Insurance clause intelligent generation method, computer device and readable storage medium

Abstract

The application relates to the technical field of intelligent decision-making, is applicable to the financial field, and discloses an insurance clause intelligent generation method, a computer device and a readable storage medium. The method comprises the following steps: determining a constraint information list based on constraint weights and data source types of each piece of constraint information; identifying a conflict constraint pair and determining conflict performance information of the conflict constraint pair; executing a constraint conflict processing strategy matched with the conflict constraint pair on the conflict constraint pair based on the conflict performance information and priorities of two pieces of constraint information in the constraint information list; performing constraint supplement processing on the updated constraint information list based on a predetermined insurance domain knowledge graph; and determining input information for a predetermined large language model based on known generation guidance information of a target insurance product and the current constraint information list, and generating an insurance clause corresponding to the input information through the large language model. The scheme improves the efficiency and accuracy of insurance clause generation.

Description

Technical Field

[0001] This application relates to the field of intelligent decision-making technology, applicable to the financial sector, and particularly to a method for intelligently generating insurance terms, a computer device, and a readable storage medium. Background Technology

[0002] When intelligently generating insurance product terms, the process must simultaneously meet constraints from multiple dimensions, including regulatory requirements, business logic, and specific product design. These constraints are often numerous and complex, with potential conflicts between them. Current technologies primarily rely on general-purpose language models to directly generate insurance product terms, or on templates based on simple rules. These methods are simplistic, lack adaptability, and cannot effectively identify potential conflicts among numerous constraints. Consequently, they cannot make reasonable adjustments when constraints conflict, making it difficult to cope with the demanding and ever-changing insurance business environment.

[0003] Therefore, how to flexibly identify and resolve conflicts between constraints during the intelligent generation of insurance clauses, and improve the accuracy of the generated clauses, has become an urgent technical problem to be solved. Summary of the Invention

[0004] This application provides a method, computer device, and readable storage medium for intelligently generating insurance clauses, aiming to solve the technical problem in related technologies where the generation of insurance clauses is unreasonable due to the inability to identify and resolve conflicts between constraints.

[0005] In a first aspect, embodiments of this application provide a method for intelligently generating insurance clauses, including: Obtain multi-source constraint information for the target insurance product, and convert each constraint information in the multi-source constraint information into a quadruple, wherein the quadruple includes constraint scope, constraint predicate, constraint object and constraint weight; Based on the constraint weight and data source type of each constraint information, the priority of each constraint information is determined, and the quadruplets of each constraint information are arranged in descending order of priority to obtain a constraint information list; Based on the constraint information list, conflicting constraint pairs are identified, and conflict performance information of the conflicting constraint pairs is determined, wherein the conflict performance information is used to reflect the identity characteristics and conflict performance characteristics of the two constraint information in the conflicting constraint pair. Based on the conflict performance information and the priority of the two constraints in the conflict constraint pair in the constraint information list, a constraint conflict handling strategy matching the conflict constraint pair is executed on the conflict constraint pair to update the constraint information list. Based on a pre-defined knowledge graph in the insurance field, the updated constraint information list is supplemented with constraints to update the constraint information list again. Based on the known generation guidance information and the current list of constraint information for the target insurance product, input information is determined for a predetermined large language model, and the insurance terms corresponding to the input information are generated through the large language model.

[0006] In one embodiment of this application, optionally, the data source types of the multi-source constraint information include: regulatory constraint type, business constraint type, and insurance product constraint type. Then, before determining the priority of each constraint information based on its constraint weight and data source type, the method further includes: Obtain the regulatory weight, business impact, and product relevance of each constraint; The process of determining the priority of each constraint based on its constraint weight and data source type includes: For each piece of constraint information, obtain the first weight of the constraint information under the legal constraint type, obtain the second weight of the constraint information under the business constraint type, and obtain the third weight of the constraint information under the insurance product constraint type; Based on the first weight, the second weight, and the third weight, the regulatory weight, business impact, and product relevance are weighted and summed to obtain the priority of the constraint information.

[0007] In one embodiment of this application, optionally, obtaining the regulatory weight, business impact, and product relevance of each constraint information item includes: For each of the aforementioned constraint information If the constraint information is related to any regulation, the regulatory weight of the constraint information is determined based on the importance level of the regulation; if the constraint information is not related to any regulation, the regulatory weight of the constraint information is set to a first predetermined value. Determine the business risk level of the target insurance product when the constraint information is violated, and determine the business impact level of the constraint information based on the risk level; Obtain the constraint product type of the constraint information. If the constraint product type includes the target insurance product, determine the product relevance of the constraint information as a second predetermined value; otherwise, determine the product relevance of the constraint information as a third predetermined value.

[0008] In one embodiment of this application, optionally, identifying conflicting constraint pairs based on the constraint information list includes: Based on the constraint information list, a conflict detection matrix is ​​constructed, wherein the element in the m-th row and n-th column of the conflict detection matrix is ​​the conflict relationship value between the m-th constraint information and the n-th constraint information in the constraint information list; If the conflict relation value at any matrix position is non-zero, the row constraint information and column constraint information corresponding to the matrix position are determined to be a conflict constraint pair.

[0009] In one embodiment of this application, optionally, if the conflict relationship value at the matrix position is zero, the row constraint information and column constraint information corresponding to the matrix position do not conflict; If the conflict relationship value at the matrix position is 1, the row constraint information and column constraint information corresponding to the matrix position have a mutual exclusive conflict where the constraint scopes intersect. If the conflict relationship value at the matrix position is 2, the row constraint information and column constraint information corresponding to the matrix position have an overlap in their constraint scopes, indicating a conflict. The conflict behavior information of the conflict constraint pair includes: conflict type, constraint number, and conflict content.

[0010] In one embodiment of this application, optionally, the step of executing a constraint conflict handling strategy matching the conflict constraint pair based on the conflict performance information and the priority of the two constraint information items in the conflict constraint pair in the constraint information list includes: If two constraints in a conflicting constraint pair are mutually exclusive, the constraint with the higher priority in the conflicting constraint pair is retained in the constraint information list. Modify the lower priority constraint information in the conflicting constraint pair so that the lower priority constraint information does not conflict with the higher priority constraint information, or add a mutual exclusion conflict mark to the lower priority constraint information in the conflicting constraint pair. The mutual exclusion conflict mark indicates that when a mutual exclusion conflict occurs, the lower priority constraint information only takes effect when the higher priority constraint information is not applicable or when a predetermined applicable condition is met. If two constraints in the conflicting constraint pair contain conflicting information, the constraint with the smaller constraint range in the conflicting constraint pair is retained in the constraint information list. For the constraint information with a larger constraint range in the conflict constraint pair, a conflict-including marker is added. The conflict-including marker indicates that the constraint information with a larger constraint range is only used as a verification condition for the final generated insurance clause and does not participate in the generation of the insurance clause.

[0011] In one embodiment of this application, optionally, the step of executing a constraint conflict handling strategy matching the conflict constraint pair based on the conflict performance information and the priority of the two constraint information items in the conflict constraint pair in the constraint information list includes: If the priority difference between the two constraint information in the conflict constraint pair is less than a predetermined difference threshold, the comprehensive constraint corresponding to the two constraint information is constructed by the union of the constraint ranges of the two constraint information. If it is determined that both the comprehensive constraint and the two constraint information contain conflicts, return to the step of retaining the constraint information with the smaller constraint range in the conflicting constraint pair in the constraint information list.

[0012] In one embodiment of this application, optionally, the constraint supplementation processing of the updated constraint information list based on a predetermined insurance domain knowledge graph includes: For each constraint in the updated constraint information list, retrieve the set of entities related to the constraint scope and constraint object of the constraint information in the predetermined insurance domain knowledge graph; Based on the semantic relationship edges of each entity in the entity set, perform relation chain reasoning to obtain supplementary constraints that are associated with the constraint information but not displayed in the updated constraint information list; Detect whether the supplementary constraint and the constraint information are mutually exclusive or contain conflicts; If a mutual exclusion conflict exists, the supplementary constraint is determined to be invalid; If a conflict exists, and the scope of the supplementary constraint is included by the scope of the constraint information, the supplementary constraint is determined to be invalid. If a conflict exists, and the scope of the supplementary constraint includes the scope of the constraint information, the supplementary constraint is added to the constraint information list, and the conflict-included flag is added to the supplementary constraint.

