Content conflict detection method and device based on knowledge graph, equipment and medium

Abstract

The invention belongs to the field of artificial intelligence, and relates to a clause conflict detection method and device based on a knowledge graph, equipment and a medium, and the method comprises the steps: obtaining multi-source data which comprises historical clause data, risk case data and supervision policy data; based on the historical clause data, the risk case data and the supervision policy data, constructing a dynamic knowledge graph; receiving term combination data of the target object; based on the dynamic knowledge graph, calculating a global risk value of the term combination data through a preset graph neural network, the global risk value reflecting a conflict degree between terms in the term combination data; if the global risk value is greater than a preset threshold value, generating conflict attribution information of the term combination data; and adjusting the clause combination data based on the conflict attribution information to obtain a target clause combination of the target object. The method can be applied to the business fields of financial insurance and the like, and the conflict detection accuracy and efficiency of the clause combination can be improved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology and is applied to online processing business scenarios such as finance and insurance. In particular, it relates to a method, device, equipment and medium for detecting clause conflicts based on knowledge graphs. Background Technology

[0002] In the field of insurance clause conflict detection, the accuracy and timeliness of clause conflict detection are crucial for the compliant operation and risk control of insurance business. However, current clause conflict detection technologies in the industry still face many technical challenges that urgently need to be addressed.

[0003] Current mainstream clause conflict detection systems primarily rely on manually maintained, fixed rule bases for static rule detection. For example, they might set fixed rules like "the sum insured must not exceed 200% of the industry average." While this method can identify explicit conflicts to some extent, it is ineffective against implicit and complex contradictions between clauses. For cases like "the main insurance sum insured is high, but the exclusion clauses do not cover emerging risks," the fixed nature of the rule base makes it difficult for the system to automatically identify them, resulting in insufficient conflict coverage and failing to fully guarantee the compliance of the clauses. Furthermore, existing verification processes have significant shortcomings in data processing. They can only parse structured fields in the application form, such as the sum insured amount, but lack semantic-level parsing capabilities for free text descriptions, such as "the types of stored goods include lithium batteries." This makes it impossible to effectively identify the risks inherent in a large amount of unstructured data, resulting in a high percentage of blind spots in risk identification and severely impacting the effectiveness of risk control.

[0004] In summary, current industry-standard clause conflict detection technologies suffer from several technical problems: static rules make it difficult to identify hidden conflicts, and the lack of unstructured data processing leads to large blind spots, resulting in low accuracy and efficiency in clause combination conflict detection. Summary of the Invention

[0005] The purpose of this application is to propose a knowledge graph-based method, apparatus, computer equipment, and storage medium for detecting clause conflicts, in order to solve the problems of existing clause conflict detection technologies in the industry, such as the difficulty in identifying hidden conflicts due to static rules, the lack of unstructured data processing leading to large risk blind spots, and the low accuracy and efficiency of clause combination conflict detection.

[0006] Firstly, a knowledge graph-based method for detecting clause conflicts is provided, employing the following technical solution:

[0007] Acquire multi-source data, including historical clause data, risk case data, and regulatory policy data; construct a dynamic knowledge graph based on the historical clause data, risk case data, and regulatory policy data; receive clause combination data of the target object; calculate the global risk value of the clause combination data based on the dynamic knowledge graph and a preset graph neural network, the global risk value reflecting the degree of conflict between clauses in the clause combination data; if the global risk value is greater than a preset threshold, generate conflict attribution information for the clause combination data; adjust the clause combination data based on the conflict attribution information to obtain the target clause combination of the target object.

[0008] Secondly, a knowledge graph-based clause conflict detection device is provided, which adopts the following technical solution:

[0009] The acquisition module is used to acquire multi-source data, including historical terms data, risk case data, and regulatory policy data.

[0010] The module is used to build a dynamic knowledge graph based on historical clause data, risk case data, and regulatory policy data;

[0011] The receiving module is used to receive the terms combination data of the target object;

[0012] The calculation module is used to calculate the global risk value of the clause combination data based on the dynamic knowledge graph and through the preset graph neural network. The global risk value reflects the degree of conflict between the clauses in the clause combination data.

[0013] The generation module is used to generate conflict attribution information for the clause combination data if the global risk value is greater than a preset threshold.

[0014] The adjustment module is used to adjust the clause combination data based on conflict attribution information to obtain the target clause combination for the target object.

[0015] Thirdly, a computer device is provided, which adopts the following technical solution:

[0016] Acquire multi-source data, including historical clause data, risk case data, and regulatory policy data; construct a dynamic knowledge graph based on the historical clause data, risk case data, and regulatory policy data; receive clause combination data of the target object; calculate the global risk value of the clause combination data based on the dynamic knowledge graph and a preset graph neural network, the global risk value reflecting the degree of conflict between clauses in the clause combination data; if the global risk value is greater than a preset threshold, generate conflict attribution information for the clause combination data; adjust the clause combination data based on the conflict attribution information to obtain the target clause combination of the target object.

[0017] Fourthly, a computer-readable storage medium is provided, which adopts the following technical solution:

[0018] Acquire multi-source data, including historical clause data, risk case data, and regulatory policy data; construct a dynamic knowledge graph based on the historical clause data, risk case data, and regulatory policy data; receive clause combination data of the target object; calculate the global risk value of the clause combination data based on the dynamic knowledge graph and a preset graph neural network, the global risk value reflecting the degree of conflict between clauses in the clause combination data; if the global risk value is greater than a preset threshold, generate conflict attribution information for the clause combination data; adjust the clause combination data based on the conflict attribution information to obtain the target clause combination of the target object.

[0019] Compared with existing technologies, the embodiments of this application have the following main advantages: By constructing a dynamic knowledge graph from multi-source data such as historical clauses, risk cases, and regulatory policies, the static limitations of traditional rule bases are broken. The dynamic knowledge graph can be updated and evolved in real time. Combined with graph neural networks, it can deeply mine implicit and complex contradictory relationships in clause combinations, accurately calculate global risk values, effectively solve the problem that static rules are difficult to identify implicit conflicts, and significantly improve conflict coverage. Furthermore, by comparing the global risk value with a preset threshold to generate conflict attribution information and adjusting clause combinations accordingly, the automation and intelligence of clause conflict detection are achieved, significantly improving the accuracy and efficiency of clause combination conflict detection. Attached Figure Description

[0020] To more clearly illustrate the solutions in this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is an exemplary system architecture diagram to which this application can be applied;

[0022] Figure 2 A flowchart of an embodiment of the knowledge graph-based clause conflict detection method according to this application;

[0023] Figure 3 This is a schematic diagram of the structure of one embodiment of the knowledge graph-based clause conflict detection device according to this application;

[0024] Figure 4 This is a schematic diagram of the structure of one embodiment of the computer device according to this application. Detailed Implementation

[0025] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application, are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.

[0026] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0027] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0028] like Figure 1 As shown, system architecture 100 may include terminal device 101, network 102, and server 103. Terminal device 101 may be a laptop 1011, tablet 1012, or mobile phone 1013. Network 102 is used as a medium to provide a communication link between terminal device 101 and server 103. Network 102 may include various connection types, such as wired, wireless communication links, or fiber optic cables.

[0029] Users can use terminal device 101 to interact with server 103 via network 102 to receive or send messages, etc. Various communication client applications can be installed on terminal device 101, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social media platform software, etc.

[0030] Terminal device 101 can be various electronic devices with a display screen and support web browsing. In addition to laptop computer 1011, tablet computer 1012 or mobile phone 1013, terminal device 101 can also be e-book reader, MP3 player (Moving Picture Experts Group Audio Layer III), MP4 player (Moving Picture Experts Group Audio Layer IV), laptop computer and desktop computer, etc.

