Policy risk identification methods, devices, computer equipment and storage media

Abstract

This application belongs to the fields of artificial intelligence and fintech, and relates to a method, device, computer equipment, and storage medium for identifying insurance policy risks. The method includes: acquiring the special agreement text and the insurance policy text to be identified; extracting a set of special liabilities from the special agreement text, and extracting a set of liability clauses and a set of exclusion clauses from the insurance policy text; calculating the difference between the set of special liabilities and the set of liability clauses, and the intersection between the set of special liabilities and the set of exclusion clauses, using a semantic understanding model; determining the extended liability scope based on the difference, and determining the exclusion liability scope based on the intersection; determining the risk level of the special liability clauses included in each liability type based on the liability types of the extended liability scope and the exclusion liability scope, and determining the warning content of the special liability clauses. Furthermore, this application also relates to blockchain technology, allowing the special agreement text and insurance policy text, among other data, to be stored on the blockchain. This application enables efficient risk identification for special clauses.

Description

Technical Field

[0001] This application relates to the fields of artificial intelligence and financial technology, and in particular to a method, device, computer equipment and storage medium for identifying insurance policy risks. Background Technology

[0002] In the course of insurance business, it is often necessary to use special clauses (i.e., special liability clauses) to fulfill additional agreements with clients. These special clauses include fixed special clauses, semi-fixed special clauses, and freely editable special clauses, which have legal effect beyond the existing clauses. Among them, freely editable special clauses, due to their high degree of flexibility and strong legal effect, often include coverage content inconsistent with the existing clauses. While this extension of liability is flexible, it also brings the risk of claims leakage. Currently, the risk assessment of freely editable special clauses mainly relies on the case-by-case evaluation of professional underwriters, a time-consuming, labor-intensive, and inefficient process. Summary of the Invention

[0003] The purpose of this application is to provide a method, device, computer equipment, and storage medium for identifying policy risks, which can efficiently identify risks related to specific clauses.

[0004] To address the aforementioned technical problems, this application provides a method for identifying policy risks, employing the following technical solution:

[0005] Obtain the special agreement text to be identified and the corresponding insurance clause text;

[0006] Using a preset liability set extraction rule, a special liability set is extracted from the special text, and a liability clause set and an exemption clause set are extracted from the insurance clause text. Each of the special liability set, liability clause set, and exemption clause set is composed of the smallest semantic unit.

[0007] The set of special liabilities, the set of liability clauses, and the set of exemption clauses are input into a pre-trained semantic understanding model. The dot product attention mechanism is used to calculate the difference between the set of special liabilities and the set of liability clauses, as well as the intersection between the set of special liabilities and the set of exemption clauses.

[0008] The extended liability scope corresponding to the special text is determined based on the difference set, and the exemption liability scope corresponding to the special text is determined based on the intersection set.

[0009] Based on the types of liability included in the extended liability scope and the exemption liability scope, the risk level of the special liability clauses included in each type of liability is determined, and the content of the notice of the special liability clauses is determined based on the risk level.

[0010] To address the aforementioned technical problems, this application also provides a policy risk identification device, which employs the following technical solution:

[0011] The acquisition module is used to acquire the special text to be identified and the insurance clause text corresponding to the special text;

[0012] The extraction module is used to extract a set of special liabilities from the special text and a set of liability clauses and a set of exemption clauses from the insurance clause text using preset liability set extraction rules. The set of special liabilities, the set of liability clauses and the set of exemption clauses are all composed of the smallest semantic units.

[0013] The calculation module is used to input the set of special liabilities, the set of liability clauses, and the set of exemption clauses into a pre-trained semantic understanding model, and to use a dot product attention mechanism to calculate the difference between the set of special liabilities and the set of liability clauses, as well as the intersection between the set of special liabilities and the set of exemption clauses.

[0014] The determining module is used to determine the extended liability scope corresponding to the special text based on the difference set, and to determine the exemption liability scope corresponding to the special text based on the intersection set;

[0015] The grading module is used to determine the risk level of the special liability clauses included in each liability type based on the liability types of the extended liability scope and the exemption liability scope, and to determine the prompt content of the special liability clauses based on the risk level.