[0013] Secondly, embodiments of this application provide an intelligent insurance clause generation device, comprising: The constraint information acquisition and conversion unit is used to acquire multi-source constraint information for the target insurance product and convert each constraint information in the multi-source constraint information into a quadruple, wherein the quadruple includes constraint scope, constraint predicate, constraint object and constraint weight; The constraint information list generation unit is used to determine the priority of each constraint information based on the constraint weight and data source type of each constraint information, and to arrange the quadruplets of each constraint information in descending order of the priority to obtain a constraint information list; The conflict identification unit is used to identify conflicting constraint pairs based on the constraint information list and determine the conflict performance information of the conflicting constraint pairs, wherein the conflict performance information is used to reflect the identity characteristics and conflict performance characteristics of the two constraint information in the conflicting constraint pair. The first list updating unit is used to update the constraint information list by performing a constraint conflict handling strategy that matches the conflict constraint pair based on the conflict performance information and the priority of the two constraint information in the constraint information list. The second list update unit is used to perform constraint supplementation processing on the updated constraint information list based on a predetermined insurance domain knowledge graph, so as to update the constraint information list again. The insurance clause generation unit is used to determine input information for a predetermined large language model based on the known generation guidance information and the current constraint information list of the target insurance product, and generate the insurance clauses corresponding to the input information through the large language model.

[0014] In one embodiment of this application, optionally, if the data source types of the multi-source constraint information include: regulatory constraint type, business constraint type, and insurance product constraint type, then the device further includes: The constraint information feature acquisition unit is used to acquire the regulatory weight, business impact, and product relevance of each constraint information before the constraint information list generation unit determines the priority of each constraint information; The constraint information list generation unit is used to: for each constraint information, obtain a first weight of the constraint information under the legal constraint type, obtain a second weight of the constraint information under the business constraint type, and obtain a third weight of the constraint information under the insurance product constraint type; and based on the first weight, the second weight, and the third weight, perform a weighted summation of the legal weight, business impact, and product relevance to obtain the priority of the constraint information.

[0015] In one embodiment of this application, optionally, the constraint information feature acquisition unit is configured to: for each piece of constraint information, if the constraint information is related to any regulation, determine the regulatory weight of the constraint information based on the importance level of the regulation; if the constraint information is not related to any regulation, set the regulatory weight of the constraint information to a first predetermined value; determine the business risk level of the target insurance product when the constraint information is violated, and determine the business impact level of the constraint information based on the risk level; acquire the constraint product type of the constraint information; if the constraint product type includes the target insurance product, determine the product relevance of the constraint information to a second predetermined value; otherwise, determine the product relevance of the constraint information to a third predetermined value.

[0016] In one embodiment of this application, optionally, the conflict identification unit is configured to: construct a conflict detection matrix based on the constraint information list, wherein the element in the m-th row and n-th column of the conflict detection matrix is ​​the conflict relationship value between the m-th constraint information and the n-th constraint information in the constraint information list; if the conflict relationship value at any matrix position is non-zero, determine that the row constraint information and column constraint information corresponding to the matrix position are a conflict constraint pair.

[0017] In one embodiment of this application, optionally, if the conflict relationship value at the matrix position is zero, the row constraint information and column constraint information corresponding to the matrix position do not conflict; if the conflict relationship value at the matrix position is 1, the row constraint information and column constraint information corresponding to the matrix position have a mutually exclusive conflict with overlapping constraint scopes; if the conflict relationship value at the matrix position is 2, the row constraint information and column constraint information corresponding to the matrix position have an inclusive conflict with overlapping constraint scopes; the conflict performance information of the conflict constraint pair includes: conflict type, constraint number, and conflict content.

[0018] Optionally, in one embodiment of this application, the first list update unit includes: The mutual exclusion conflict handling unit is configured to, if there is a mutual exclusion conflict between two constraint information in the conflict constraint pair, retain the constraint information with higher priority in the conflict constraint pair in the constraint information list, and modify the constraint information with lower priority in the conflict constraint pair so that the constraint information with lower priority does not conflict with the constraint information with higher priority, or add a mutual exclusion conflict mark to the constraint information with lower priority in the conflict constraint pair, wherein the mutual exclusion conflict mark indicates that when a mutual exclusion conflict occurs, the constraint information with lower priority is only effective when the constraint information with higher priority is not applicable or when a predetermined applicable condition is met; The system includes a conflict handling unit, which, if two constraint information in the conflict constraint pair contain a conflict, retains the constraint information with a smaller constraint range in the constraint information list, and adds a conflict-containing flag to the constraint information with a larger constraint range in the conflict constraint pair. The conflict-containing flag indicates that the constraint information with a larger constraint range is only used as a verification condition for the final generated insurance clause and does not participate in the generation of the insurance clause.

[0019] Optionally, in one embodiment of this application, the first list update unit further includes: The low-priority difference conflict handling unit is used to construct a comprehensive constraint corresponding to the two constraint information by using the union of the constraint ranges of the two constraint information if the priority difference between the two constraint information in the conflict constraint pair is less than a predetermined difference threshold; if it is determined that the comprehensive constraint and the two constraint information are both conflict-inclusive, it returns to the conflict-inclusive handling unit and retains the constraint information with the smaller constraint range in the conflict constraint pair in the constraint information list.

[0020] In one embodiment of this application, optionally, the second list updating unit is configured to: for each constraint information in the updated constraint information list, retrieve an entity set related to the constraint scope and constraint object of the constraint information in a predetermined insurance domain knowledge graph; perform relation chain reasoning based on the semantic relation edges of each entity in the entity set to obtain supplementary constraints associated with the constraint information that are not displayed in the updated constraint information list; detect whether the supplementary constraints and the constraint information have mutual exclusion or inclusion conflicts; if there is a mutual exclusion conflict, determine that the supplementary constraint is invalid; if there is an inclusion conflict, and the constraint scope of the supplementary constraint is included by the constraint scope of the constraint information, determine that the supplementary constraint is invalid; if there is an inclusion conflict, and the constraint scope of the supplementary constraint includes the constraint scope of the constraint information, add the supplementary constraint to the constraint information list, and add the inclusion conflict marker to the supplementary constraint.

[0021] Thirdly, embodiments of this application provide a computer device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being configured to perform the method described in the first aspect above.

[0022] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions for performing the method described in the first aspect above.

[0023] The above technical solution addresses the technical problem in related technologies where the intelligent generation of insurance clauses results in unreasonable clauses due to the inability to identify and resolve conflicts between constraints. It transforms multi-source constraint information into structured quadruples and forms a constraint information list based on quantified priority. This list is then updated twice: first, to identify and resolve conflicts between constraints within the list; and second, to use an insurance domain knowledge graph for semantic reasoning to supplement implicit constraints. Finally, it automatically generates insurance clauses using a large language model. This automates the entire process of generating insurance clauses, from rule understanding and conflict resolution to intelligent generation. It transforms the previously time-consuming and error-prone clause design process, which relied on human experience, into a highly efficient and traceable automated standard process. This significantly improves the efficiency of insurance clause generation, reduces compliance risks and labor costs, and ensures the rigor of the generated clauses under multi-dimensional rules by quantifying multi-source constraint information and identifying and resolving conflicts between them, thus increasing the reliability of the insurance clauses. Attached Figure Description

[0024] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 A flowchart of an insurance clause smart generation method according to an embodiment of this application is shown; Figure 2 A schematic diagram of an intelligent insurance clause generation and consistency verification system based on multidimensional constraint reasoning is shown. Figure 3 A block diagram of a computer device according to one embodiment of this application is shown; Figure 4 A block diagram of a computer device according to another embodiment of this application is shown. Detailed Implementation

[0026] 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.

[0027] Figure 1 A flowchart of an insurance clause smart generation method according to an embodiment of this application is shown.

[0028] like Figure 1 As shown, the process of the intelligent generation method for insurance terms according to an embodiment of this application includes: Step 102: Obtain multi-source constraint information for the target insurance product, and convert each constraint information in the multi-source constraint information into a quadruple.