[0031] Server 103 can be a server that provides various services, such as a backend server that provides support for the pages displayed on terminal device 101.

[0032] It should be noted that the knowledge graph-based clause conflict detection method provided in this application embodiment is generally executed by a server / terminal device, and correspondingly, the knowledge graph-based clause conflict detection device is generally set in the server / terminal device.

[0033] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0034] Continue to refer to Figure 2 The diagram illustrates a flowchart of an embodiment of a knowledge graph-based clause conflict detection method according to this application. The knowledge graph-based clause conflict detection method includes the following steps:

[0035] Step S201: Obtain multi-source data, which includes historical terms data, risk case data, and regulatory policy data.

[0036] Multi-source data refers to a diverse and complementary set of data obtained from multiple different data sources. These data originate from various channels, such as historical policy storage systems, risk case management systems, and regulatory policy publishing platforms, and represent multi-dimensional information involved in the formulation and execution of insurance policy terms.

[0037] Historical policy data refers to the texts and related attribute information of past insurance policies accumulated during the course of insurance business operations. It originates from the policy database in the insurance company's historical business system and represents the specific content and characteristics of insurance policies from different periods and for different types of insurance. This data provides the foundational policy knowledge for constructing a dynamic knowledge graph. For example, it includes detailed information such as policy number, type of insurance, insurance liability, and exclusions.

[0038] Among them, risk case data refers to the recorded data of various risk events that actually occur in the insurance industry and their handling results. It represents the risk scenarios and consequences that insurance clauses may face in actual application, and is used to inject practical risk experience into the dynamic knowledge graph. For example, it includes key information such as case number, time of occurrence, cause of occurrence, amount of loss, and applicable clauses.

[0039] Among them, regulatory policy data refers to laws, regulations, normative documents, and guidelines issued by insurance regulatory agencies related to the formulation and implementation of insurance clauses. This data represents the compliance requirements and regulatory orientation that insurance clauses must follow, ensuring that the dynamic knowledge graph conforms to industry regulatory standards. For example, it includes policy numbers, issuing agencies, effective dates, and specific clause requirements.

[0040] Step S202: Construct a dynamic knowledge graph based on historical clause data, risk case data, and regulatory policy data.

[0041] Among them, dynamic knowledge graph refers to a knowledge representation model that is built based on multi-source data and can be updated and evolved in real time. It uses a graph structure to associate and integrate entities and relationships in historical clause data, risk case data and regulatory policy data, representing the dynamic changes and complex relationships of knowledge in the insurance clause domain, and providing an intelligent knowledge reasoning basis for clause conflict detection.

[0042] Step S203: Receive the terms combination data of the target object.

[0043] The target entity refers to the specific insurance business entity that needs to undergo clause combination conflict detection and optimization. This could be an insurance product plan designed by an insurance company for a specific customer group, or a single policyholder's insurance clause combination, representing the specific application object of clause conflict detection. For example, a company's group accident insurance clause combination purchased for its employees.

[0044] Among them, the clause combination data refers to a complete data set designed for the target object, consisting of multiple insurance clauses, including the specific textual content of each clause, the logical relationships between clauses, and the applicable conditions. For example, it includes the main insurance clauses, supplementary insurance clauses, and the rules governing their interrelationships.

[0045] Step S204: Based on the dynamic knowledge graph, the global risk value of the clause combination data is calculated through a preset graph neural network. The global risk value reflects the degree of conflict between the clauses in the clause combination data.

[0046] Graph neural networks (GNNs) are deep learning models that learn from graph-structured data. By transmitting and aggregating information at nodes and edges in a dynamic knowledge graph, they can automatically learn the complex relationships and potential conflict patterns between clauses in a combination of clauses. Examples include graph convolutional networks (CNNs) and graph attention networks (GANs).

[0047] The global risk value is a quantitative indicator that reflects the overall conflict level of a clause combination, calculated using a graph neural network. For example, it ranges from 0 to 1, with higher values ​​indicating a higher risk of conflict.

[0048] Each clause refers to a single insurance clause that constitutes the clause combination data. These clauses include different types such as main insurance clauses, supplementary insurance clauses, and special agreement clauses. Each clause has its own independent rules regarding insurance liability, exclusions, and premium calculation, representing the basic building blocks of the clause combination.

[0049] Among them, the degree of conflict refers to the degree of inconsistency or contradiction between the various clauses in terms of insurance liability, scope of exclusion, premium calculation, and applicable conditions.

[0050] Step S205: If the global risk value is greater than the preset threshold, then generate conflict attribution information for the clause combination data.

[0051] The threshold refers to a pre-set global risk value threshold used to determine whether there are conflicts in the clause combination data. It can be determined based on the compliance requirements and risk tolerance of the insurance business. For example, a threshold of 0.7 can be set, and a clause combination is considered to have a significant conflict when the global risk value is greater than this value.

[0052] Among them, conflict attribution information refers to detailed information generated by analyzing the results of dynamic knowledge graphs and graph neural networks when the global risk value of the combined clause data exceeds a preset threshold. This information explains the reasons for clause conflicts and may include the clause number involved, the type of conflict (such as overlapping liability, loopholes in exemptions, etc.), and the scope of the conflict's impact. For example, "The main insurance clause and the supplementary insurance clause have overlapping definitions of accidental disability liability, which may lead to double payouts."

[0053] Step S206: Based on the conflict attribution information, adjust the clause combination data to obtain the target clause combination for the target object.

[0054] Among them, adjustment refers to the process of modifying and optimizing the clause combination data of the target object based on conflict attribution information. This may include operations such as deleting conflicting clauses, modifying the content of clauses, and adjusting the logical relationship between clauses, with the aim of eliminating the conflict risk in the clause combination and improving the compliance and risk control capabilities of the clauses.

[0055] Among them, the target clause combination refers to the adjusted clause combination data that meets compliance requirements and risk control standards, and is the final output of the clause conflict detection and optimization process.

[0056] This application's embodiments construct a dynamic knowledge graph by acquiring multi-source data such as historical clauses, risk cases, and regulatory policies, breaking through the static limitations of traditional rule bases. The dynamic knowledge graph can be updated and evolved in real time. Combined with graph neural networks, it can deeply mine implicit and complex contradictory relationships in clause combinations, accurately calculate global risk values, effectively solve the problem of static rules making it difficult to identify implicit conflicts, and significantly improve conflict coverage. Furthermore, by comparing the global risk value with preset thresholds to generate conflict attribution information and adjusting clause combinations accordingly, it achieves automated and intelligent clause conflict detection, significantly improving the accuracy and efficiency of clause combination conflict detection.

[0057] In some optional implementations of this embodiment, step 202, constructing a dynamic knowledge graph based on clause data, risk case data, and regulatory policy data, specifically includes the following steps:

[0058] Node identification is performed on clause data, risk case data, and regulatory policy data to obtain graph nodes of clause elements, risk tags, regulatory provisions, and historical cases; a semantic similarity algorithm is used to calculate the semantic similarity between graph nodes to obtain the association strength between graph nodes; based on the association strength, the association paths between graph nodes are established to generate a dynamic knowledge graph.

[0059] Node identification refers to the process of accurately extracting entity information with specific semantic and business meanings from multi-source data and transforming it into structured nodes that can be processed in a dynamic knowledge graph. For example, insurance liability, exclusions, and other clause elements can be identified from historical clause data as graph nodes, and risk labels such as cause of loss and loss type can be extracted from risk case data as graph nodes.

