths, inference path data is obtained, which can retain the complete inference link and achieve traceability of the risk assessment process.

[0059] For example, the risk assessment scheme of the present invention can be applied to intelligent underwriting scenarios in the financial insurance industry, mainly for automated risk verification and intelligent risk assessment in the process of applying for personal health insurance, critical illness insurance, and medical insurance.

[0060] For example, during the operation of the intelligent underwriting system, the initial business interaction data of the insured user is first acquired, including the user's self-reported health declaration information, previous insurance records, basic physical examination information, etc. Semantic parsing technology is used to extract core entities such as symptoms, diseases, physical examination indicators, and medical history from massive amounts of unstructured business data, forming a first entity set, which serves as the initial benchmark data for this underwriting risk assessment. Based on a pre-built and continuously updated dynamic knowledge graph, the first entity nodes corresponding to the first entity set are matched. This knowledge graph is constructed based on multi-source medical knowledge, underwriting rules, and historical business data, and can continuously learn from new business data through graph neural networks, updating entity features and association weights in real time to align with the latest underwriting standards and medical common sense.

[0061] The underwriting system, based on the first entity node, retrieves related nodes with pre-defined relationships in the knowledge graph and generates a hierarchical, dynamic information collection sequence according to a gradient of granularity from coarse to fine. The first round generates broad, basic health questions. After the user provides feedback, the system uses this feedback data to jump between knowledge graph nodes, delving deeper into specific health dimensions to generate more refined secondary questions. This iterative collection process continues until a termination condition is met, ultimately obtaining complete user feedback data. This dynamic questioning method breaks through the limitations of traditional fixed questionnaires, enabling adaptive adjustments to the question content based on different users' health scenarios.

[0062] After information collection is completed, semantic analysis is performed on user feedback data to extract new semantic entities and relationships from the feedback content, forming a second entity set. This second entity set is then cross-referenced with the initial first entity set to verify for any inconsistencies in the user's previously submitted health information. If conflicts are detected in medical history, symptoms, or physical examination indicators, the risk level is determined by the risk level attribute of the conflicting entity. Low-level conflicts automatically trigger a manual review process, while high-level conflicts automatically send a request for supplementary materials to the user, requiring them to submit supporting documents such as physical examination reports and medical records, thus mitigating underwriting risks associated with false reporting, omissions, or errors.

[0063] If the user information comparison is conflict-free and the information is deemed authentic and valid, the system uses a dynamic knowledge graph to match the initial first entity node with the feedback second entity node, establishing a connection path. It then retrieves various pre-defined high-risk underwriting nodes from the graph and filters out all candidate connection paths that can connect to these risk nodes. Combining the risk level attributes of each path node and the probability of risk transmission between nodes, a weighted comprehensive risk score is calculated for each path. Finally, a pre-defined risk threshold is used to filter out target risk paths with potential risks.

[0064] Then, based on the risk score and transmission logic of the target risk path, a user-specific intelligent underwriting risk assessment result is generated. Simultaneously, the node sequence and relationships within the path are organized to form traceable reasoning path data. Underwriters can use this risk assessment result and reasoning path data to quickly determine the user's insurance risk level and provide underwriting conclusions such as acceptance, rejection, acceptance with increased premiums, or deferred acceptance. This solution significantly improves the intelligence level of insurance underwriting and the accuracy of risk management.

[0065] As can be seen, in the above scheme, by obtaining the initial business interaction data of the evaluation object and performing semantic parsing on the initial business interaction data to obtain the first entity set, it can be compatible with multimodal data and complete structured purification, realizing the fusion and regularization of multi-source data; by using a pre-constructed dynamic knowledge graph to determine the first entity node corresponding to the first entity set, obtaining at least one associated node of the first entity node, and generating a dynamic information collection sequence for the evaluation object based on at least one associated node, and sequentially obtaining the feedback data of the evaluation object for the dynamic information collection sequence, it can rely on the dynamic knowledge graph to generate personalized dynamic information collection sequences based on the initial entity nodes, realize adaptive expansion of related queries based on user feedback data, so that the collected user information has logical correlation, and thoroughly improve the problems of data fragmentation and lack of linkage correlation; subsequently, By extracting a second entity set from the feedback data and comparing it with the first entity set for consistency verification, the authenticity and accuracy of the collected information are verified. A dynamic knowledge graph is used to determine the second entity node corresponding to the second entity set and the transmission path between the first and second entity nodes. Based on this transmission path and preset risk nodes in the dynamic knowledge graph, chain-like reasoning is performed to generate the risk assessment result of the assessed object and the corresponding reasoning path data. This approach leverages the dynamic knowledge graph to organize the risk transmission path and combines preset risk nodes to complete chain-like progressive risk reasoning. It overcomes the shortcomings of traditional isolated judgments of single risk indicators, supports in-depth mining of the inherent connections and progressive evolution logic of health risks, accurately identifies potential related risks, and effectively improves the accuracy and reliability of insurance underwriting risk assessment in complex scenarios.