[0029] Constraint information defines the content and boundaries of insurance clauses. It is a rule-based description extracted from specific data sources. Multi-source constraint information refers to constraint information from multiple data sources. The data source types for multi-source constraint information include: regulatory constraint types, business constraint types, and insurance product constraint types. In other words, constraint information can come from rules in regulations and policy documents, business requirements for the target insurance product from within the company or industry, or conditions that the target insurance product itself must meet or avoid. Therefore, through multi-dimensional constraint information, the conditions that the target insurance product needs to meet or avoid under multiple dimensions such as regulations, business, and product can be determined. Thus, multi-dimensional constraint information can serve as the basis for automatically generating insurance clauses for the target insurance product.

[0030] The four-tuple includes the constraint scope, constraint predicate, constraint object, and constraint weight. Converting constraint information into a four-tuple decomposes the overall rule-based description into four structured pieces of information: constraint scope, constraint predicate, constraint object, and constraint weight. Specifically, the constraint scope describes the scope within which the constraint information is effective; the constraint predicate describes the judgment or operation that the constraint information should perform; the constraint object describes the specific clause elements or content that the constraint information applies to; and the constraint weight describes the importance of the constraint information to the target insurance product.

[0031] Step 104: Based on the constraint weight and data source type of each constraint information, determine the priority of each constraint information, and arrange the quadruplets of each constraint information in descending order of priority to obtain a constraint information list.

[0032] The constraint weight describes the importance of the constraint information to the target insurance product and also reflects the strictness of the constraint information itself, while the data source type reflects the reliability of the constraint information's source. Based on this, constraint information with high constraint weights and strong source reliability has a greater impact on the target insurance product. Therefore, to a certain extent, constraint information with high constraint weights and strong source reliability can be given higher priority. The constraint information list is a set of four-tuples sorted by priority, reflecting all rules that must be followed during the intelligent generation of insurance clauses and the distribution of their importance. It can serve as a guide for the orderly execution of clause content generation.

[0033] Specifically, before step 102, the regulatory weight, business impact, and product relevance of each constraint can be obtained. The regulatory weight reflects the degree of legal enforceability of the constraint, the business impact reflects the potential impact on the business if the constraint is not met, and the product relevance reflects the degree of relevance between the constraint and the target insurance product, and can directly reflect whether the constraint is applicable to the target insurance product.

[0034] Based on this, for each piece of constraint information, a first weight is obtained under the regulatory constraint type, a second weight is obtained under the business constraint type, and a third weight is obtained under the insurance product constraint type. The first, second, and third weights respectively reflect the degree of influence of regulatory weight, business influence, and product relevance on the contribution of the constraint information to the target insurance product. Next, based on the first, second, and third weights, a weighted sum is calculated for the regulatory weight, business influence, and product relevance to obtain the priority of the constraint information. This priority reflects the influence of the constraint information on the target insurance property rights when it is affected by regulations, business, and product factors, under the respective influence capabilities of regulatory weight, business influence, and product relevance.

[0035] Thus, the complex multidimensional correlation attributes of constraint information are quantified into a single numerical value, providing a reliable basis for subsequent conflict identification and resolution that is easy to calculate.

[0036] Step 106: Based on the constraint information list, identify conflicting constraint pairs and determine the conflict performance information of the conflicting constraint pairs.

[0037] A conflicting constraint pair refers to two constraints in the constraint information list that have overlapping scopes and whose constraint predicates and objects are mutually exclusive or inclusive. More simply, a conflicting constraint pair is two constraints whose scopes, predicates, and objects are related. By constructing a conflict detection matrix and iterating through the non-zero elements, conflicting constraint pairs and their conflict characteristics in the constraint information list can be identified to resolve constraint conflicts.

[0038] Specifically, a conflict detection matrix can be constructed based on the constraint information list, wherein the element in the m-th row and n-th column of the conflict detection matrix is ​​the conflict relationship value between the m-th constraint information and the n-th constraint information in the constraint information list.

[0039] Based on this, if the conflict relation value at any matrix position is non-zero, the row constraint information and column constraint information corresponding to that matrix position are determined to be a conflicting constraint pair; if the conflict relation value at the matrix position is 1, the row constraint information and column constraint information corresponding to that matrix position have a mutual exclusion conflict with overlapping constraint scopes; if the conflict relation value at the matrix position is 2, the row constraint information and column constraint information corresponding to that matrix position have an inclusion conflict with overlapping constraint scopes. Thus, comprehensive and automated detection of all constraint conflict pairs is achieved through matrix traversal, significantly improving the comprehensiveness of conflict detection.

[0040] The conflict performance information is used to reflect the identity characteristics and conflict performance characteristics of the two constraint information in the conflict constraint pair, including conflict type, constraint number and conflict content. It can be said that the conflict performance information provides a reliable decision basis for subsequent steps to accurately select a conflict handling strategy that matches the conflict type and constraint object.

[0041] Step 108: Based on the conflict performance information and the priority of the two constraint information in the conflict constraint pair in the constraint information list, execute a constraint conflict handling strategy that matches the conflict constraint pair to update the constraint information list.

[0042] The conflict manifestation information directly reveals the specific type of conflict, such as mutual exclusion or inclusion, and also directly shows the content of the conflict. The priority of constraint information in the list provides a quantifiable bias standard for conflict handling, so that more critical constraint information can be satisfied first, thereby improving the reliability of the basis for generating insurance terms.

[0043] In one possible design, if two constraint information in the conflicting constraint pair are mutually exclusive, the constraint information with higher priority in the conflicting constraint pair is retained in the constraint information list, and the constraint information with lower priority in the conflicting constraint pair is modified so that the lower priority constraint information does not conflict with the higher priority constraint information, or a mutual exclusion conflict mark is added to the lower priority constraint information in the conflicting constraint pair, wherein the mutual exclusion conflict mark indicates that when a mutual exclusion conflict occurs, the lower priority constraint information only takes effect when the higher priority constraint information is not applicable or when a predetermined applicable condition is met.

[0044] In other words, if the scope of two constraints overlaps and their requirements cannot be met simultaneously, they are mutually exclusive. When resolving mutual exclusion conflicts, the constraint with higher priority can be retained, ensuring its satisfaction, while the constraint with lower priority is modified or marked as effective only when the higher priority constraint is inapplicable or when predetermined applicable conditions are met. This effectively resolves mutual exclusion conflicts, ensuring that high-priority constraints are unconditionally satisfied while flexibly preserving the applicability of lower-priority constraints under specific conditions. This maintains both the compliance baseline of the clauses and the flexibility of business rules.

[0045] In one possible design, if two constraint information in the conflicting constraint pair contain a conflict, the constraint information with a smaller constraint range in the conflicting constraint pair is retained in the constraint information list, and a conflict-containing flag is added to the constraint information with a larger constraint range in the conflicting constraint pair. The conflict-containing flag indicates that the constraint information with a larger constraint range is only used as a verification condition for the final generated insurance clause and does not participate in the generation of the insurance clause.

[0046] In other words, if the scope of two constraints overlaps, and the scope of one constraint completely encompasses the scope of the other, they are considered to be in conflict. For such conflicts, the intersection of the two constraints—the one with the smaller scope—is retained as the primary basis for generating the clause, while the constraint with the larger scope is marked as a verification condition for the final generated insurance clause. This allows for the retention of the most stringent constraint requirement in the conflicting pair while downgrading broad constraints to boundary verification conditions. This ensures the compliance of the generated clause, avoids interference from broad constraints in the generation process, and further enhances the effectiveness of the broad constraint, improving the accuracy of the generated clause.

[0047] In another possible design, if the priority difference between the two constraint information in the conflicting constraint pair is less than a predetermined difference threshold, a comprehensive constraint corresponding to the two constraint information is constructed by the union of the constraint ranges of the two constraint information; if it is determined that both the comprehensive constraint and the two constraint information contain conflict, the step of retaining the constraint information with the smaller constraint range in the conflicting constraint pair in the constraint information list is returned.