[0016] To address the aforementioned technical problems, this application also provides a computer device that employs the following technical solution: the computer device includes a memory and a processor, the memory stores computer-readable instructions, and the processor executes the computer-readable instructions to implement the steps of the policy risk identification method described in any of the above claims.

[0017] To address the aforementioned technical problems, this application also provides a computer-readable storage medium, employing the following technical solution: the computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the policy risk identification method described above.

[0018] Compared with the prior art, the embodiments of this application have the following main advantages:

[0019] This application's embodiment, by employing preset liability set extraction rules, can accurately extract liability sets from the contract text and insurance clause text, including the contract liability set, liability clause set, and exemption clause set. By using a pre-trained semantic understanding model, it can deeply understand the semantic information in the contract text and insurance clause text, and by employing a dot product attention mechanism to calculate the difference between the contract liability set and the liability clause set, as well as the intersection between the contract liability set and the exemption clause set, it can accurately identify the extended liability scope and the exemption liability scope corresponding to the contract text. By abstracting the semantic problem into a mathematical problem and processing the liability definition in a set-based manner, compared with traditional semantic comparison reasoning, it not only has a clear working mechanism and is easy to backtrack and explain, but also improves the accuracy and efficiency of identification. By determining the risk level of the contract liability clauses included in each liability type according to the liability type of the extended liability scope and the exemption liability scope, and determining the prompt content of the contract liability clauses according to the risk level, the business process becomes more scientific and reasonable, enabling hierarchical risk control and improving the efficiency and effectiveness of business processing. Attached Figure Description

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

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

[0022] Figure 2 This is a flowchart illustrating one embodiment of the policy risk identification method of this application;

[0023] Figure 3 This is a schematic diagram of the structure of one embodiment of the policy risk identification device of this application;

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

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

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

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

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

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

[0030] Terminal device 101 can be various electronic devices with a display screen and support web browsing. In addition to laptops 1011, tablets 1012, or mobile phones 1013, terminal device 101 can also be e-book readers, MP3 players (Moving Picture Experts Group Audio Layer III), MP4 players (Moving Picture Experts Group Audio Layer IV), laptops, and desktop computers.

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

[0032] It should be noted that the policy risk identification method provided in this application embodiment is generally executed by a server / terminal device, and correspondingly, the policy risk identification device is generally set in the server / terminal device.

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

[0034] Continue to refer to Figure 2 A flowchart of an embodiment of the question-and-answer method according to this application is shown. The policy risk identification method includes the following steps:

[0035] Step S201: Obtain the special text to be identified and the insurance clause text corresponding to the special text;

[0036] In this embodiment, the policy risk identification method operates on an electronic device (e.g., Figure 1 The server / terminal device shown can obtain the contract text to be identified and the corresponding insurance terms text through a wired or wireless connection. It should be noted that the aforementioned wireless connection methods may include, but are not limited to, 2G / 3G / 4G / 5G connections, WiFi connections, Bluetooth connections, WiMAX connections, Zigbee connections, UWB (ultra wideband) connections, and other currently known or future-developed wireless connection methods.

[0037] Specifically, this involves obtaining the special agreement text to be identified and the corresponding insurance policy text. The special agreement text refers to additional liability clauses agreed upon by both parties (the insurance company and the policyholder) outside the main contract during the insurance contract signing process, based on specific needs or circumstances. These special liability clauses may cover specific coverage, liability limitations, exclusions, and claim conditions. The insurance policy text is the core content of the insurance contract, stipulating the rights and obligations between the insurance company and the policyholder, including insurance liabilities, exclusions, and claims procedures. The insurance policy text can include the main insurance policy text and supplementary insurance policy texts. The main insurance policy text stipulates the basic content of the insurance contract and the basic rights and obligations of both parties; the supplementary insurance policy text supplements or expands upon the main insurance policy text.

[0038] Step S202: Using a preset liability set extraction rule, extract the special liability set from the special text, and extract the liability clause set and the exemption clause set from the insurance clause text. The special liability set, liability clause set and exemption clause set are all composed of the smallest semantic unit.