[0060] Among them, the clause elements refer to the key business information units that constitute the insurance clause text. They are extracted from historical clause data through natural language processing technology, representing the core content and rules of the insurance clause, and are used as graph nodes in the dynamic knowledge graph to describe the features of the clause.

[0061] Among them, risk labels refer to the identification information formed after classifying and annotating the risk events in risk case data. They are obtained through text mining and semantic analysis of risk case data, and represent the characteristics and attributes of different types of risks. They are used to identify risk-related graph nodes in dynamic knowledge graphs.

[0062] Among them, regulatory provisions refer to the specific provisions of laws, regulations and normative requirements that are clearly stipulated in insurance regulatory policy data and must be followed by insurance clauses. They are directly extracted from the text of regulatory policy data and used as graph nodes of constraint clause elements in dynamic knowledge graphs.

[0063] Among them, historical cases refer to complete records of representative risk events that actually occurred in the insurance industry and their handling results. They are derived from specific case texts in the risk case data, representing the risk scenarios and consequences that insurance clauses may face in actual application, and are used to provide practical risk experience references for dynamic knowledge graphs.

[0064] In this context, a knowledge graph node refers to the basic building block of a dynamic knowledge graph. It represents various business entities and concepts in the insurance clause domain, such as clause elements, risk tags, regulatory provisions, and historical cases, obtained through node identification of multi-source data. This knowledge graph serves to construct the structural framework and store business knowledge. Examples include a knowledge graph node for "Accidental Injury Insurance Liability," a knowledge graph node for "Fire Risk" risk tags, and a knowledge graph node for "Injunctions Exempting Insurers from Legally Liable Compensation."

[0065] Semantic similarity algorithms refer to mathematical models and calculation methods used to calculate the degree of semantic similarity between two texts or semantic representations, specifically for calculating the degree of semantic association between graph nodes. Examples include cosine similarity algorithms and word vector-based similarity calculation methods.

[0066] Semantic similarity refers to a quantitative indicator calculated using a semantic similarity algorithm, reflecting the degree of semantic association between two graph nodes. A higher value indicates higher semantic similarity, representing the closeness of graph nodes at the semantic level. It is used to measure the strength of semantic association between nodes such as clause elements and risk labels. For example, the semantic similarity between the "Accidental Injury Insurance Liability" graph node and the "Personal Accident Disability Compensation" graph node is 0.8, indicating that they are highly semantically related.

[0067] Among them, association strength refers to a comprehensive index that describes the degree of association between graph nodes, determined based on semantic similarity calculation results and other business rules. It characterizes the degree of association between graph nodes in actual business and is used to establish association paths between graph nodes.

[0068] Among them, the association path refers to a path with certain business logic significance that connects two or more graph nodes in a dynamic knowledge graph. It is established using graph algorithms (such as shortest path algorithms, random walk algorithms, etc.) based on the association strength. It represents the business association relationship and knowledge transfer path between graph nodes and is used to analyze potential conflict relationships between nodes such as clause elements and risk tags in clause conflict detection.

[0069] In one example, a property insurance company possesses a large amount of historical policy data (such as the insured amount and compensation scope for different car models in auto insurance policies), risk case data (such as claims cases of a large number of vehicles being flooded due to heavy rain in a certain region), and regulatory policy data (such as regulatory provisions regarding the fluctuation of auto insurance rates). First, node identification is performed on the aforementioned multi-source data. From the historical policy data, graph nodes are extracted as policy elements, such as "vehicle damage insurance insured amount" and "third-party liability insurance compensation limit"; from the risk case data, graph nodes are labeled with risk tags such as "heavy rain risk" and "vehicle flooding loss"; from the regulatory policy data, graph nodes are identified as regulatory provisions, such as "auto insurance rate fluctuation shall not exceed a certain percentage"; and specific risk cases are used as graph nodes for historical cases. Next, a cosine similarity algorithm is used to calculate the semantic similarity between graph nodes, thereby obtaining the association strength. For example, "vehicle damage insurance insured amount" and "vehicle flooding loss" are closely related in auto insurance claims scenarios, resulting in a high association strength. Based on the strength of association, a shortest path algorithm is used to establish association paths, generating a dynamic knowledge graph. This dynamic knowledge graph enables a more comprehensive analysis of the complex relationships between clauses, allowing for the early detection of hidden conflicts, improving the accuracy of clause conflict detection, and reducing blind spots in risk prevention and control.

[0070] This application's embodiments accurately extract graph nodes such as clause elements and risk tags by identifying nodes in clause data, risk case data, and regulatory policy data. This lays the foundation for a comprehensive analysis of insurance clause-related elements and effectively solves the problem that static rules cannot cover all information. A semantic similarity algorithm is used to calculate the semantic similarity between graph nodes to determine the strength of association, enabling in-depth mining of implicit and complex semantic connections between clauses and overcoming the limitations of traditional methods in identifying implicit conflicts. Based on the strength of association, a dynamic knowledge graph is generated by establishing association paths, which can reflect the relationships between nodes in real time, reducing risk blind spots caused by insufficient processing of unstructured data and greatly improving the accuracy and comprehensiveness of clause combination conflict detection.

[0071] In some optional implementations, the step "identifying nodes in clause data, risk case data, and regulatory policy data to obtain a graph of clause elements, risk labels, regulatory provisions, and historical cases" specifically includes the following steps:

[0072] A pre-trained language model is used to semantically encode clause data, risk case data, and regulatory policy data to obtain semantic feature vectors. A sequence labeling model is used to identify entities from the semantic feature vectors to obtain a set of key entities, which includes clause element entities, risk label entities, regulatory clause entities, and historical case entities. Based on the clause element entities, risk label entities, regulatory clause entities, and historical case entities, graph nodes for clause elements, risk labels, regulatory clauses, and historical cases are generated.

[0073] Among them, the pre-trained language model is a model that has been trained in advance on a large-scale general text dataset through unsupervised learning. It is used to perform preliminary processing on the input insurance-related text data, such as terms data, risk case data, and regulatory policy data, to provide a foundation for subsequent analysis.

[0074] Semantic encoding refers to using pre-trained language models to convert textual information such as input clause data, risk case data, and regulatory policy data into a numerical representation that computers can understand and process.

[0075] The semantic feature vector is a numerical vector obtained after semantic encoding, used to accurately describe the semantics of text in a computer. For example, the vector obtained after encoding the text "The types of stored goods include lithium batteries" contains the semantic features of the text regarding stored goods and lithium batteries.

[0076] Sequence labeling models are a type of machine learning model that analyzes a sequence of input semantic feature vectors and assigns a label category to each element in the sequence to identify entities with specific meanings from the semantic feature vectors. For example, Hidden Markov Models (HMMs) can be used to label the semantic feature vector sequence of insurance text to identify key entities within it.

[0077] Entity recognition is a process that uses sequence labeling models to process semantic feature vectors, identify entities with specific semantics from text, and determine their categories. For example, it can identify entities such as "car insurance terms," ​​"rainstorm," and "vehicle damage" from the statement "a certain car insurance policy stipulates that vehicle damage caused by heavy rain is within the scope of compensation."

[0078] Among them, the key entity set is a set of entities with specific meanings obtained after entity identification. These entities contain key information from data such as insurance terms, risk cases, and regulatory policies, and are used to generate nodes for the knowledge graph.

[0079] Among them, the clause element entities are entities identified from the insurance clause text that constitute the basic content of the insurance clause. They represent the key components of the insurance clause, clarify the specific provisions and scope of the insurance clause, and provide a basis for generating knowledge graph nodes and detecting clause conflicts. For example, "insured," "insured amount," and "insurance period" are common clause element entities.