[0048] In other words, if the priorities of two constraints in a conflicting constraint pair are similar, discarding either one based solely on a difference in magnitude would negatively impact the accuracy of the generated clause. Therefore, a comprehensive constraint corresponding to the two constraints can be constructed by referring to the constraint scope of the first constraint. This comprehensive constraint integrates the original constraint information when priorities are similar, ensuring that the comprehensive constraint and both constraints contain conflict. Then, based on the conflict-containing handling method, conflict resolution is applied between the comprehensive constraint and one constraint, and between the comprehensive constraint and the other constraint. Thus, using the comprehensive constraint as a bridge, a compromise resolution path is automatically created for conflicting parties with similar priorities, downgrading the conflict between them. This avoids the rule gaps that might result from simply discarding either constraint, balances the demands of both constraints, improves the reliability of the conflict resolution result, and thus ensures the rigor of the generated clause.

[0049] Step 110: Based on the predetermined insurance domain knowledge graph, perform constraint supplementation processing on the updated constraint information list to update the constraint information list again.

[0050] The constraint supplementation process leverages the semantic reasoning capabilities of existing insurance domain knowledge graphs to automatically discover implicit constraints related to the current constraint information list, thereby improving the completeness of the constraint information list and providing a more reliable basis for the generation of subsequent clauses.

[0051] Specifically, for each constraint in the updated constraint information list, a set of entities related to the constraint scope and constraint object of the constraint information is retrieved from a predetermined insurance domain knowledge graph; relation chain reasoning is performed based on the semantic relation edges of each entity in the entity set to obtain supplementary constraints associated with the constraint information that are not displayed in the updated constraint information list.

[0052] Therefore, other constraint entities associated with the constraint information in the updated constraint information list can be automatically discovered and extracted, thereby identifying supplementary constraints. The reliability of supplementary constraints is determined by the strength of their association with relevant entities such as regulations, business, and products in the knowledge graph, as well as the confidence of the reasoning path in the knowledge graph.

[0053] Next, it is checked whether there is a mutual exclusion conflict or a conflict between the supplementary constraint and the constraint information; if there is a mutual exclusion conflict, the supplementary constraint is determined to be invalid. If there is a mutual exclusion conflict, it means that the supplementary constraint and the existing constraint are logically directly contradictory and cannot be simultaneously valid. Therefore, the supplementary constraint is invalid for the current clause generation process.

[0054] If a conflict exists and the scope of the supplementary constraint is included by the scope of the constraint information, the supplementary constraint is determined to be invalid. This indicates that the requirements proposed by the supplementary constraint have been completely covered by the existing constraints. Therefore, the supplementary constraint makes little contribution to improving the reliability of the insurance terms and is redundant content, and can be directly confirmed as invalid.

[0055] If a conflict exists, and the scope of the supplementary constraint includes the scope of the constraint information, the supplementary constraint is added to the constraint information list, and the conflict-including flag is added to the supplementary constraint. This indicates that the supplementary constraint proposes broader boundary requirements than the existing constraints. Although it is not directly used to generate specific clause content, it can serve as an important boundary verification condition to ensure that the final generated clause does not exceed the maximum allowable range defined by the supplementary constraint, thus playing a supplementary constraint role.

[0056] Step 112: Based on the known generation guidance information and the current constraint information list of the target insurance product, determine the input information for the predetermined large language model, and generate the insurance terms corresponding to the input information through the large language model.

[0057] Finally, the known generation guidance information and current constraint information list of the target insurance product are combined with prompt words and input into the large language model, which then automatically generates the insurance terms of the target insurance product.

[0058] In summary, this technical solution transforms multi-source constraint information into structured quadruples and forms a constraint information list based on quantified priority sorting. This list is then updated twice: first, to identify and resolve conflicts between constraints within the list; and second, to utilize an insurance domain knowledge graph for semantic reasoning to supplement implicit constraints. Finally, it automatically generates insurance clauses using a large language model. This automates the entire process of generating insurance clauses, from rule understanding and conflict resolution to intelligent generation. It transforms the previously time-consuming and error-prone clause design process, which relied heavily on human experience, into a highly efficient and traceable automated standard process. This significantly improves the efficiency of insurance clause generation, reduces compliance risks and labor costs, and ensures the rigor of the generated clauses under multi-dimensional rules by quantifying multi-source constraint information and identifying and resolving conflicts between them, thus increasing the reliability of the insurance clauses.

[0059] In addition, after the insurance clauses are automatically generated, a consistency check can be performed. Specifically, this check can detect whether the terminology corresponding to the same professional term in the insurance clauses is consistent; whether the format of the insurance clauses conforms to a predetermined industry standard template; whether there are logical contradictions, overlapping or omissions in the coverage, or conflicting effective conditions among the sub-clauses of the insurance clauses; whether the insurance clauses include all risk items of the target insurance product; whether the logical dependency chain between the sub-clauses of the insurance clauses is closed-loop; whether there are any missing links in the relevant business processes of the insurance clauses; and whether the insurance clauses match the predetermined legal database, business regulatory requirements database, and industry standard database. Finally, based on the detection results, problem identification and resolution are determined.

[0060] Let's take the development of a new comprehensive household property insurance product by a property insurance company as an example. First, the system extracts legal constraints from Article X and regulatory measures Y, such as the property insurance contract should clearly define the insured object and insured value. It also extracts business constraints from the company's internal product development specifications, such as earthquake liability must be listed separately as an additional insurance. Finally, it extracts product constraints from the product requirements document, such as the main insurance coverage must include fire, explosion, and typhoon. Each constraint is then converted into a quaternion.

[0061] Subsequently, the priority of each constraint is calculated. For example, regulatory constraints have the highest priority, followed by core product protection constraints, forming an ordered list of constraint information.

[0062] During the conflict identification phase, a business constraint was detected—suggesting a deductible of 500 yuan for water pipe bursting and leakage losses—and a product constraint—requiring a uniform deductible of 300 yuan for household property insurance. The two constraints overlap in scope and are mutually exclusive, constituting a conflict. Therefore, if the core product constraint has higher priority, the 300 yuan deductible requirement can be retained, and the 500 yuan business constraint can be marked to indicate that it only applies to specific promotional programs.

[0063] Next, the insurance knowledge graph was used for reasoning, and it was found that home property insurance is associated with the implicit regulatory requirement that the insured object must have a clearly defined location address. This requirement was then added to the list as a supplementary constraint and used as a verification condition for the final terms.

[0064] Finally, all the processed and supplemented constraint information, along with the product description of the comprehensive family property insurance, is input into the big data model to automatically generate logically rigorous, compliant, and business-designed insurance terms.

[0065] Figure 2 A schematic diagram of an intelligent insurance clause generation and consistency verification system based on multidimensional constraint reasoning is shown.

[0066] like Figure 2As shown, following its hierarchical order from input to output, the first layer is the input layer, where the system receives various input information, including: business requirement specifications, regulatory compliance requirements, product parameter configurations, historical clause database, and expert knowledge base.

[0067] The second layer is the Intelligent Constraint Modeling Layer. Based on the input information, the system performs constraint modeling. The Intelligent Constraint Modeling Layer includes: a regulatory constraint extractor, a business rule modeler, a product constraint generator, a constraint conflict detector, and a constraint priority analyzer.

[0068] The third layer is the Semantic Constraint Reasoning Engine. Based on constraint modeling, the system performs semantic reasoning and verification. The Semantic Constraint Reasoning Engine includes: constraint logic inferrer, semantic consistency verifier, clause verification analyzer, and compliance reasoning engine.

[0069] The fourth layer is the Intelligent Clause Generation Core. The system generates and optimizes clauses based on the reasoning results. This layer includes: an insurance domain fine-tuning model, a clause template matcher, a constraint-aware generator, and a multi-round iterative optimizer.

[0070] The fifth layer is the ConsistencyValidation Layer, which performs multi-level consistency checks after the terms are generated. This layer includes: a syntax consistency checker, a semantic conflict detector, a logical integrity validator, a compliance assessment evaluator, and a terms relationship graph builder.

[0071] The sixth layer is the Intelligent Feedback Optimization Layer, where the system analyzes problems and generates optimization suggestions based on the verification results. This layer includes: a problem location analyzer, a repair suggestion generator, a quality scoring calculator, and a manual review interface.

[0072] The final layer is the output layer, which outputs the following: generated insurance terms, consistency verification report, compliance analysis report, optimization suggestion list, and quality assessment report.

[0073] The entire process starts with multi-source input, goes through steps such as constraint modeling, semantic reasoning, clause generation, consistency verification, and feedback optimization, and finally outputs complete insurance clauses and related analysis reports.