[0039] In this embodiment, a pre-defined set of liability extraction rules is established. These rules are formulated based on the characteristics of the insurance terms and may include rules for extracting special liabilities, rules for extracting liability clauses, and rules for extracting exclusion clauses. The pre-defined set of liability extraction rules are obtained, and then used to analyze the special liability text sentence by sentence, breaking it down into the smallest semantic units to extract the set of special liabilities. Similarly, the insurance terms text is analyzed sentence by sentence using the same rules, breaking it down into the smallest semantic units to extract the set of liability clauses and the set of exclusion clauses.

[0040] In some optional implementations, the above-mentioned method of extracting a set of liabilities from the special agreement text and extracting a set of liability clauses and a set of exemption clauses from the insurance clause text may include the following steps:

[0041] Obtain a preset liability description template, which includes liability tags related to the liability content. The liability tags are set according to preset liability set extraction rules, and include special liability tags, liability scope tags, and exemption scope tags.

[0042] Specifically, a preset liability description template is obtained. This template may include liability tags related to the liability content. These tags are set according to preset liability set extraction rules, which may include rules for extracting special liabilities, liability clauses, and exclusion clauses. The liability tags may include special liability tags, liability scope tags, and exclusion scope tags. Each liability tag corresponds to specific liability content, such as the liable party (insured, policyholder), the liable event (accidental injury, diagnosis of illness), the amount of compensation, the compensation ratio, and the conditions for compensation.

[0043] Using the aforementioned responsibility description template, the smallest semantic unit that matches the aforementioned responsibility tag is extracted from the special text as a set of special responsibilities;

[0044] Specifically, the obtained responsibility description template is applied to the special text, which is then divided into multiple smallest semantic units. The special responsibility tags in the responsibility description template are compared with each of the divided smallest semantic units, and the smallest semantic units that match the special responsibility tags are extracted as the special responsibility set.

[0045] Using the liability description template, the smallest semantic unit that matches the liability scope label is extracted from the insurance clause text as a liability clause set, and the smallest semantic unit that matches the exemption scope label is extracted from the insurance clause text as an exemption clause set.

[0046] Specifically, the obtained liability description template is applied to the insurance policy text, which is then broken down into multiple smallest semantic units. The liability scope tags in the liability description template are compared with each of the broken-down smallest semantic units, and the smallest semantic units matching each liability scope tag are extracted as a set of liability clauses. Similarly, the exemption scope tags in the liability description template are compared with each of the broken-down smallest semantic units, and the smallest semantic units matching each exemption scope tag are extracted as a set of exemption clauses.

[0047] This application embodiment sets a liability description template according to the liability set extraction rules, and compares the liability tags and various text contents in the liability description template. It can accurately identify the smallest semantic unit that matches the liability tag, efficiently extract key liability information from the special text and insurance clause text, and avoid semantic misunderstanding and omission.

[0048] Step S203: Input the set of special liabilities, the set of liability clauses, and the set of exemption clauses into the pre-trained semantic understanding model, and use the dot product attention mechanism to calculate the difference between the set of special liabilities and the set of liability clauses, as well as the intersection between the set of special liabilities and the set of exemption clauses;

[0049] In this embodiment, the semantic understanding model is a machine learning model used to understand and process semantic information in text. This model is built on deep learning technology and can capture information at the lexical, syntactic, and semantic levels in the text. The dot product attention mechanism is a method for calculating the relevance between texts. It is used to evaluate the similarity between vectors by calculating the dot product between two vectors based on dot product operations in vector space.

[0050] Specifically, the set of special liabilities, the set of liability clauses, and the set of disclaimers are input into a pre-trained semantic understanding model. In this model, these sets are converted into vector representations. A dot-product attention mechanism is used to calculate the difference between the vector representations of the special liabilities set and the set of liability clauses, and the intersection between the vector representations of the special liabilities set and the set of disclaimers. The difference includes liability clauses that exist in the set of special liabilities but not in the set of liability clauses; the intersection includes liability clauses that exist in both the set of special liabilities and the set of disclaimers.