[0066] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0067] In one embodiment, a risk assessment device is provided, which corresponds one-to-one with the risk assessment methods described in the above embodiments. For example... Figure 6 As shown, the risk assessment device includes a data parsing module 101, an information acquisition module 102, an entity comparison module 103, a path determination module 104, and a risk assessment module 105. Detailed descriptions of each functional module are as follows: Data parsing module 101 is used to obtain the initial business interaction data of the evaluation object, perform semantic parsing on the initial business interaction data, and determine the first entity set; The information acquisition module 102 is used to determine the first entity node corresponding to the first entity set using a pre-built dynamic knowledge graph, obtain at least one associated node of the first entity node, generate a dynamic information acquisition sequence for the evaluation object based on at least one associated node, and obtain feedback data of the evaluation object for the dynamic information acquisition sequence in sequence. The entity comparison module 103 is used to extract a second entity set from the feedback data and perform a consistency comparison between the second entity set and the first entity set. The path determination module 104 is used to determine the second entity node corresponding to the second entity set by using a dynamic knowledge graph if the comparison passes, and to determine the transmission path between the first entity node and the second entity node. The risk assessment module 105 is used to perform chain reasoning based on the transmission path and preset risk nodes in the dynamic knowledge graph, and generate the risk assessment results of the assessed object and the corresponding reasoning path data.

[0068] In one embodiment, the risk assessment device further includes: The multi-source data acquisition module is used to acquire multi-source heterogeneous domain knowledge data and historical business interaction data; The knowledge graph construction module is used to extract business entities and the relationships between business entities from domain knowledge data and historical business interaction data to build an initial knowledge graph. The graph update module is used to learn the representation of newly added business data through a graph neural network when new business data is detected, and to update the embedding vectors of business entities and the weight attributes of relationships in the initial knowledge graph to obtain a dynamic knowledge graph.

[0069] In one embodiment, the map update module is specifically used for: When new business data is detected, the new business data will be converted into incremental graph structure data; The incremental graph structure data is input into a pre-trained graph neural network model, and the aggregation function in the graph neural network model is used to aggregate the features of the neighbor nodes of each node in the incremental graph structure data. The embedding vectors of the corresponding business entities in the initial knowledge graph are updated based on the aggregated neighbor node features, and the weight attributes of the associations are updated based on the co-occurrence frequency in the newly added business data to obtain the dynamic knowledge graph.

[0070] In one embodiment, the information acquisition module 102 is specifically used for: S31, determine the first entity node corresponding to the first entity set in the pre-constructed dynamic knowledge graph; S32, Obtain the first associated node that has a preset association relationship with the first entity node from the dynamic knowledge graph; S33, Based on the attribute information of the first associated node, generate a first-level information collection request and add it to the dynamic information collection sequence; S34, Receive feedback data from the evaluation object regarding the information collection request for the first level; S35, based on feedback data, perform node jumping in the dynamic knowledge graph to determine the second associated node; S36, Generate a second-level information collection request based on the attribute information of the second associated node, and add the second-level information collection request to the dynamic information collection sequence in sequence, wherein the granularity of the second-level information collection request is smaller than that of the first-level information collection request. S37. Repeat steps S34 to S36 until all the associated nodes of the feedback data in the dynamic knowledge graph have been traversed, or the number of iterations reaches the preset number of loop requests, then stop information collection and summarize the dynamic information collection sequence and feedback data.