[0074] The following is combined Figure 2The system architecture shown in the diagram illustrates the technical solution of this application in detail.

[0075] This technical solution adopts a six-step progressive process: constraint acquisition, conflict detection, intelligent reasoning, clause generation, consistency verification, and feedback optimization. The output of each step serves as the input for the next step, forming a complete technical processing chain. The specific implementation process of each step is described in detail below.

[0076] Step 1: Obtaining and structurally modeling multi-level constraints.

[0077] Step Objective: To obtain various constraints related to insurance terms from multiple sources and transform them into a unified structured representation, laying the foundation for subsequent conflict detection and reasoning analysis.

[0078] Information obtained: This step obtains constraint information from three levels.

[0079] Regulatory constraints layer: Extract regulatory constraints from insurance laws and regulations, policy documents issued by regulatory agencies, and industry standards and normative documents.

[0080] Business constraint layer: Extract business constraints from the company's internal product design specifications, risk control guidelines, and business process manuals, such as the waiting period for medical insurance not being less than 30 days.

[0081] Product constraint layer: Extract product characteristic constraints from the specific product's requirements specification, actuarial report, and product instruction manual template. For example, the coverage period of this product is 1 year and it is not renewable.

[0082] Information processing procedure: For each constraint obtained, the system uses a "quadruple" data structure for structured modeling. Specifically, each constraint rule is represented as: C_i=(S_i,P_i,O_i,W_i), and the meanings of these four elements are as follows.

[0083] S_i (Constraint Scope): Describes the scope within which this constraint is effective. The scope defines the boundaries of the constraint's application, such as whether it applies to all health insurance products or only to group insurance. By specifying the scope, the system can determine whether a constraint applies to the currently generated insurance product terms.

[0084] P_i (Constraint Predicate): Describes what judgment or operation the constraint requires to be performed. The predicate is the core logical part of the constraint, defining the specific rules, such as must be included, must not exceed, and should be explicitly stated. The predicate determines whether the constraint is a mandatory requirement, a prohibition, or a suggestion.

[0085] O_i (Constraint Object): Describes which specific clause elements or content the constraint applies to. The constraint object specifies the concrete target being constrained, such as coverage clauses, exclusion clauses, waiting periods, and sum insured. By clearly defining the constraint object, the system can precisely apply the constraint to the corresponding parts of the clause.

[0086] W_i (Constraint Weight): A numerical value describing the importance and priority of this constraint. The weight value reflects the priority order of handling constraints when conflicts occur, and typically ranges from 0 to 1. Regulatory mandatory constraints usually have a weight close to 1, while general business recommendations have a relatively lower weight.

[0087] Processing result: After this step, the system obtains a structured set of constraints, each with a uniform data format, which facilitates automated computer processing in subsequent steps.

[0088] Step 2: Calculation and sorting of constraint priorities.

[0089] Step objective: Calculate the priority of all constraints obtained in the first step to determine the processing order of each constraint when a conflict occurs.

[0090] Information obtained: This step uses the structured constraint set output in the first step, combined with the following three types of evaluation information: the legal enforceability of the constraint, indicating whether it originates from laws and regulations, whether there are clear penalties, etc.; the impact of the constraint on the business, indicating the assessment of business losses that may be caused by violating the constraint; and the relevance of the constraint to the current product, i.e., the degree of matching between the constraint and the current product type.

[0091] Information processing procedure: The system uses a weighted summation formula to calculate the comprehensive priority score of each constraint, Priority(C_i)=α×Legal_Weight(C_i)+β×Business_Impact(C_i)+γ×Product_Relevance(C_i), where the specific meanings of the three parameters are as follows.

[0092] Legal_Weight: Measures the legal enforceability of constraints. This parameter assesses whether the constraint originates from laws, regulations, regulatory requirements, or mandatory industry standards. Constraints explicitly stipulated by law have a Legal_Weight value close to 1 (e.g., insurance clauses must be filed with regulatory authorities); while internal guidelines have a lower Legal_Weight value. This parameter ensures that legal constraints are prioritized, avoiding the generation of clauses that violate regulations.

[0093] Business_Impact: Measures the potential impact of violating this constraint on the business. This parameter assesses the business risks and losses that might result from ignoring the constraint. For example, constraints involving core safeguard responsibilities have a higher Business_Impact value, while constraints that only affect the wording of the terms have a lower Business_Impact value. This parameter ensures that constraints with a significant impact on the business are prioritized for protection.

[0094] Product_Relevance: Measures the relevance of constraints to the currently generated product. This parameter assesses whether a constraint is directly applicable to the current product type. For example, for a critical illness insurance product being generated, the Product_Relevance value for the constraint that the waiting period for critical illness insurance does not exceed 180 days is very high; while the Product_Relevance value for the constraint that a car insurance claim must be filed within 24 hours of the incident is close to 0. This parameter ensures that the system prioritizes constraints that are highly relevant to the current product.

[0095] α, β, and γ are three weighting coefficients used to adjust the relative importance of the three parameters, and they satisfy α + β + γ = 1. In practical applications, α is usually set to a relatively high value (e.g., 0.5) to ensure that regulatory compliance takes priority; β and γ are adjusted according to specific business needs.

[0096] Processing result: After this step, the system obtains a list of constraints sorted by priority, with higher priority constraints listed first. This sorting result will be used to determine the conflict resolution strategy in the conflict detection step in the third step.

[0097] Step 3: Detection and identification of constraint conflicts.

[0098] Step objective: To perform pairwise comparisons on the constraint sets sorted in the second step, detect and identify conflicting constraint pairs, and provide a basis for subsequent conflict resolution.

[0099] Information obtained: This step uses the priority-sorted constraint set output from the second step to perform pairwise analysis on the constraints.

[0100] Information processing procedure: The system uses a conflict detection matrix method to systematically detect conflict relationships between constraints. The specific detection process is as follows: Phase 1: Constructing the conflict detection matrix.

[0101] The system constructs an n×n matrix (where n is the total number of constraints), where each position i stores the conflict relationship value between constraints C_i and C_j. The matrix construction process is as follows: the system sequentially extracts each pair of constraints (C_i, C_j) and analyzes whether there is a conflict between them.

[0102] Phase Two: Determine the type of conflict.

[0103] For each pair of constraints (C_i, C_j), the system determines the conflict type and assigns a value according to the following rules.

[0104] No conflict (matrix value = 0): Two constraints are considered conflict-free when their scopes are completely disjoint. For example, constraint C_i has a scope of life insurance products, and constraint C_j has a scope of property insurance products. They will not act on the same product simultaneously, therefore there is no conflict. The condition is mathematically represented as: C_i ∩ C_j = (The intersection of the scopes of the two constraints is an empty set).

[0105] Mutually exclusive conflict (matrix value = 1): When the scopes of two constraints overlap, and their requirements contradict each other and cannot be satisfied simultaneously, they are considered mutually exclusive. For example, constraint C_i requires a waiting period of 30 days, and constraint C_j requires a waiting period of 90 days. The specific values ​​required are different, and they cannot be satisfied simultaneously. The condition for determination is: C_i ∩ C_j ≠ ∧C_i⊕C_j means that the scopes have an intersection and the two constraints are mutually exclusive.

[0106] Inclusion Conflict (Matrix Value = 2): When the scopes of two constraints overlap, and the scope of one constraint completely encompasses the scope of the other, it is considered an inclusion conflict. For example, constraint C_i requires a guarantee period of no more than 20 years, and constraint C_j requires a guarantee period of no more than 10 years; the latter's scope is completely encompassed by the former. The determination condition is: C_i ∩ C_j ≠ ∧C_i C_j means that the scopes have an intersection, and the scope of C_i includes C_j.

[0107] Phase 3: Record conflict information.

[0108] For each pair of conflicting constraints detected, the system records the type of conflict, the constraint number involved, and a detailed description of the conflict, forming a conflict list.

[0109] Processing result: After this step, the system obtains a complete conflict detection matrix and a detailed conflict list, which lists all conflicting constraint pairs and their conflict types.

[0110] Step 4: Resolving constraint conflicts and optimizing constraint sets.