[0051] In some optional implementations, the above-mentioned inputting the set of special liabilities, the set of liability clauses, and the set of disclaimers into a pre-trained semantic understanding model, and using a dot product attention mechanism to calculate the difference between the set of special liabilities and the set of liability clauses, as well as the intersection between the set of special liabilities and the set of disclaimers, may include the following steps:

[0052] The set of special liabilities, the set of liability clauses, and the set of exemption clauses are input into a pre-trained semantic understanding model, and features are extracted from the set of special liabilities, the set of liability clauses, and the set of exemption clauses respectively to obtain feature vectors of special liabilities, liability clauses, and exemption clauses.

[0053] Specifically, the sets of special liability agreements, liability clauses, and disclaimers are input into a pre-trained semantic understanding model. Optionally, before inputting the data into the pre-trained semantic understanding model, the sets of special liability agreements, liability clauses, and disclaimers can be pre-processed to obtain pre-processed sets of special liability agreements, liability clauses, and disclaimers. Using the pre-trained semantic understanding model, features are extracted from these pre-processed sets of special liability agreements, liability clauses, and disclaimers, respectively, resulting in feature vectors for special liability agreements, liability clauses, and disclaimers.

[0054] The dot product attention mechanism is used to calculate the first dot product between the special liability feature vector and the liability clause feature vector. The first similarity is determined based on the first dot product. The special liability feature vectors with the first similarity less than the first preset threshold are selected to determine the difference between the special liability set and the liability clause set.

[0055] Specifically, in this model, a dot product attention mechanism is used to calculate the first dot product between the special liability feature vector and the liability clause feature vector. Based on the calculated first dot product, a first similarity is calculated through normalization or a softmax function. A preset first threshold is obtained, and the first similarity is compared with the first preset threshold. Special liability feature vectors with a first similarity less than the first preset threshold are selected, and the selected special liability feature vectors are determined as the difference between the special liability set and the liability clause set.

[0056] A second dot product is calculated between the special liability feature vector and the disclaimer feature vector using a dot product attention mechanism. A second similarity is determined based on the second dot product. Special liability feature vectors with a second similarity greater than a second preset threshold are selected to determine the intersection of the special liability set and the disclaimer set.

[0057] Specifically, in this model, a dot product attention mechanism is used to calculate the second dot product between the feature vectors of the contractual liability and the feature vectors of the disclaimers. Based on the calculated second dot product, a second similarity is calculated through normalization or a softmax function. A preset second threshold is obtained, and the second similarity is compared with the second preset threshold. Feature vectors of the contractual liability that have a second similarity greater than the second preset threshold are selected, and the selected feature vectors of the contractual liability set and the disclaimer set are determined as the intersection of the contractual liability set and the disclaimer set.

[0058] This application embodiment efficiently extracts the feature vectors of the respective sets of special liability, liability clauses, and disclaimers by inputting them into a pre-trained semantic understanding model. By employing a dot product attention mechanism to calculate the dot product between feature vectors, subtle differences between them can be captured, thereby achieving accurate measurement of similarity and significantly improving computational efficiency and accuracy. By setting a preset threshold and selecting feature vectors based on the magnitude of similarity, the difference between the special liability set and the liability clause set, as well as the intersection between the special liability set and the disclaimer set, can be accurately determined, avoiding semantic misunderstandings and omissions.

[0059] In some alternative implementations, before inputting the aforementioned set of special liabilities, set of liability clauses, and set of disclaimers into the pre-trained semantic understanding model, the following steps may also be included:

[0060] Obtain a pre-collected training dataset, which includes a set of historical special liabilities, a set of historical liability clauses, and a set of historical disclaimers;

[0061] The training dataset is input into the initial semantic understanding model, and features are extracted from the historical special liability set, the historical liability clause set, and the historical disclaimer set to obtain the historical special liability feature vector, the historical liability clause feature vector, and the historical disclaimer feature vector. The initial semantic understanding model is constructed using a bidirectional encoder to represent the BERT model.

[0062] The third dot product between the historical special liability feature vector and the historical liability clause feature vector is calculated using a dot product attention mechanism. The third similarity is determined based on the third dot product. Historical special liability feature vectors with a third similarity less than a third preset threshold are selected to determine the training difference set between the historical special liability set and the historical liability clause set.