[0071] In one embodiment, the entity comparison module 103 is specifically used for: Semantic parsing is performed on the feedback data to extract semantic entities and entity relationships, resulting in a second entity set. Perform a cross-comparison between the second entity set and the first entity set; If there is an entity conflict between the second entity set and the first entity set, the consistency comparison is determined to have failed, and an early warning signal is generated based on the content of the entity conflict. If the second entity set has no conflict with the first entity set, then the consistency comparison is deemed successful.

[0072] In one embodiment, the entity comparison module 103 is further configured to: Identify the target entities that conflict with the first entity set in the second entity set, and determine the conflict nodes corresponding to the target entities in the dynamic knowledge graph; Obtain the first risk level attribute of the conflict node; If the first risk level attribute is lower than the first preset risk threshold, a first warning signal is generated, and a review process is triggered based on the first warning signal. If the first risk level attribute is higher than or equal to the first preset risk threshold, a second early warning signal is generated, and a request for supplementary supporting materials is generated based on the second early warning signal and sent to the assessment object.

[0073] In one embodiment, the risk assessment module 105 is specifically used for: Obtain the pre-defined risk nodes in the dynamic knowledge graph, and determine all candidate paths to reach the risk nodes from the transmission path; Obtain the second risk level attribute of each node on each candidate path and the transmission probability of the correlation between each node; Based on the second risk level attributes and transmission probability, calculate the comprehensive risk score for each candidate path; Candidate paths with a comprehensive risk score higher than the second preset risk threshold are identified as target risk paths, and risk assessment results are generated based on the comprehensive risk score of the target risk paths. The node sequences and relationships of each node in the target risk path are transformed into inference path data.

[0074] This invention provides a risk assessment device that acquires initial business interaction data of the assessment object and performs semantic parsing on the initial business interaction data to obtain a first entity set. It is compatible with multimodal data and performs structured purification, achieving multi-source data fusion and regularization. By using a pre-constructed dynamic knowledge graph to determine the first entity nodes corresponding to the first entity set, it acquires at least one associated node of the first entity node and generates a dynamic information collection sequence for the assessment object based on at least one associated node. It sequentially acquires feedback data from the assessment object in response to the dynamic information collection sequence. It can generate personalized dynamic information collection sequences based on the initial entity nodes using the dynamic knowledge graph, enabling adaptive expansion of related queries based on user feedback data. This ensures that the collected user information has logical coherence, thoroughly improving the problems of data fragmentation and lack of linkage and correlation. Subsequently, by extracting a second entity set from the feedback data and comparing it with the first entity set for consistency verification, the authenticity and accuracy of the collected information are verified. A dynamic knowledge graph is used to determine the second entity node corresponding to the second entity set, and the transmission path between the first and second entity nodes is established. Based on this transmission path and preset risk nodes in the dynamic knowledge graph, chain-like reasoning is performed to generate the risk assessment results of the assessed object and the corresponding reasoning path data. This approach leverages the dynamic knowledge graph to organize risk transmission paths and combines preset risk nodes to complete chain-like progressive risk reasoning. It overcomes the shortcomings of traditional isolated judgments of single risk indicators, supports in-depth mining of the inherent connections and progressive evolution logic of health risks, accurately identifies potential related risks, and effectively improves the accuracy and reliability of insurance underwriting risk assessment in complex scenarios.

[0075] Specific limitations regarding the risk assessment device can be found in the limitations of the intelligent risk assessment method described above, and will not be repeated here. Each module in the aforementioned risk assessment device 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 in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0076] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 7 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with external clients via a network connection. When the computer program is executed by the processor, it implements the functions or steps of a risk assessment method on the server side.

[0077] In one embodiment, a computer device is provided, which may be a client, and its internal structure diagram may be as follows: Figure 8 As shown, the computer device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with an external server via a network connection. When the computer program is executed by the processor, it implements the functions or steps of a risk assessment method on the client side.

[0078] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the following steps: Obtain the initial business interaction data of the evaluation object, perform semantic parsing on the initial business interaction data, and determine the first entity set; The first entity node corresponding to the first entity set is determined by using a pre-constructed dynamic knowledge graph, at least one associated node of the first entity node is obtained, a dynamic information collection sequence for the evaluation object is generated based on at least one associated node, and feedback data of the evaluation object for the dynamic information collection sequence is obtained in sequence. Extract the second entity set from the feedback data and perform a consistency comparison between the second entity set and the first entity set; If the comparison passes, a dynamic knowledge graph is used to determine the second entity node corresponding to the second entity set, and to determine the transmission path between the first entity node and the second entity node. Chain reasoning is performed based on the transmission path and preset risk nodes in the dynamic knowledge graph to generate risk assessment results for the assessed object and corresponding reasoning path data.