[0111] Step objective: For the conflicting constraint pairs detected in step 3, process them according to the conflict type and constraint priority, and use the corresponding resolution strategies to generate a conflict-free set of optimization constraints.

[0112] Information obtained: This step uses the conflict list and conflict detection matrix output in step 3, combined with the constraint priority information calculated in step 2.

[0113] Information processing: The system adopts different resolution strategies for different types of conflicts.

[0114] Strategy 1: Priority resolution strategy, applicable to mutual exclusion conflicts.

[0115] When two constraints conflict with each other, the system compares their priority scores. The constraint with higher priority is retained, while the constraint with lower priority is adjusted or marked as conditionally applicable.

[0116] Specific execution process: If Priority(C_i) > Priority(C_j), then the full requirement of C_i is retained, and C_j is marked as a supplementary requirement under the condition of C_i or applicable only in specific circumstances. For example, if regulatory constraints require a waiting period of no more than 180 days (priority 0.9), and business constraints suggest a waiting period of 90 days (priority 0.6), the system retains the regulatory constraint as a mandatory upper limit and uses the business constraint as a suggested value.

[0117] Strategy Two: Negotiation-based resolution strategy, applicable when both parties have similar priorities.

[0118] When the priority scores of two conflicting constraints are close, the system uses constraint relaxation methods to find a compromise that can simultaneously satisfy the core requirements of both constraints.

[0119] The specific execution process is as follows: The system analyzes the common feasible interval of the two constraints and generates a new comprehensive constraint. This comprehensive constraint satisfies both the key requirements of constraint C_i and the main demands of constraint C_j. For example, if constraint C_i requires a compensation ratio of no less than 80%, and constraint C_j requires a compensation ratio of no more than 90%, the system generates a comprehensive constraint with a compensation ratio of 80%-90%.

[0120] Strategy 3: Layered resolution strategy, applicable to situations involving conflict.

[0121] When two constraints have an inclusion relationship, the system determines the retention strategy based on the hierarchical origin of the constraints. Generally, the constraint with the stricter scope is retained as the actual execution standard, while the constraint with the broader scope is retained as a boundary condition.

[0122] Specific execution process: If C_i includes C_j, and the scope of C_i is wider, the system will use C_j as the actual execution constraint, while C_i will be retained as an external boundary constraint. For example, if regulations require the deductible to not exceed 10,000 yuan (C_i), and product constraints require the deductible to be 5,000 yuan (C_j, which is included by C_i), the system will use the deductible of 5,000 yuan as the actual constraint, while retaining the regulatory boundary of not exceeding 10,000 yuan.

[0123] Processing result: After this step, the system obtains a set of optimized constraints that have been resolved. The constraints in this set no longer conflict with each other and can be directly used to guide clause generation.

[0124] Step 5: Semantic reasoning and clause generation based on knowledge graphs.

[0125] Step Objective: Utilize a pre-built insurance domain knowledge graph to perform semantic reasoning on the optimized constraint set obtained in step four, supplement implicit constraint relationships, and guide the large language model to generate insurance clause content that meets the requirements based on the reasoning results.

[0126] Regarding the knowledge graph: The insurance knowledge graph used in this solution is pre-built based on the full range of knowledge in the insurance industry, not just based on the current conflict query results. The construction of the knowledge graph is an independent preliminary project, and its data sources include: a full-text library of insurance industry laws and regulations; a historical clause library of various insurance products; a dictionary of insurance terminology and definition of concepts; insurance business operation standards and industry practices; and product review guidelines and penalty cases issued by regulatory agencies.

[0127] By extracting, organizing, and associating knowledge from the above sources, a knowledge graph structure containing entity, relation, and time dimensions is formed: E (entity set): including insurance product entities, clause entities, legal provisions entities, conceptual term entities, etc.; R (relationship set): including semantic relations such as "containment", "dependency", "conflict", "mutual exclusion", "applicable"; T (time dimension): recording the effective time of regulations, product iteration versions, and other temporal information to ensure that the latest and most valid knowledge is used.

[0128] Explanation of the purpose of the reasoning process: The semantic reasoning in this step is not for conflict detection, which was completed in step three, but is for the following purpose.

[0129] Supplementing implicit constraints: By deriving relationships from the knowledge graph, implicit constraints that users have not explicitly provided but are logically required to comply with are discovered. For example, if a user specifies the type of critical illness insurance product, the system can use the knowledge graph to deduce a series of regulatory requirements that this product type must comply with.

[0130] Verify constraint completeness: Check whether the current constraint set covers all the necessary constraints that this type of product should have, and identify any possible missing constraints.

[0131] Obtain generation guidance information: Retrieve clause templates, standard expressions, professional terms, etc. related to the current constraints from the knowledge graph to provide professional reference for clause generation of the large language model.

[0132] Information processing process: The reasoning process is carried out according to the following steps.

[0133] Constraint coverage calculation: For each constraint in the optimization constraint set, calculate all relevant entities involved in its scope of application in the knowledge graph.

[0134] Related entity retrieval: Based on the scope of the constraints, search for all related entity nodes in the knowledge graph, such as products, terms, regulations, concepts, etc.

[0135] Relationship chain reasoning: Reason along the relationship edges in the knowledge graph to discover other constraints, clause templates, standard expressions, etc. that are related to the current constraint.

[0136] Consistency check of reasoning results: Verify the consistency of supplementary information obtained from reasoning and filter out content that contradicts existing constraints.

[0137] Generate guidance information summary: Compile the clause templates, professional terms, standard expressions, etc. obtained through reasoning into generation guidance information.

[0138] Clause generation process: Based on the complete constraint set obtained through reasoning and generation guidance information, the system calls a large language model fine-tuned for the insurance domain to generate the clause content. The generation process adopts a mechanism of constraint injection, template fusion, and iterative optimization: constraint rules are encoded as special control signals and injected into the model; the historical clause template with the highest matching degree is used as the starting point for generation; and the generation quality is gradually improved through multiple rounds of generation-verification-optimization loops.

[0139] Processing result: After this step, the system obtains a preliminary version of the insurance terms that meets all the constraints.

[0140] Step 6: Consistency verification and intelligent feedback optimization.

[0141] Step Objective: To perform multi-level consistency checks on the initial draft of the terms generated in step 5, identify existing problems and generate repair suggestions, and obtain the final high-quality terms through iterative optimization.

[0142] Information obtained: This step uses the initial draft of the terms output in step 5 and the optimized constraint set in step 4 as the verification criteria.

[0143] Information processing process: Phase 1: Multi-level Consistency Check. The system performs consistency checks on the generated clauses from four levels.

[0144] Syntactic consistency check: Check whether the terminology for the same concept is used consistently in different clauses, whether the clause format conforms to the standard template, and whether the clause structure is complete and without omissions.

[0145] Semantic conflict detection: Check for logical contradictions, overlapping or omissions in the scope of protection, and conflicting conditions for effectiveness among the clauses.

[0146] Logical integrity verification: Check whether the risk coverage is complete, whether the logical chain of the terms is closed, and whether there are any missing links in the business process.

[0147] Compliance assessment: This examines the degree to which the terms of the inspection comply with current regulations, regulatory requirements, and industry standards.

[0148] Phase Two: Problem Identification and Remediation Recommendation Generation. For problems discovered during the inspection, the system performs the following steps: categorizes the problems, such as constraint violations, semantic conflicts, logical incompleteness, and compliance issues; locates the specific position of the problem within the relevant clauses; analyzes the root cause of the problem; and matches the appropriate remediation strategy based on the problem type, generating specific remediation recommendations.

[0149] Phase 3: Iterative optimization.

[0150] The system provides feedback on the repair suggestions to the clause generation module, which then makes targeted modifications to the problematic clauses and performs a consistency check again. This check-repair-re-check process is repeated until the generated clauses pass all checks or the preset maximum number of iterations is reached.

[0151] Processing result: After this step, the system outputs the final high-quality insurance terms, which meet all constraints, pass the consistency check, and can be used for product launch.