[0080] Among them, risk label entities are identified from risk case data and used to describe the type and characteristics of risks. They reflect the various risk situations that may be faced in insurance business and are used to identify risk information in the knowledge graph to help analyze the relationship between terms and risks. Examples of risk label entities include "earthquake risk," "theft risk," and "disease risk."

[0081] Among them, regulatory clause entities are entities identified from regulatory policy data that represent the specific provisions of insurance regulatory regulations. They embody the legal requirements that insurance businesses must comply with and are used to integrate regulatory information into the knowledge graph to ensure compliance checks on the clauses. Examples of regulatory clause entities include "restrictions on the proportion of insurance funds used" and "regulations on insurance premium rate approval."

[0082] Historical case entities are identified from historical risk case data and record actual risk events and related handling in past insurance business. They provide a reference for analyzing the risks of current clauses and are used to link historical risk information in the knowledge graph to help predict and prevent similar risks. For example, "a claim case in a certain region where a large number of houses were damaged by floods" can be used as a historical case entity.

[0083] In one example, taking auto insurance as an example in an insurance clause conflict detection scenario, multi-source data is first acquired, including historical auto insurance clause data, such as "vehicle damage insurance coverage is determined based on the new vehicle purchase price"; risk case data, such as claims cases of numerous vehicles damaged by flooding due to heavy rain in a certain region; and regulatory policy data, such as "regulations on the floating range of auto insurance rates". A pre-trained language model is used to semantically encode the above data, converting the text into semantic feature vectors, which accurately preserve the semantic information of the text. Next, a sequence labeling model is used to perform entity recognition on the semantic feature vectors, obtaining a set of key entities, including clause element entities "vehicle damage insurance coverage" and "new vehicle purchase price", risk label entities "heavy rain" and "vehicle flooding", and regulatory clause entity "auto insurance rate floating range". Based on these entities, graph nodes are generated to construct a dynamic knowledge graph. This approach allows for in-depth analysis of the complex relationships between terms, risks, and regulations, effectively identifying hidden conflicts. For example, it can uncover potential contradictions between high-coverage vehicle damage insurance terms and exclusion clauses that do not cover the risk of heavy rain, thereby increasing conflict coverage, reducing blind spots in risk identification, and providing a solid foundation for subsequent accurate detection of conflicts in clause combinations.

[0084] This application employs a pre-trained language model to semantically encode clause data, risk case data, and regulatory policy data. This enables deep mining of textual semantic information, transforming complex text into precise semantic feature vectors. This effectively addresses the shortcomings of existing technologies in processing unstructured data and reduces blind spots in risk identification. Next, a sequence labeling model is used for entity recognition, accurately obtaining a set of key entities covering multiple important entities. Based on these entities, graph nodes are generated, comprehensively clarifying the complex relationships between clauses, risks, and regulations. This breaks the limitations of static rules, accurately identifies implicit and complex contradictions between clauses, and significantly improves the coverage of clause conflict detection. This lays a solid foundation for subsequent generation of dynamic knowledge graphs and accurate detection of clause combination conflicts.

[0085] In some optional implementations, step S204, based on a dynamic knowledge graph, calculates the global risk value of the clause combination data through a preset graph neural network, specifically including the following steps:

[0086] Extract graph nodes and edge relationships related to the clause combination data from the dynamic knowledge graph; process the graph nodes and edge relationships using a graph neural network to calculate node risk weights and edge relationship conflict coefficients; obtain the first dynamic adjustment parameter, the second dynamic adjustment parameter, and the regulatory policy adjustment parameter; and calculate the global risk value of the clause combination data based on the node risk weights, edge relationship conflict coefficients, the first dynamic adjustment parameter, the second dynamic adjustment parameter, and the regulatory policy adjustment parameter.

[0087] The global risk value R can be calculated using the formula R = α·Σ (node ​​risk weight) + β·Σ (edge ​​relationship conflict coefficient) + γ·Δ. Here, α is the first dynamic adjustment parameter, β is the second dynamic adjustment parameter, and γ·Δ is the regulatory policy adjustment parameter (i.e., a dynamic adjustment term). Specifically, γ, which constitutes the dynamic adjustment term, is a fixed adjustment coefficient, while Δ is a value that dynamically adjusts with changes in regulatory policies.

[0088] In this context, "relevant" refers to the graph nodes and edges selected from the dynamic knowledge graph that have a logical connection or informational correspondence with the clause combination data of the target object. For example, if the clause combination data involves theft insurance for auto insurance, then the "relevant" graph nodes may include historical theft insurance claim case nodes, regulatory policy clauses regarding theft insurance coverage limits, etc.

[0089] Among them, edge relationships refer to directed or undirected logical connections between different graph nodes, representing the inherent connections and interaction logic between different clause elements, risk labels, regulatory provisions and historical cases.

[0090] The processing refers to a series of complex calculations and feature extraction operations performed by graph neural networks on the graph nodes and edge relationships related to the clause combination data extracted from the dynamic knowledge graph. For example, the graph neural network calculates the weight value of each graph node in the overall risk through feature aggregation and propagation of graph nodes and edge relationships.

[0091] Among them, node risk weight refers to the proportion of each node in the overall risk assessment obtained after processing the graph nodes related to the clause combination data in the dynamic knowledge graph through graph neural network.

[0092] Among them, the edge relationship conflict coefficient refers to the value obtained after processing the edge relationships related to the clause combination data in the dynamic knowledge graph using graph neural networks, which reflects the degree of conflict between the nodes at both ends of the edge relationship.

[0093] Among them, the first dynamic adjustment parameter is a parameter value that can be dynamically adjusted according to the actual situation. It represents the relative importance of node risk weight in global risk assessment and is used to flexibly adjust the contribution ratio of node risk weight to global risk value.

[0094] The second dynamic adjustment parameter is a parameter that can change dynamically according to the actual business situation. It represents the relative impact of the edge relationship conflict coefficient on the global risk and is used to flexibly adjust the weight of the edge relationship conflict coefficient in the calculation of the global risk value.

[0095] Among them, the regulatory policy adjustment parameter is derived from the quantitative analysis of the impact of regulatory policies on the risk of insurance clauses. It represents the adjustment effect of changes in regulatory policies on the overall risk of the clause combination data. It is used to consider the impact of regulatory policy factors on risk assessment when calculating the overall risk value, so that the risk assessment results are more in line with the actual business environment and regulatory requirements.

[0096] In one example, taking auto insurance as an example, the system receives auto insurance policy combination data for a target object, including high-coverage main insurance and specific exclusion clauses. First, relevant graph nodes and edge relationships are extracted from a constructed dynamic knowledge graph, such as historical high-coverage auto insurance claim case nodes, regulatory restrictions on auto insurance coverage clauses nodes, and the edge relationships connecting them. Through graph neural network processing, node risk weights are calculated, such as a risk weight of 0.6 for high-coverage main insurance nodes and 0.4 for specific exclusion clause nodes; the edge relationship conflict coefficient is 0.5. The first dynamic adjustment parameter α = 0.4 and the second dynamic adjustment parameter β = 0.3 are obtained. Recent regulatory policies have strengthened control over high-coverage auto insurance, and the calculated regulatory policy adjustment parameter γ·Δ = 0.2. According to the formula R = α·Σ(node ​​risk weight) + β·Σ(edge ​​relationship conflict coefficient) + γ·Δ, the global risk value R = 0.4 × (0.6 + 0.4) + 0.3 × 0.5 + 0.2 = 0.65 is calculated. Because the value exceeds the preset threshold, conflict attribution information is generated. Based on this, the combination of clauses is adjusted, which improves the compliance of the clauses and reduces blind spots in risk prevention and control.