[0079] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor: Obtain the initial business interaction data of the evaluation object, perform semantic parsing on the initial business interaction data, and determine the first entity set; The first entity node corresponding to the first entity set is determined by using a pre-constructed dynamic knowledge graph, at least one associated node of the first entity node is obtained, a dynamic information collection sequence for the evaluation object is generated based on at least one associated node, and feedback data of the evaluation object for the dynamic information collection sequence is obtained in sequence. Extract the second entity set from the feedback data and perform a consistency comparison between the second entity set and the first entity set; If the comparison passes, a dynamic knowledge graph is used to determine the second entity node corresponding to the second entity set, and to determine the transmission path between the first entity node and the second entity node. Chain reasoning is performed based on the transmission path and preset risk nodes in the dynamic knowledge graph to generate risk assessment results for the assessed object and corresponding reasoning path data.

[0080] It should be noted that the functions or steps that can be implemented by the computer-readable storage medium or computer device described above can be referred to the relevant descriptions on the server side and client side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0081] 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 in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of 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 memory bus dynamic RAM (RDRAM), etc.

[0082] 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 device can be divided into different functional units or modules to complete all or part of the functions described above.

[0083] It should be noted that any AI models, software tools, or components not belonging to this company appearing in the embodiments of this application are merely illustrative examples and do not represent actual use. All user personal information involved in the embodiments of this application has been authorized (with the knowledge and consent) by the relevant parties or has been fully authorized by all parties, and the executing entity may obtain it through various legal and compliant means. The collection, storage, use, processing, transmission, provision, and disclosure of the information, data, and signals involved all comply with relevant laws and regulations and do not violate public order and good morals.

[0084] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A risk assessment method, characterized in that, The method includes: Obtain the initial business interaction data of the evaluation object, perform semantic parsing on the initial business interaction data, and determine the first entity set; The first entity node corresponding to the first entity set is determined by using a pre-constructed dynamic knowledge graph, at least one associated node of the first entity node is obtained, a dynamic information collection sequence for the evaluation object is generated based on the at least one associated node, and feedback data of the evaluation object for the dynamic information collection sequence is obtained in sequence. Extract a second entity set from the feedback data, and perform a consistency comparison between the second entity set and the first entity set; If the comparison passes, the dynamic knowledge graph is used to determine the second entity node corresponding to the second entity set, and the transmission path between the first entity node and the second entity node is determined. Based on the transmission path and the preset risk nodes in the dynamic knowledge graph, chain reasoning is performed to generate the risk assessment result of the assessment object and the corresponding reasoning path data.

2. The risk assessment method as described in claim 1, characterized in that, Before obtaining the initial business interaction data of the evaluation object, performing semantic parsing on the initial business interaction data, and determining the first entity set, the method further includes: Acquire multi-source heterogeneous domain knowledge data and historical business interaction data; Extract the business entities from the domain knowledge data and the historical business interaction data, as well as the relationships between the business entities, to construct an initial knowledge graph; When new business data is detected, the new business data is represented by a graph neural network to update the embedding vectors of the business entities and the weight attributes of the relationships in the initial knowledge graph, thereby obtaining the dynamic knowledge graph.

3. The risk assessment method as described in claim 2, characterized in that, When new business data is detected, a graph neural network is used to learn the representation of the new business data, updating the embedding vectors of the business entities and the weight attributes of the relationships in the initial knowledge graph to obtain the dynamic knowledge graph, including: When new business data is detected, the new business data is converted into incremental graph structure data; The incremental graph structure data is input into a pre-trained graph neural network model, and the aggregation function in the graph neural network model is used to aggregate the neighbor node features of each node in the incremental graph structure data. The embedding vectors of the corresponding business entities in the initial knowledge graph are updated based on the aggregated neighbor node features, and the weight attributes of the associations are updated based on the co-occurrence frequency in the newly added business data to obtain the dynamic knowledge graph.