[0152] This application also provides an intelligent insurance clause generation device, including: The constraint information acquisition and conversion unit is used to acquire multi-source constraint information for the target insurance product and convert each constraint information in the multi-source constraint information into a quadruple, wherein the quadruple includes constraint scope, constraint predicate, constraint object and constraint weight; The constraint information list generation unit is used to determine the priority of each constraint information based on the constraint weight and data source type of each constraint information, and to arrange the quadruplets of each constraint information in descending order of the priority to obtain a constraint information list; The conflict identification unit is used to identify conflicting constraint pairs based on the constraint information list and determine the conflict performance information of the conflicting constraint pairs, wherein the conflict performance information is used to reflect the identity characteristics and conflict performance characteristics of the two constraint information in the conflicting constraint pair. The first list updating unit is used to update the constraint information list by performing a constraint conflict handling strategy that matches the conflict constraint pair based on the conflict performance information and the priority of the two constraint information in the constraint information list. The second list update unit is used to perform constraint supplementation processing on the updated constraint information list based on a predetermined insurance domain knowledge graph, so as to update the constraint information list again. The insurance clause generation unit is used to determine input information for a predetermined large language model based on the known generation guidance information and the current constraint information list of the target insurance product, and generate the insurance clauses corresponding to the input information through the large language model.

[0153] In one embodiment of this application, optionally, if the data source types of the multi-source constraint information include: regulatory constraint type, business constraint type, and insurance product constraint type, then the device further includes: The constraint information feature acquisition unit is used to acquire the regulatory weight, business impact, and product relevance of each constraint information before the constraint information list generation unit determines the priority of each constraint information; The constraint information list generation unit is used to: for each constraint information, obtain a first weight of the constraint information under the legal constraint type, obtain a second weight of the constraint information under the business constraint type, and obtain a third weight of the constraint information under the insurance product constraint type; and based on the first weight, the second weight, and the third weight, perform a weighted summation of the legal weight, business impact, and product relevance to obtain the priority of the constraint information.

[0154] In one embodiment of this application, optionally, the constraint information feature acquisition unit is configured to: for each piece of constraint information, if the constraint information is related to any regulation, determine the regulatory weight of the constraint information based on the importance level of the regulation; if the constraint information is not related to any regulation, set the regulatory weight of the constraint information to a first predetermined value; determine the business risk level of the target insurance product when the constraint information is violated, and determine the business impact level of the constraint information based on the risk level; acquire the constraint product type of the constraint information; if the constraint product type includes the target insurance product, determine the product relevance of the constraint information to a second predetermined value; otherwise, determine the product relevance of the constraint information to a third predetermined value.

[0155] In one embodiment of this application, optionally, the conflict identification unit is configured to: construct a conflict detection matrix based on the constraint information list, wherein the element in the m-th row and n-th column of the conflict detection matrix is ​​the conflict relationship value between the m-th constraint information and the n-th constraint information in the constraint information list; if the conflict relationship value at any matrix position is non-zero, determine that the row constraint information and column constraint information corresponding to the matrix position are a conflict constraint pair.

[0156] In one embodiment of this application, optionally, if the conflict relationship value at the matrix position is zero, the row constraint information and column constraint information corresponding to the matrix position do not conflict; if the conflict relationship value at the matrix position is 1, the row constraint information and column constraint information corresponding to the matrix position have a mutually exclusive conflict with overlapping constraint scopes; if the conflict relationship value at the matrix position is 2, the row constraint information and column constraint information corresponding to the matrix position have an inclusive conflict with overlapping constraint scopes; the conflict performance information of the conflict constraint pair includes: conflict type, constraint number, and conflict content.

[0157] Optionally, in one embodiment of this application, the first list update unit includes: The mutual exclusion conflict handling unit is configured to, if there is a mutual exclusion conflict between two constraint information in the conflict constraint pair, retain the constraint information with higher priority in the conflict constraint pair in the constraint information list, and modify the constraint information with lower priority in the conflict constraint pair so that the constraint information with lower priority does not conflict with the constraint information with higher priority, or add a mutual exclusion conflict mark to the constraint information with lower priority in the conflict constraint pair, wherein the mutual exclusion conflict mark indicates that when a mutual exclusion conflict occurs, the constraint information with lower priority is only effective when the constraint information with higher priority is not applicable or when a predetermined applicable condition is met; The system includes a conflict handling unit, which, if two constraint information in the conflict constraint pair contain a conflict, retains the constraint information with a smaller constraint range in the constraint information list, and adds a conflict-containing flag to the constraint information with a larger constraint range in the conflict constraint pair. The conflict-containing flag indicates that the constraint information with a larger constraint range is only used as a verification condition for the final generated insurance clause and does not participate in the generation of the insurance clause.

[0158] Optionally, in one embodiment of this application, the first list update unit further includes: The low-priority difference conflict handling unit is used to construct a comprehensive constraint corresponding to the two constraint information by using the union of the constraint ranges of the two constraint information if the priority difference between the two constraint information in the conflict constraint pair is less than a predetermined difference threshold; if it is determined that the comprehensive constraint and the two constraint information are both conflict-inclusive, it returns to the conflict-inclusive handling unit and retains the constraint information with the smaller constraint range in the conflict constraint pair in the constraint information list.

[0159] In one embodiment of this application, optionally, the second list updating unit is configured to: for each constraint information in the updated constraint information list, retrieve an entity set related to the constraint scope and constraint object of the constraint information in a predetermined insurance domain knowledge graph; perform relation chain reasoning based on the semantic relation edges of each entity in the entity set to obtain supplementary constraints associated with the constraint information that are not displayed in the updated constraint information list; detect whether the supplementary constraints and the constraint information have mutual exclusion or inclusion conflicts; if there is a mutual exclusion conflict, determine that the supplementary constraint is invalid; if there is an inclusion conflict, and the constraint scope of the supplementary constraint is included by the constraint scope of the constraint information, determine that the supplementary constraint is invalid; if there is an inclusion conflict, and the constraint scope of the supplementary constraint includes the constraint scope of the constraint information, add the supplementary constraint to the constraint information list, and add the inclusion conflict marker to the supplementary constraint.

[0160] The device uses the solution described in any one of the above embodiments, and therefore has all the above-mentioned technical effects, which will not be repeated here.

[0161] In another embodiment, this application provides a computer device, which may be a server, and its internal structure diagram may be as follows. Figure 3 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 can implement the methods described in any of the above embodiments.

[0162] In one embodiment, this application also provides a computer device, which can be a client, and its internal structure diagram can be as follows: Figure 4As shown, the computer device includes a processor, memory, network interface, display screen, and input device connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. 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 can implement the methods described in any of the above embodiments.

[0163] Any of the computer devices described in the embodiments of this application exist in various forms, including but not limited to: (1) Mobile communication devices: These devices are characterized by their mobile communication capabilities and primarily aim to provide voice and data communication. These terminals include smartphones, multimedia phones, feature phones, and low-end phones.

[0164] (2) Ultra-mobile personal computer devices: These devices fall under the category of personal computers, possessing computing and processing capabilities, and generally also have mobile internet access features. These terminals include PDAs, MIDs, and UMPCs, etc.

[0165] (3) Portable entertainment devices: These devices can display and play multimedia content. This category includes: audio and video players, handheld game consoles, e-books, as well as smart toys, wearable devices, and portable car navigation devices.

[0166] (4) Server: A device that provides computing services. The components of a server include a processor, hard disk, memory, system bus, etc. Servers are similar to general computer architectures, but because they need to provide highly reliable services, they have higher requirements in terms of processing power, stability, reliability, security, scalability, and manageability.

[0167] (5) Other electronic devices with data interaction functions.

[0168] Additionally, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which are used to perform the following steps: Obtain multi-source constraint information for the target insurance product, and convert each constraint information in the multi-source constraint information into a quadruple, wherein the quadruple includes constraint scope, constraint predicate, constraint object and constraint weight; Based on the constraint weight and data source type of each constraint information, the priority of each constraint information is determined, and the quadruplets of each constraint information are arranged in descending order of priority to obtain a constraint information list; Based on the constraint information list, conflicting constraint pairs are identified, and conflict performance information of the conflicting constraint pairs is determined, wherein the conflict performance information is used to reflect the identity characteristics and conflict performance characteristics of the two constraint information in the conflicting constraint pair. Based on the conflict performance information and the priority of the two constraints in the conflict constraint pair in the constraint information list, a constraint conflict handling strategy matching the conflict constraint pair is executed on the conflict constraint pair to update the constraint information list. Based on a pre-defined knowledge graph in the insurance field, the updated constraint information list is supplemented with constraints to update the constraint information list again. Based on the known generation guidance information and the current list of constraint information for the target insurance product, input information is determined for a predetermined large language model, and the insurance terms corresponding to the input information are generated through the large language model.