[0097] This application's embodiments accurately extract graph nodes and edge relationships related to clause combination data from a dynamic knowledge graph, comprehensively covering explicit and implicit associations between clauses and overcoming the limitations of static rules. Utilizing graph neural networks to process these nodes and edge relationships allows for the scientific calculation of node risk weights and edge relationship conflict coefficients, effectively uncovering potential risks. Furthermore, combining the acquired first and second dynamic adjustment parameters with regulatory policy adjustment parameters to calculate the global risk value enables flexible adaptation to different business scenarios and regulatory changes. This significantly improves the accuracy of clause combination conflict detection and reduces blind spots in risk identification.

[0098] In some optional implementations, after generating conflict attribution information for the clause combination data in step S205 if the global risk value is greater than a preset threshold, the following steps are also included:

[0099] Based on global risk values ​​and clause combination data, a conflict attribution heatmap is generated, which marks the conflict paths between clauses in the clause combination data. The conflict attribution heatmap is displayed through an interactive interface. When a triggering operation targeting a conflict path is detected, the conflict attribution information corresponding to the conflict path is displayed on the interactive interface.

[0100] The conflict attribution heatmap is a visualization chart generated based on a dynamic knowledge graph and related calculation results. It obtains information from a dynamic knowledge graph constructed from historical clause data, risk case data, and regulatory policy data. Through a preset algorithm, it uses different shades of color to represent the degree of conflict between clauses in a clause combination; for example, a darker color indicates a higher degree of conflict. This heatmap is used to visually present the distribution of potential conflicts in clause combinations and quickly locate areas of concentrated conflict. In insurance clause conflict detection scenarios, it allows business personnel to quickly understand which clause combinations pose significant risks, providing clear guidance for subsequent clause adjustments and risk control. For example, a darker color in a car insurance clause combination indicates a conflict between a high-coverage main insurance policy and a specific exclusion clause.

[0101] Among them, a conflict path is one or more related paths identified in the conflict attribution process. It represents the logical connection and transmission route of conflicts between clauses in a clause portfolio, revealing how conflicts originate from some clauses and affect others. For example, in a health insurance clause portfolio, there may be a conflict path that starts from a specific disease coverage clause, passes through the deductible clause, and ultimately affects the reimbursement ratio clause. This path clearly demonstrates the generation and transmission process of conflicts between clauses.

[0102] The interactive interface displays key information such as conflict attribution heatmaps and conflict attribution information corresponding to conflict paths. Users can obtain more detailed conflict attribution explanations by performing actions on the interface, such as clicking on specific nodes or line segments on the conflict path.

[0103] In one example, let's take the conflict detection of corporate property insurance clauses by a property insurance company. First, a dynamic knowledge graph is constructed based on historical clauses, risk cases, and regulatory policy data. After receiving clause combination data, a graph neural network calculates a global risk value of 0.7 (with a preset threshold of 0.5). Based on this global risk value and clause combination data, a conflict attribution heatmap is generated. The heatmap uses different shades of color to mark the conflict paths between clauses, with darker colors indicating more severe conflicts. This heatmap is displayed through an interactive interface. Business personnel can click on darker-colored conflict paths, and the interface immediately displays the corresponding conflict attribution information, such as "stored goods contain flammable chemicals, but the exclusion clauses do not explicitly cover this risk." Based on this information, the insurance company adjusts the clause combinations and adds relevant exclusion clauses.

[0104] This application's embodiments generate conflict attribution heatmaps based on global risk values ​​and clause combination data. This provides an intuitive and visual representation of the conflict distribution among clauses within a clause combination, transforming abstract and complex conflict relationships into easily understandable color and graphic displays. This effectively solves the problem of hidden conflicts being difficult to detect. Marking conflict paths allows for precise location of the logical chain of conflict generation and propagation. The user-friendly interface quickly displays corresponding conflict attribution information after an action is triggered, helping business personnel quickly understand the root cause of the conflict.

[0105] In some optional implementations, step S206, based on conflict attribution information, adjusts the clause combination data to obtain the target clause combination for the target object, specifically including the following steps:

[0106] Based on conflict attribution information, multiple conflict fields are identified from the clause combination data; a clause adjustment algorithm is used to resolve conflicts in each conflict field to obtain the target clause combination of the target object.

[0107] Among them, multiple conflict fields represent the specific data fields that cause conflicts in the policy terms. These fields involve various key elements of the policy terms, such as the scope of insurance liability, exclusions, sum insured, and insurance period. For example, in a health insurance policy term set, it was found that the "Specific Disease Coverage" field and the "Pre-existing Condition Exclusion" field conflict. These two fields are multiple conflict fields. By identifying them, the policy term conflict issues can be resolved in a targeted manner.

[0108] The clause adjustment algorithm is a computational method built upon insurance business rules, risk assessment models, and interconnected information in a dynamic knowledge graph. This algorithm characterizes the rules and processes for reasonably modifying and optimizing clauses based on conflict attribution information and multiple conflict fields. For example, a clause adjustment algorithm based on risk weight allocation can dynamically adjust the clause content of each field according to the degree of impact of different conflict fields on the overall risk.

[0109] Conflict resolution refers to the process of using clause adjustment algorithms to process multiple conflicting fields in order to resolve contradictions and conflicts between clauses in a clause combination. Starting from the identified conflicting fields, it modifies, supplements, or deletes these fields according to the rules and strategies determined by the clause adjustment algorithm. For example, regarding the conflicts in the aforementioned health insurance clause combination, conflict resolution can be used to remove diseases related to pre-existing conditions from the "Specific Disease Coverage" section, or to provide more detailed explanations of the "Pre-existing Condition Exclusion" clause, thereby eliminating the conflict between the two.

[0110] In one example, we'll examine the handling of policy conflicts in a comprehensive protection plan from a large life insurance company. This plan includes coverage for critical illness, medical expenses, and accidents, resulting in a complex combination of policy terms. After preliminary steps, the calculated global risk value exceeds a preset threshold, generating conflict attribution information that reveals a conflict between the "payment conditions for a specific rare disease in critical illness insurance" and the "reimbursement scope for treatment costs of the same rare disease in medical insurance." Based on this conflict attribution information, these two related fields are identified as multiple conflicting fields within the policy combination data. Subsequently, a policy adjustment algorithm is employed, which comprehensively considers factors such as disease incidence, treatment costs, and industry practices. For the critical illness insurance payout conditions field, the algorithm is adjusted to explicitly state that payment is made upon diagnosis of the specific rare disease; for the medical insurance reimbursement scope field, the algorithm is expanded to cover all reasonable and necessary treatment costs for the rare disease. After conflict resolution, the target policy combination is obtained.

[0111] This application's embodiments accurately locate multiple conflict fields from clause combination data based on conflict attribution information. This method overcomes the limitation of traditional static rule detection, which can only detect explicit conflicts, and can delve deeper into implicit and complex contradictions between clauses, effectively improving conflict coverage. A clause adjustment algorithm is used to resolve conflicts specifically for each conflict field, intelligently adjusting clause content according to specific conflict situations, avoiding the subjectivity and uncertainty of manual adjustments. The final target clause combination ensures both the compliance of the clauses and improves the accuracy of clause combination conflict detection.