[0063] The fourth dot product between the historical special liability feature vector and the historical disclaimer feature vector is calculated using a dot product attention mechanism. The fourth similarity is determined based on the fourth dot product. Historical special liability feature vectors with a fourth similarity greater than a fourth preset threshold are selected to determine the training intersection of the historical special liability set and the historical disclaimer set.

[0064] Using a preset loss function, calculate the first loss value between the historical difference set labels and the training difference set, and the second loss value between the historical intersection set labels and the training intersection set;

[0065] Based on the first loss value and the second loss value, the model parameters of the initial semantic understanding model are updated. When the model meets the convergence condition, the pre-trained semantic understanding model is obtained.

[0066] Specifically, the initial semantic understanding model was built upon the bidirectional encoder BERT model, which incorporates a dot product attention mechanism. This mechanism calculates the dot product between feature vectors at different locations, enabling precise computation of the third dot product between the historical special liability feature vector and the historical liability clause feature vector, as well as the fourth dot product between the historical special liability feature vector and the historical disclaimer feature vector.

[0067] More specifically, the process of training the initial semantic understanding model is as follows: First, a pre-collected training dataset is acquired, which includes a set of historical special liabilities, a set of historical liability clauses, and a set of historical disclaimers. Then, the pre-collected training dataset is input into the initial semantic understanding model, which extracts features from the historical special liabilities, historical liability clauses, and historical disclaimers respectively, obtaining feature vectors for historical special liabilities, historical liability clauses, and historical disclaimers. Next, a dot product attention mechanism is used to calculate the third dot product between the historical special liabilities feature vector and the historical liability clause feature vector. A third similarity is determined based on the calculated third dot product, and a pre-set third threshold is obtained. The third similarity is compared with the third threshold, and historical special liabilities feature vectors with a third similarity less than the third threshold are selected, thus determining the training difference set between the historical special liabilities set and the historical liability clause set.

[0068] Furthermore, a fourth dot product is calculated between the historical special liability feature vector and the historical disclaimer feature vector using a dot product attention mechanism. The fourth similarity is determined based on the calculated fourth dot product. A pre-set fourth threshold is obtained. The fourth similarity is compared with the fourth threshold. Historical special liability feature vectors with a fourth similarity greater than the fourth threshold are selected. The training intersection of the historical special liability set and the historical disclaimer set is determined.

[0069] Next, obtain the preset loss function, historical difference labels, and historical intersection labels. Using the preset loss function, calculate the first loss value between the historical difference labels and the training difference set, and calculate the second loss value between the historical intersection labels and the training intersection set. Based on the calculated first and second loss values, update the model parameters of the initial semantic understanding model, and iterate the model training process until the model meets the convergence condition. Then, stop the model training process to obtain the pre-trained semantic understanding model.

[0070] Step S204: Determine the extended liability scope corresponding to the special text based on the difference set, and determine the exemption liability scope corresponding to the special text based on the intersection set;

[0071] In this embodiment, extended liability refers to the additional coverage added to the special agreement text relative to the insurance policy text; exclusions refer to the specific areas listed in the special agreement text where the insurance company is not liable for compensation. The extended liability scope corresponding to the special agreement text is determined based on the difference between the set of special liabilities and the set of liability clauses; the exclusions corresponding to the special agreement text are determined based on the intersection of the set of special liabilities and the set of exclusions.

[0072] Step S205: Based on the liability types of the extended liability scope and the exemption liability scope, determine the risk level of the special liability clauses included in each liability type, and determine the prompt content of the special liability clauses based on the risk level.

[0073] In this embodiment, the liability type to which each special liability clause in the extended liability scope and the exclusion of liability belongs is determined. The liability type may include, but is not limited to, personal injury liability, property damage liability, medical expense liability, legal liability, and other specific liabilities. Different liability types correspond to different risk assessment rules. For each liability type, the risk level of each special liability clause is assessed according to the risk assessment rules corresponding to that liability type. The risk level can be divided into two levels: low risk and high risk, or more precisely, multiple levels such as low risk, lower risk, medium risk, higher risk, and high risk. Based on the assessed risk level of each special liability clause, the corresponding prompt content for each special liability clause is determined. This prompt content may include, but is not limited to, underwriter review prompts, highlighted prompts, and transfer for review prompts.