4. The risk assessment method as described in claim 1, characterized in that, The process involves using a pre-constructed dynamic knowledge graph to determine the first entity node corresponding to the first entity set, obtaining at least one associated node of the first entity node, generating a dynamic information collection sequence for the evaluation object based on the at least one associated node, and sequentially obtaining feedback data from the evaluation object for the dynamic information collection sequence, including: S31, determine the first entity node corresponding to the first entity set in the pre-constructed dynamic knowledge graph; S32, Obtain a first associated node that has a preset association relationship with the first entity node from the dynamic knowledge graph; S33, Based on the attribute information of the first associated node, generate a first-level information collection request and add it to the dynamic information collection sequence; S34, Receive feedback data from the evaluation object regarding the information collection request for the first level; S35, based on the feedback data, perform node jumping in the dynamic knowledge graph to determine the second associated node; S36, Generate a second-level information collection request based on the attribute information of the second associated node, and add the second-level information collection request to the dynamic information collection sequence in sequence, wherein the granularity of the second-level information collection request is smaller than the granularity of the first-level information collection request. S37. Repeat steps S34 to S36 until all associated nodes of the feedback data in the dynamic knowledge graph have been traversed, or the number of iterations reaches the preset number of loop requests, then stop information collection and summarize the dynamic information collection sequence and the feedback data.

5. The risk assessment method as described in claim 1, characterized in that, The step of extracting a second entity set from the feedback data and performing a consistency comparison between the second entity set and the first entity set includes: The feedback data is semantically parsed to extract semantic entities and entity relationships, thus obtaining the second entity set. Perform a cross-comparison between the second entity set and the first entity set; If there is an entity conflict between the second entity set and the first entity set, the consistency comparison is determined to have failed, and an early warning signal is generated based on the content of the entity conflict. If the second entity set has no entity conflict with the first entity set, then the consistency comparison is deemed to have passed.

6. The risk assessment method as described in claim 5, characterized in that, The generation of early warning signals based on entity conflict content includes: Identify the target entities that conflict between the second entity set and the first entity set, and determine the conflicting nodes corresponding to the target entities in the dynamic knowledge graph; Obtain the first risk level attribute of the conflict node; If the first risk level attribute is lower than the first preset risk threshold, a first warning signal is generated, and a review process is triggered based on the first warning signal. If the first risk level attribute is higher than or equal to the first preset risk threshold, a second warning signal is generated, a supplementary supporting document request is generated based on the second warning signal, and the supplementary supporting document request is sent to the assessment object.

7. The risk assessment method as described in claim 1, characterized in that, The step of performing chain-like reasoning based on the transmission path and preset risk nodes in the dynamic knowledge graph to generate the risk assessment result of the assessment object and the corresponding reasoning path data includes: Obtain the preset risk nodes in the dynamic knowledge graph, and determine all candidate paths to the risk nodes from the transmission path; Obtain the second risk level attribute of each node on each candidate path and the transmission probability of the correlation between each node; Based on the second risk level attribute and the transmission probability, calculate the comprehensive risk score for each candidate path; Candidate paths whose comprehensive risk scores are higher than the second preset risk threshold are identified as target risk paths, and risk assessment results are generated based on the comprehensive risk scores of the target risk paths. The node sequences and relationships of each node in the target risk path are converted into inference path data.

8. A risk assessment device, characterized in that, include: The data parsing module is used to obtain the initial business interaction data of the evaluation object, perform semantic parsing on the initial business interaction data, and determine the first entity set; The information acquisition module is used to determine the first entity node corresponding to the first entity set using a pre-constructed dynamic knowledge graph, obtain at least one associated node of the first entity node, generate a dynamic information acquisition sequence for the evaluation object based on the at least one associated node, and obtain feedback data of the evaluation object for the dynamic information acquisition sequence in sequence. The entity comparison module is used to extract a second entity set from the feedback data and perform a consistency comparison between the second entity set and the first entity set. The path determination module is used to determine the second entity node corresponding to the second entity set using the dynamic knowledge graph if the comparison passes, and to determine the transmission path between the first entity node and the second entity node. The risk assessment module is used to perform chain reasoning based on the transmission path and preset risk nodes in the dynamic knowledge graph to generate the risk assessment result of the assessment object and the corresponding reasoning path data.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the risk assessment method as described in 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 a processor, it implements the steps of the risk assessment method as described in any one of claims 1 to 7.

Images