[0112] In some optional implementations, after adjusting the clause combination data based on conflict attribution information to obtain the target clause combination of the target object in step S206, the following steps are further included:

[0113] When a change in regulatory policy is detected, the target regulatory provision is acquired as the new regulatory policy data. Based on the new regulatory policy data, the dynamic knowledge graph is updated to obtain the updated dynamic knowledge graph. The explicit rules in the target clause combination are validated through a hard-coded rule engine to obtain the explicit rule validation results. The implicit rules in the target clause combination are constrained and validated through the updated dynamic knowledge graph to obtain the implicit rule validation results. Based on the explicit rule validation results and the implicit rule validation results, the compliance verification result of the target clause combination is determined.

[0114] Among them, the target regulatory provisions are derived from the specific policy clauses involved in changes to regulatory policies. They represent the newly issued, amended, or repealed binding written regulations in the field of insurance business supervision.

[0115] Among them, the hard-coded rule engine is a rule processing module built based on pre-written fixed code logic. It originates from the code implementation of explicit and stable rules during program development and represents a set of rules stored and executed in hard-coded form.

[0116] Explicit rules are those directly stated in insurance clauses, regulatory policies, and other texts. They represent business norms with clear and intuitive judgment standards, used to regulate behaviors or conditions in insurance business that can be directly quantified and judged. For example, an insurance clause stipulating that "the insured must be between 18 and 60 years old to be eligible for insurance" is an explicit rule, which can be directly used by a hard-coded rule engine to verify the insured's age information.

[0117] The explicit rule validation result is the conclusive information obtained by the hard-coded rule engine after validating the explicit rules in the target clause combination.

[0118] Implicit rules are hidden behind insurance business data, historical terms, and risk cases. They are difficult to express explicitly in words, but can be discovered through data correlation and logical reasoning. They are used to discover implicit and complex contradictions and conflicts in the combination of terms. For example, by analyzing historical risk cases, it can be found that "when the main insurance coverage is too high and the exclusion clause does not cover specific emerging risks, it is easy to cause claims disputes," which is an implicit rule.

[0119] Among them, constraint verification refers to the process of using the updated dynamic knowledge graph to perform logical reasoning and correlation analysis on the implicit rules in the target clause combination in order to determine whether the clauses meet the implicit rule constraints.

[0120] The implicit rule verification result is a conclusive information based on the constraint verification process, indicating whether the implicit rules in the target clause combination are satisfied.

[0121] In one example, in the auto insurance business of a property insurance company, the regulatory authorities issued a new policy stipulating that "for compensation for battery damage of new energy vehicles in auto insurance, the deductible must be clearly defined as 10% of the battery value." This is the target regulatory provision. This provision is obtained as new regulatory policy data, and the dynamic knowledge graph is updated to include information related to this new rule. For a set of auto insurance clauses to be reviewed, the explicit rules are first validated using a hard-coded rule engine, such as checking whether the insurance period is within the prescribed range. The explicit rule validation result is deemed compliant. Then, using the updated dynamic knowledge graph, the implicit rules in the clauses, such as those concerning compensation for new energy vehicle batteries, are constrained and validated. It is found that the original clauses do not clearly define the deductible limit, resulting in an implicit rule validation result of non-compliance. Based on the above two validation results, the compliance verification result for this clause set is determined to be unsuccessful.

[0122] This application's embodiments, upon detecting changes in regulatory policies and obtaining the updated dynamic knowledge graph of the target regulatory provisions, ensure that the knowledge graph keeps pace with the latest regulatory requirements, providing real-time and accurate evidence for clause verification. By verifying explicit rules through a hard-coded rule engine, it can quickly and accurately determine whether clauses comply with explicit regulations, such as rates and terms. Utilizing the updated dynamic knowledge graph to constrain and verify implicit rules can uncover complex relationships between clauses and identify potential conflicts, such as mismatches between coverage amounts and exclusions. Combining the verification results of explicit and implicit rules to determine compliance comprehensively covers clause issues and effectively improves the accuracy and efficiency of clause conflict detection.

[0123] It should be emphasized that, in order to further ensure the privacy and security of the aforementioned historical clause data, risk case data, regulatory policy data, clause combination data, global risk value, conflict attribution information, and target clause combination, these data can also be stored in a blockchain node.

[0124] The blockchain referred to in this application is a novel application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanisms, and encryption algorithms. Essentially, a blockchain is a decentralized database, a chain of data blocks linked together using cryptographic methods. Each data block contains information about a batch of network transactions, used to verify the validity of the information (anti-counterfeiting) and generate the next block. A blockchain can include an underlying blockchain platform, a platform product service layer, and an application service layer.

[0125] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0126] Foundational technologies for artificial intelligence generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies mainly encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0127] 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 instructing related hardware with computer-readable instructions. These computer-readable instructions can be stored in a computer-readable storage medium. When executed, the program can include the processes of the embodiments of the above methods. The aforementioned storage medium can be a non-volatile storage medium such as a magnetic disk, optical disk, or read-only memory (ROM), or random access memory (RAM).

[0128] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0129] Further reference Figure 3 As a response to the above Figure 2 To implement the method shown, this application provides an embodiment of a knowledge graph-based clause conflict detection device, which is similar to... Figure 2 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.

[0130] like Figure 3 As shown, the knowledge graph-based clause conflict detection device 400 of this embodiment includes: an acquisition module 401, a construction module 402, a receiving module 403, a calculation module 404, a generation module 405, and an adjustment module 406. Wherein:

[0131] Module 401 is used to acquire multi-source data, which includes historical terms data, risk case data, and regulatory policy data.

[0132] Module 402 is used to build a dynamic knowledge graph based on historical clause data, risk case data, and regulatory policy data.

[0133] Receiver module 403 is used to receive the terms combination data of the target object;

[0134] The calculation module 404 is used to calculate the global risk value of the clause combination data based on the dynamic knowledge graph and through the preset graph neural network. The global risk value reflects the degree of conflict between the clauses in the clause combination data.

[0135] The generation module 405 is used to generate conflict attribution information for the clause combination data if the global risk value is greater than a preset threshold.

[0136] The adjustment module 406 is used to adjust the clause combination data based on conflict attribution information to obtain the target clause combination of the target object.

[0137] This embodiment constructs a dynamic knowledge graph by acquiring multi-source data such as historical clauses, risk cases, and regulatory policies, breaking through the static limitations of traditional rule bases. The dynamic knowledge graph can be updated and evolved in real time. Combined with graph neural networks, it can deeply mine implicit and complex contradictory relationships in clause combinations, accurately calculate global risk values, effectively solve the problem of static rules making it difficult to identify implicit conflicts, and significantly improve conflict coverage. Furthermore, by comparing the global risk value with preset thresholds to generate conflict attribution information and adjusting clause combinations accordingly, it achieves automation and intelligence in clause conflict detection, significantly improving the accuracy and efficiency of clause combination conflict detection.

[0138] In one embodiment, the construction module 402 includes:

[0139] The identification submodule is used to identify nodes in clause data, risk case data, and regulatory policy data to obtain graph nodes of clause elements, risk labels, regulatory provisions, and historical cases;

[0140] The similarity calculation submodule is used to calculate the semantic similarity between graph nodes using a semantic similarity algorithm, thereby obtaining the association strength between graph nodes;

[0141] A submodule is created to establish the association paths between graph nodes based on the association strength, thereby generating a dynamic knowledge graph.

[0142] This application's embodiments accurately extract graph nodes such as clause elements and risk tags by identifying nodes in clause data, risk case data, and regulatory policy data. This lays the foundation for a comprehensive analysis of insurance clause-related elements and effectively solves the problem that static rules cannot cover all information. A semantic similarity algorithm is used to calculate the semantic similarity between graph nodes to determine the strength of association, enabling in-depth mining of implicit and complex semantic connections between clauses and overcoming the limitations of traditional methods in identifying implicit conflicts. Based on the strength of association, a dynamic knowledge graph is generated by establishing association paths, which can reflect the relationships between nodes in real time, reducing risk blind spots caused by insufficient processing of unstructured data and greatly improving the accuracy and comprehensiveness of clause combination conflict detection.