[0074] In some alternative implementations, determining the risk level of the special liability clauses included in each liability type based on the types of liability for the extended liability scope and the exemption liability scope may include the following steps:

[0075] The liability types of each special liability clause included in the extended liability scope and the exemption liability scope shall be determined respectively;

[0076] Specifically, the expanded scope of liability and the exclusion of liability may include one or more special liability clauses. Based on the characteristics of each type of liability, the type of liability for each special liability clause included in the expanded scope of liability and the exclusion of liability shall be determined separately. The type of liability may include, but is not limited to, personal injury liability, property damage liability, medical expense liability, legal liability, and other specific liabilities.

[0077] Based on the type of liability, obtain the preset risk assessment rules;

[0078] Specifically, different risk assessment rules are pre-set according to different liability types, and these risk assessment rules are stored in a pre-set risk assessment rule base. The various special liability clauses included in the extended liability scope and the exemption liability scope are categorized and aggregated according to liability type to obtain the special liability clauses included in each liability type. For each special liability clause included in a liability type, the corresponding risk assessment rule is retrieved from the pre-set risk assessment rule base based on that liability type.

[0079] Using the aforementioned risk assessment rules, a risk assessment is conducted on the special liability clauses to obtain the risk level corresponding to the special liability clauses.

[0080] Specifically, for each type of liability and its specific liability clauses, the risk assessment rules corresponding to that type of liability are used to evaluate the risk level of each specific liability clause under that type of liability. This risk level can be divided into two levels: low risk and high risk, or more specifically, into multiple levels such as low risk, lower risk, medium risk, higher risk, and high risk.

[0081] This application's embodiments, by separately determining the liability types of each special liability clause included in the extended liability scope and the exemption liability scope, can accurately identify key risk points in the special liability text. These risk points may involve multiple aspects such as personal injury, property damage, medical expenses, and legal liability. Obtaining preset risk assessment rules based on the liability type and using these rules to assess the risks of the special liability clauses can greatly improve the accuracy and objectivity of risk assessment.

[0082] In some optional implementations, the above-mentioned risk assessment rules are used to assess the risk of the special liability clauses and obtain the risk level corresponding to the special liability clauses. This may include the following steps:

[0083] Based on the aforementioned risk assessment rules, obtain the number of liability clauses included in the target liability type;

[0084] Specifically, the target responsibility type is determined, and the number of responsibility clauses included in the target responsibility type is obtained according to the risk assessment rule corresponding to the target responsibility type. The risk assessment rule can be a quantity-based assessment rule.

[0085] If the number of liability clauses exceeds a preset threshold, the risk level of the special liability clauses included in the target liability type will be determined as a high-risk level.

[0086] If the number of liability clauses is less than or equal to a preset threshold, the risk level of the special liability clauses included in the target liability type will be determined as low risk.

[0087] Specifically, a preset quantity threshold is extracted from the quantity-based assessment rule, and the number of liability clauses obtained is compared with this preset quantity threshold. If the number of liability clauses is greater than the preset quantity threshold, it indicates that the special liability clauses under the target liability type are relatively complex and may involve higher compensation risks; therefore, the risk level of the special liability clauses included in the target liability type is determined to be high-risk. If the number of liability clauses is less than or equal to the preset quantity threshold, it indicates that the special liability clauses under the target liability type are relatively simple and the compensation risks are relatively low; therefore, the risk level of the special liability clauses included in the target liability type is determined to be low-risk.

[0088] This application embodiment, by counting the number of special liability clauses under the target liability type and comparing them with a preset number threshold, can perform preliminary risk screening on a large number of liability clauses in a short time, quickly classify the liability type into high-risk or low-risk levels, and greatly improve the efficiency of risk assessment.

[0089] In some alternative implementations, the above-mentioned prompt content for determining the special liability clause based on the risk level may include the following steps:

[0090] If the risk level is low risk, a highlighted prompt message will be generated based on the special liability clauses corresponding to the low risk level.

[0091] Specifically, if the risk level of the special liability clause is low, the low-risk level prompt content generation process will be executed. Based on the special liability clause corresponding to the low-risk level, a highlighted prompt message will be generated. The highlight can be displayed through color, bold font, or other visually emphasizing methods to attract the attention of users or reviewers.