[0169] 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 in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0170] The technical solution of this application has been described in detail above with reference to the accompanying drawings. This technical solution converts multi-source constraint information into structured quadruples, and forms a constraint information list based on quantified priority sorting. The constraint information list is then updated twice: first, to identify and resolve conflicts between constraints within the list; and second, to use an insurance domain knowledge graph for semantic reasoning to supplement implicit constraints. Finally, a large language model is used to automatically generate insurance clauses. This achieves an automated process for generating insurance clauses, from rule understanding and conflict resolution to intelligent generation. It transforms the original clause design process, which relied on human experience, was time-consuming, and prone to errors, into an efficient and traceable automated standard process. This significantly improves the efficiency of insurance clause generation, reduces compliance risks and labor costs, and ensures the rigor of the generated clauses under multi-dimensional rules by quantifying multi-source constraint information and identifying and resolving conflicts between them, thus increasing the reliability of the insurance clauses.

[0171] Depending on the context, the word "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrase "if determination" or "if detection (of the stated condition or event)" can be interpreted as "when determination," "in response to determination," "when detection (of the stated condition or event)," or "in response to detection (of the stated condition or event)."

[0172] The terminology used in the embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. The singular forms “a,” “the,” and “the” used in the embodiments of this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise.

[0173] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0174] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in a combination of hardware and software functional units.

[0175] 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, and when executed, it 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 can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0176] The above-described 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 method for intelligently generating insurance clauses, characterized in that, include: Obtain multi-source constraint information for the target insurance product, and convert each constraint information in the multi-source constraint information into a quadruple, wherein the quadruple includes constraint scope, constraint predicate, constraint object and constraint weight; Based on the constraint weight and data source type of each constraint information, the priority of each constraint information is determined, and the quadruplets of each constraint information are arranged in descending order of priority to obtain a constraint information list; Based on the constraint information list, conflicting constraint pairs are identified, and conflict performance information of the conflicting constraint pairs is determined, wherein the conflict performance information is used to reflect the identity characteristics and conflict performance characteristics of the two constraint information in the conflicting constraint pair. Based on the conflict performance information and the priority of the two constraints in the conflict constraint pair in the constraint information list, a constraint conflict handling strategy matching the conflict constraint pair is executed on the conflict constraint pair to update the constraint information list. Based on a pre-defined knowledge graph in the insurance field, the updated constraint information list is supplemented with constraints to update the constraint information list again. Based on the known generation guidance information and the current list of constraint information for the target insurance product, input information is determined for a predetermined large language model, and the insurance terms corresponding to the input information are generated through the large language model.

2. The method according to claim 1, characterized in that, The data source types for the multi-source constraint information include: regulatory constraint type, business constraint type, and insurance product constraint type. Therefore, before determining the priority of each constraint information based on its constraint weight and data source type, the method further includes: Obtain the regulatory weight, business impact, and product relevance of each constraint; The process of determining the priority of each constraint based on its constraint weight and data source type includes: For each piece of constraint information, obtain the first weight of the constraint information under the legal constraint type, obtain the second weight of the constraint information under the business constraint type, and obtain the third weight of the constraint information under the insurance product constraint type; Based on the first weight, the second weight, and the third weight, the regulatory weight, business impact, and product relevance are weighted and summed to obtain the priority of the constraint information.

3. The method according to claim 2, characterized in that, The acquisition of the regulatory weight, business impact, and product relevance of each constraint information item includes: For each of the aforementioned constraint information If the constraint information is related to any regulation, the regulatory weight of the constraint information is determined based on the importance level of the regulation; if the constraint information is not related to any regulation, the regulatory weight of the constraint information is set to a first predetermined value. Determine the business risk level of the target insurance product when the constraint information is violated, and determine the business impact level of the constraint information based on the risk level; Obtain the constraint product type of the constraint information. If the constraint product type includes the target insurance product, determine the product relevance of the constraint information as a second predetermined value; otherwise, determine the product relevance of the constraint information as a third predetermined value.

4. The method according to any one of claims 1 to 3, characterized in that, The process of identifying conflicting constraint pairs based on the constraint information list includes: Based on the constraint information list, a conflict detection matrix is ​​constructed, wherein the element in the m-th row and n-th column of the conflict detection matrix is ​​the conflict relationship value between the m-th constraint information and the n-th constraint information in the constraint information list; If the conflict relation value at any matrix position is non-zero, the row constraint information and column constraint information corresponding to the matrix position are determined to be a conflict constraint pair.

5. The method according to claim 4, characterized in that, If the conflict relationship value at the matrix position is zero, the row constraint information and column constraint information corresponding to the matrix position do not conflict. If the conflict relationship value at the matrix position is 1, the row constraint information and column constraint information corresponding to the matrix position have a mutual exclusive conflict where the constraint scopes intersect. If the conflict relationship value at the matrix position is 2, the row constraint information and column constraint information corresponding to the matrix position have an overlap in their constraint scopes, indicating a conflict. The conflict behavior information of the conflict constraint pair includes: conflict type, constraint number, and conflict content.

6. The method according to claim 5, characterized in that, The step of implementing a constraint conflict handling strategy matching the conflict constraint pair based on the conflict performance information and the priority of the two constraint information items in the constraint information list includes: If two constraints in a conflicting constraint pair are mutually exclusive, the constraint with the higher priority in the conflicting constraint pair is retained in the constraint information list. Modify the lower priority constraint information in the conflicting constraint pair so that the lower priority constraint information does not conflict with the higher priority constraint information, or add a mutual exclusion conflict mark to the lower priority constraint information in the conflicting constraint pair. The mutual exclusion conflict mark indicates that when a mutual exclusion conflict occurs, the lower priority constraint information only takes effect when the higher priority constraint information is not applicable or when a predetermined applicable condition is met. If two constraints in the conflicting constraint pair contain conflicting information, the constraint with the smaller constraint range in the conflicting constraint pair is retained in the constraint information list. For the constraint information with a larger constraint range in the conflict constraint pair, a conflict-including marker is added. The conflict-including marker indicates that the constraint information with a larger constraint range is only used as a verification condition for the final generated insurance clause and does not participate in the generation of the insurance clause.

7. The method according to claim 6, characterized in that, The step of implementing a constraint conflict handling strategy matching the conflict constraint pair based on the conflict performance information and the priority of the two constraint information items in the constraint information list includes: If the priority difference between the two constraint information in the conflict constraint pair is less than a predetermined difference threshold, the comprehensive constraint corresponding to the two constraint information is constructed by the union of the constraint ranges of the two constraint information. If it is determined that both the comprehensive constraint and the two constraint information contain conflicts, return to the step of retaining the constraint information with the smaller constraint range in the conflicting constraint pair in the constraint information list.

8. The method according to claim 6, characterized in that, The process of supplementing the updated constraint information list based on a predetermined insurance domain knowledge graph includes: For each constraint in the updated constraint information list, retrieve the set of entities related to the constraint scope and constraint object of the constraint information in the predetermined insurance domain knowledge graph; Based on the semantic relationship edges of each entity in the entity set, perform relation chain reasoning to obtain supplementary constraints that are associated with the constraint information but not displayed in the updated constraint information list; Detect whether the supplementary constraint and the constraint information are mutually exclusive or contain conflicts; If a mutual exclusion conflict exists, the supplementary constraint is determined to be invalid; If a conflict exists, and the scope of the supplementary constraint is included by the scope of the constraint information, the supplementary constraint is determined to be invalid. If a conflict exists, and the scope of the supplementary constraint includes the scope of the constraint information, the supplementary constraint is added to the constraint information list, and the conflict-included flag is added to the supplementary constraint.

9. A computer device, characterized in that, include: At least one processor; And, a memory communicatively connected to the at least one processor; The memory stores instructions executable by the at least one processor, the instructions being configured to cause the processor to perform the method according to any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The device stores computer-executable instructions configured to perform the method as described in any one of claims 1 to 8.

Images