[0143] In one embodiment, the identification submodule is further configured to use a pre-trained language model to semantically encode the clause data, risk case data, and regulatory policy data to obtain a semantic feature vector; to use a sequence labeling model to perform entity identification on the semantic feature vector to obtain a set of key entities, which includes clause element entities, risk label entities, regulatory clause entities, and historical case entities; and to generate graph nodes for clause elements, risk labels, regulatory clauses, and historical cases based on the clause element entities, risk label entities, regulatory clause entities, and historical case entities.

[0144] This application employs a pre-trained language model to semantically encode clause data, risk case data, and regulatory policy data. This enables deep mining of textual semantic information, transforming complex text into precise semantic feature vectors. This effectively addresses the shortcomings of existing technologies in processing unstructured data and reduces blind spots in risk identification. Next, a sequence labeling model is used for entity recognition, accurately obtaining a set of key entities covering multiple important entities. Based on these entities, graph nodes are generated, comprehensively clarifying the complex relationships between clauses, risks, and regulations. This breaks the limitations of static rules, accurately identifies implicit and complex contradictions between clauses, and significantly improves the coverage of clause conflict detection. This lays a solid foundation for subsequent generation of dynamic knowledge graphs and accurate detection of clause combination conflicts.

[0145] In one embodiment, the calculation module 404 includes:

[0146] The extraction submodule is used to extract graph nodes and edge relationships related to the clause combination data from the dynamic knowledge graph.

[0147] The processing submodule is used to process graph nodes and edge relationships through graph neural networks, and calculate node risk weights and edge relationship conflict coefficients.

[0148] The acquisition submodule is used to acquire the first dynamic adjustment parameter, the second dynamic adjustment parameter, and the regulatory policy adjustment parameter;

[0149] The risk value calculation submodule is used to calculate the global risk value of the clause combination data based on the node risk weight, edge relationship conflict coefficient, first dynamic adjustment parameter, second dynamic adjustment parameter and regulatory policy adjustment parameter.

[0150] This application's embodiments accurately extract graph nodes and edge relationships related to clause combination data from a dynamic knowledge graph, comprehensively covering explicit and implicit associations between clauses and overcoming the limitations of static rules. Utilizing graph neural networks to process these nodes and edge relationships allows for the scientific calculation of node risk weights and edge relationship conflict coefficients, effectively uncovering potential risks. Furthermore, combining the acquired first and second dynamic adjustment parameters with regulatory policy adjustment parameters to calculate the global risk value enables flexible adaptation to different business scenarios and regulatory changes. This significantly improves the accuracy of clause combination conflict detection and reduces blind spots in risk identification.

[0151] In one embodiment, the adjustment module 406 includes:

[0152] The determination submodule is used to identify multiple conflict fields from the clause combination data based on conflict attribution information;

[0153] The conflict resolution submodule is used to resolve conflicts in each conflict field using a clause adjustment algorithm to obtain the target clause combination of the target object.

[0154] This application's embodiments accurately locate multiple conflict fields from clause combination data based on conflict attribution information. This method overcomes the limitation of traditional static rule detection, which can only detect explicit conflicts, and can delve deeper into implicit and complex contradictions between clauses, effectively improving conflict coverage. A clause adjustment algorithm is used to resolve conflicts specifically for each conflict field, intelligently adjusting clause content according to specific conflict situations, avoiding the subjectivity and uncertainty of manual adjustments. The final target clause combination ensures both the compliance of the clauses and improves the accuracy of clause combination conflict detection.

[0155] In one embodiment, the knowledge graph-based clause conflict detection device 400 further includes:

[0156] The graph generation module is used to generate a conflict attribution heatmap based on the global risk value and clause combination data. The conflict attribution heatmap is marked with the conflict paths between clauses in the clause combination data.

[0157] The graph display module is used to display conflict attribution heatmaps through an interactive interface;

[0158] The information display module is used to display the conflict attribution information corresponding to the conflict path on the interactive interface when a triggering operation for the conflict path is detected.

[0159] This application's embodiments generate conflict attribution heatmaps based on global risk values ​​and clause combination data. This provides an intuitive and visual representation of the conflict distribution among clauses within a clause combination, transforming abstract and complex conflict relationships into easily understandable color and graphic displays. This effectively solves the problem of hidden conflicts being difficult to detect. Marking conflict paths allows for precise location of the logical chain of conflict generation and propagation. The user-friendly interface quickly displays corresponding conflict attribution information after an action is triggered, helping business personnel quickly understand the root cause of the conflict.

[0160] In one embodiment, the knowledge graph-based clause conflict detection device 400 further includes:

[0161] The data acquisition module is used to acquire the target regulatory provisions as new regulatory policy data when a change in regulatory policy is detected.

[0162] The update module is used to update the dynamic knowledge graph based on the newly added regulatory policy data, resulting in an updated dynamic knowledge graph.

[0163] The first verification module is used to verify the explicit rules in the target clause combination through the hard-coded rule engine and obtain the explicit rule verification results.

[0164] The second verification module is used to perform constraint verification on the implicit rules in the target clause combination through the updated dynamic knowledge graph, and obtain the implicit rule verification result.

[0165] The determination module is used to determine the compliance verification result of the target clause combination based on the explicit rule verification results and the implicit rule verification results.

[0166] This application's embodiments, upon detecting changes in regulatory policies and obtaining the updated dynamic knowledge graph of the target regulatory provisions, ensure that the knowledge graph keeps pace with the latest regulatory requirements, providing real-time and accurate evidence for clause verification. By verifying explicit rules through a hard-coded rule engine, it can quickly and accurately determine whether clauses comply with explicit regulations, such as rates and terms. Utilizing the updated dynamic knowledge graph to constrain and verify implicit rules can uncover complex relationships between clauses and identify potential conflicts, such as mismatches between coverage amounts and exclusions. Combining the verification results of explicit and implicit rules to determine compliance comprehensively covers clause issues and effectively improves the accuracy and efficiency of clause conflict detection.

[0167] To address the aforementioned technical problems, embodiments of this application also provide a computer device. Please refer to [link / reference needed]. Figure 4 , Figure 4 This is a basic structural block diagram of the computer device in this embodiment.

[0168] Computer device 6 includes a memory 61, a processor 62, and a network interface 63 that are interconnected via a system bus. It should be noted that only computer device 6 with memory 61, processor 62, and network interface 63 is shown in the figure; however, it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented alternatively. Those skilled in the art will understand that the computer device described herein is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0169] Computer devices can include desktop computers, laptops, handheld computers, and cloud servers. These devices allow for human-computer interaction with users through keyboards, mice, remote controls, touchpads, or voice-activated devices.

[0170] The memory 61 includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 61 may be an internal storage unit of the computer device 6, such as the hard disk or memory of the computer device 6. In other embodiments, the memory 61 may also be an external storage device of the computer device 6, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the computer device 6. Of course, the memory 61 may also include both the internal storage unit and the external storage device of the computer device 6. In this embodiment, the memory 61 is typically used to store the operating system and various application software installed on the computer device 6, such as computer-readable instructions for a knowledge graph-based clause conflict detection method. In addition, memory 61 can also be used to temporarily store various types of data that have been output or will be output.