[0092] If the risk level is high risk, a handover review prompt message will be generated based on the special liability clauses corresponding to the high risk level.

[0093] Specifically, if the risk level of a special liability clause is high, a high-risk level alert generation process is executed. Based on the special liability clause corresponding to the high-risk level, a transfer review alert is generated. This transfer review alert includes more detailed clause content, potential risk points, recommended review department, review process, or precautions, to ensure that high-risk clauses are assigned to a more specialized department for review.

[0094] This application's embodiments generate different forms of prompts for special liability clauses based on their risk levels, enabling rapid identification of low-risk and high-risk clauses. Highlighting low-risk clauses allows reviewers to quickly browse and confirm their risk controllability, while prompts for transferring high-risk clauses to a review team ensure these clauses receive in-depth review and evaluation by a professional team. This differentiated approach significantly improves the efficiency of risk identification.

[0095] This application embodiment, through the above-described scheme, specifically employs a preset liability set extraction rule to accurately extract the liability set from the contract text and insurance clause text, including the contract liability set, liability clause set, and exemption clause set. By using a pre-trained semantic understanding model, it can deeply understand the semantic information in the contract text and insurance clause text, and uses a dot product attention mechanism to calculate the difference between the contract liability set and the liability clause set, as well as the intersection between the contract liability set and the exemption clause set. This allows for precise identification of the extended liability scope and the exemption liability scope corresponding to the contract text. By abstracting the semantic problem into a mathematical problem and processing the liability definition in a set-based manner, compared to traditional semantic comparison reasoning, it not only has a clear working mechanism and is easy to backtrack and explain, but also improves the accuracy and efficiency of identification. By determining the risk level of the contract liability clauses included in each liability type based on the liability type of the extended liability scope and the exemption liability scope, and determining the prompt content of the contract liability clauses based on the risk level, the business process becomes more scientific and reasonable, enabling hierarchical risk control and improving the efficiency and effectiveness of business processing.

[0096] It should be emphasized that, in order to further ensure the privacy and security of the aforementioned contract text to be identified, insurance clause text, pre-trained semantic understanding model, and other related data, the aforementioned contract text to be identified, insurance clause text, pre-trained semantic understanding model, and other related data can also be stored in a blockchain node.

[0097] The blockchain referred to in this application is a novel application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanisms, and encryption algorithms. Essentially, a blockchain is a decentralized database, a chain of data blocks linked together using cryptographic methods. Each data block contains information about a batch of network transactions, used to verify the validity of the information (anti-counterfeiting) and generate the next block. A blockchain can include an underlying blockchain platform, a platform product service layer, and an application service layer.

[0098] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by instructing related hardware with computer-readable instructions. These computer-readable instructions can be stored in a computer-readable storage medium. When executed, the program can include the processes of the embodiments of the above methods. The aforementioned storage medium can be a non-volatile storage medium such as a magnetic disk, optical disk, or read-only memory (ROM), or random access memory (RAM).

[0099] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0100] Further reference Figure 3 As a response to the above Figure 2 The implementation of the method shown in this application provides an embodiment of a policy risk identification device, which is similar to... Figure 2 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.

[0101] like Figure 3 As shown, the policy risk identification device 400 described in this embodiment includes: an acquisition module 401, an extraction module 402, a calculation module 403, a determination module 404, and a grading module 405. Wherein:

[0102] The acquisition module 401 is used to acquire the special text to be identified and the insurance clause text corresponding to the special text;

[0103] Extraction module 402 is used to extract a set of special liabilities from the special text and a set of liability clauses and a set of exemption clauses from the insurance clause text using a preset set of liability extraction rules, wherein the set of special liabilities, the set of liability clauses and the set of exemption clauses are all composed of the smallest semantic units;

[0104] The calculation module 403 is used to input the set of special liabilities, the set of liability clauses, and the set of exemption clauses into a pre-trained semantic understanding model, and to use a dot product attention mechanism to calculate the difference between the set of special liabilities and the set of liability clauses, as well as the intersection between the set of special liabilities and the set of exemption clauses.