[0171] In some embodiments, processor 62 may be a central processing unit (CPU), controller, microcontroller, microprocessor, or other data processing chip. Processor 62 is typically used to control the overall operation of computer device 6. In this embodiment, processor 62 is used to execute computer-readable instructions stored in memory 61 or to process data, such as executing computer-readable instructions for a knowledge graph-based clause conflict detection method.

[0172] The network interface 63 may include a wireless network interface or a wired network interface, which is typically used to establish a communication connection between the computer device 6 and other electronic devices.

[0173] This application's embodiments construct a dynamic knowledge graph by acquiring multi-source data such as historical clauses, risk cases, and regulatory policies, breaking through the static limitations of traditional rule bases. The dynamic knowledge graph can be updated and evolved in real time. Combined with graph neural networks, it can deeply mine implicit and complex contradictory relationships in clause combinations, accurately calculate global risk values, effectively solve the problem of static rules making it difficult to identify implicit conflicts, and significantly improve conflict coverage. Furthermore, by comparing the global risk value with preset thresholds to generate conflict attribution information and adjusting clause combinations accordingly, it achieves automated and intelligent clause conflict detection, significantly improving the accuracy and efficiency of clause combination conflict detection.

[0174] This application also provides another implementation, namely, providing a computer-readable storage medium storing computer-readable instructions that can be executed by at least one processor to cause the at least one processor to perform the steps of the knowledge graph-based clause conflict detection method described above.

[0175] This application's embodiments construct a dynamic knowledge graph by acquiring multi-source data such as historical clauses, risk cases, and regulatory policies, breaking through the static limitations of traditional rule bases. The dynamic knowledge graph can be updated and evolved in real time. Combined with graph neural networks, it can deeply mine implicit and complex contradictory relationships in clause combinations, accurately calculate global risk values, effectively solve the problem of static rules making it difficult to identify implicit conflicts, and significantly improve conflict coverage. Furthermore, by comparing the global risk value with preset thresholds to generate conflict attribution information and adjusting clause combinations accordingly, it achieves automated and intelligent clause conflict detection, significantly improving the accuracy and efficiency of clause combination conflict detection.

[0176] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods of the various embodiments of this application.

[0177] Obviously, the embodiments described above are only some embodiments of this application, not all embodiments. The accompanying drawings show preferred embodiments of this application, but do not limit the patent scope of this application. This application can be implemented in many different forms; rather, the purpose of providing these embodiments is to provide a more thorough and comprehensive understanding of the disclosure of this application. Although this application 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 specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this application's specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the scope of patent protection of this application.

[0178] The software tools or components not belonging to our company that appear in the embodiments of this application are merely examples and do not represent actual use.

Claims

1. A knowledge graph-based method for detecting clause conflicts, characterized in that, Includes the following steps: Acquire multi-source data, including historical terms data, risk case data, and regulatory policy data; A dynamic knowledge graph is constructed based on the historical terms data, the risk case data, and the regulatory policy data. Receive the terms and conditions combination data of the target object; Based on the dynamic knowledge graph, a global risk value is calculated using a preset graph neural network. The global risk value reflects the degree of conflict between the clauses in the clause combination data. If the global risk value is greater than a preset threshold, conflict attribution information for the clause combination data is generated. Based on the conflict attribution information, the clause combination data is adjusted to obtain the target clause combination for the target object.

2. The method according to claim 1, characterized in that, The step of constructing a dynamic knowledge graph based on the clause data, the risk case data, and the regulatory policy data specifically includes: By performing node identification on the aforementioned clause data, risk case data, and regulatory policy data, a graph node of clause elements, risk tags, regulatory provisions, and historical cases is obtained; A semantic similarity algorithm is used to calculate the semantic similarity between the graph nodes to obtain the association strength between the graph nodes; Based on the association strength, establish the association paths between the graph nodes to generate a dynamic knowledge graph.

3. The method as described in claim 2, characterized in that, The step of identifying nodes in the clause data, risk case data, and regulatory policy data to obtain a graph of clause elements, risk tags, regulatory provisions, and historical cases specifically includes: A pre-trained language model is used to semantically encode the clause data, the risk case data, and the regulatory policy data to obtain semantic feature vectors. By using a sequence labeling model, entity recognition is performed on the semantic feature vector to obtain a set of key entities, which includes clause element entities, risk label entities, regulatory clause entities, and historical case entities. Based on the aforementioned clause element entities, risk label entities, regulatory clause entities, and historical case entities, generate graph nodes for clause elements, risk labels, regulatory clauses, and historical cases.

4. The method according to claim 1, characterized in that, The step of calculating the global risk value of the clause combination data based on the dynamic knowledge graph and through a preset graph neural network specifically includes: Extract graph nodes and edge relationships related to the combined clause data from the dynamic knowledge graph; The graph nodes and edge relationships are processed using a graph neural network to calculate node risk weights and edge relationship conflict coefficients. Obtain the first dynamic adjustment parameter, the second dynamic adjustment parameter, and the regulatory policy adjustment parameter; Based on the node risk weight, the edge relationship conflict coefficient, the first dynamic adjustment parameter, the second dynamic adjustment parameter, and the regulatory policy adjustment parameter, the global risk value of the clause combination data is calculated.

5. The method according to claim 1, characterized in that, After the step of generating conflict attribution information for the clause combination data if the global risk value is greater than a preset threshold, the method further includes: Based on the global risk value and the clause combination data, a conflict attribution heatmap is generated, which marks the conflict paths between the clauses in the clause combination data. The conflict attribution heatmap is displayed through an interactive interface; When a triggering operation is detected for the conflict path, the conflict attribution information corresponding to the conflict path is displayed on the interactive interface.

6. The method according to claim 1, characterized in that, The step of adjusting the clause combination data based on the conflict attribution information to obtain the target clause combination for the target object specifically includes: Based on the conflict attribution information, multiple conflict fields are identified from the clause combination data; A clause adjustment algorithm is used to resolve conflicts in each conflict field, resulting in the target clause combination for the target object.

7. The method according to claim 1, characterized in that, After the step of adjusting the clause combination data based on the conflict attribution information to obtain the target clause combination of the target object, the method further includes: When a change in regulatory policy is detected, the target regulatory provision is obtained as the newly added regulatory policy data; Based on the newly added regulatory policy data, the dynamic knowledge graph is updated to obtain the updated dynamic knowledge graph; The explicit rules in the target clause combination are validated using a hard-coded rule engine to obtain the explicit rule validation results. The implicit rules in the target clause combination are constrained and verified using the updated dynamic knowledge graph to obtain the implicit rule verification results. Based on the explicit rule verification results and the implicit rule verification results, the compliance verification result of the target clause combination is determined.

8. A knowledge graph-based clause conflict detection device, characterized in that, include: The acquisition module is used to acquire multi-source data, including historical terms data, risk case data, and regulatory policy data. The construction module is used to build a dynamic knowledge graph based on the historical terms data, the risk case data, and the regulatory policy data; The receiving module is used to receive the terms combination data of the target object; The calculation module is used to calculate the global risk value of the clause combination data based on the dynamic knowledge graph and through a preset graph neural network. The global risk value reflects the degree of conflict between the clauses in the clause combination data. The generation module is used to generate conflict attribution information for the clause combination data if the global risk value is greater than a preset threshold. The adjustment module is used to adjust the clause combination data based on the conflict attribution information to obtain the target clause combination of the target object.

9. A computer device, characterized in that, The method includes a memory and a processor, wherein the memory stores computer-readable instructions, and the processor, when executing the computer-readable instructions, implements the steps of the knowledge graph-based clause conflict detection method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-readable instructions that, when executed by a processor, implement the steps of the knowledge graph-based clause conflict detection method as described in any one of claims 1 to 7.

Images