[0105] The determining module 404 is used to determine the extended liability scope corresponding to the special text based on the difference set, and to determine the exemption liability scope corresponding to the special text based on the intersection set;

[0106] The grading module 405 is used to determine the risk level of the special liability clauses included in each liability type based on the liability types of the extended liability scope and the exemption liability scope, and to determine the prompt content of the special liability clauses based on the risk level.

[0107] In this embodiment, the acquisition module 401 is used to acquire the special agreement text to be identified and the corresponding insurance clause text. The special agreement text refers to additional liability clauses agreed upon by both parties (the insurance company and the policyholder) outside the main contract during the insurance contract signing process, based on specific needs or circumstances. These special liability clauses may involve specific coverage, liability limitations, exclusions, and claim conditions. The insurance clause text is the core content of the insurance contract, stipulating the rights and obligations between the insurance company and the policyholder, including insurance liability, exclusions, and claims procedures. The insurance clause text may include the main insurance clause text and supplementary insurance clause texts. The main insurance clause text stipulates the basic content of the insurance contract and the basic rights and obligations of both parties; the supplementary insurance clause text supplements or expands the main insurance clause text.

[0108] Specifically, the extraction module 402 is used to pre-set the liability set extraction rules. These rules are formulated based on the characteristics of the insurance terms and may include rules for extracting special liabilities, rules for extracting liability clauses, and rules for extracting exclusion clauses. The module obtains the pre-set liability set extraction rules, uses these rules to analyze the special liability text sentence by sentence, breaking it down into the smallest semantic units to extract the special liability set; it also uses these rules to analyze the insurance terms text sentence by sentence, breaking it down into the smallest semantic units to extract the liability clause set and the exclusion clause set.

[0109] Specifically, a semantic understanding model is a machine learning model used to understand and process semantic information in text. This model is built on deep learning techniques and can capture information at the lexical, syntactic, and semantic levels in text. The dot product attention mechanism is a method for calculating the relevance between texts. It is used to evaluate the similarity between vectors by calculating the dot product between two vectors based on dot product operations in vector space.

[0110] Specifically, the calculation module 403 is used to input the set of special liabilities, the set of liability clauses, and the set of disclaimers into a pre-trained semantic understanding model. In this model, the set of special liabilities, the set of liability clauses, and the set of disclaimers are converted into vector representations. Using a dot product attention mechanism, the difference between the vector representation of the set of special liabilities and the vector representation of the set of liability clauses, and the intersection between the vector representation of the set of special liabilities and the vector representation of the set of disclaimers are calculated. The difference includes liability clauses that exist in the set of special liabilities but not in the set of liability clauses; the intersection includes liability clauses that exist in both the set of special liabilities and the set of disclaimers.

[0111] Specifically, extended liability refers to the additional coverage added to the special agreement text relative to the insurance policy text; exclusions refer to the specific areas listed in the special agreement text where the insurance company is not liable for compensation. Module 404 is used to determine the extended liability scope corresponding to the special agreement text based on the difference between the set of special liabilities and the set of liability policies; and to determine the exclusions corresponding to the special agreement text based on the intersection of the set of special liabilities and the set of exclusions.

[0112] Specifically, the grading module 405 is used to determine the liability type to which each special liability clause in the extended liability scope and the exclusion of liability scope belongs. The liability type may include, but is not limited to, personal injury liability, property damage liability, medical expense liability, legal liability, and other specific liabilities. Different liability types correspond to different risk assessment rules. For each liability type, the risk level of each special liability clause is assessed according to the risk assessment rules corresponding to that liability type. The risk level can be divided into two levels: low risk and high risk, or more finely divided into multiple levels such as low risk, lower risk, medium risk, higher risk, and high risk. Based on the assessed risk level of each special liability clause, the corresponding prompts for each special liability clause are determined. These prompts may include, but are not limited to, underwriter review prompts, highlighted prompts, and transfer-for-review prompts.

[0113] The policy risk identification device 400 of this application embodiment can accurately extract the liability set from the special agreement text and the insurance clause text by adopting a preset liability set extraction rule. This includes the special liability set, the liability clause set, and the exclusion clause set. By adopting a pre-trained semantic understanding model, it can deeply understand the semantic information in the special agreement text and the insurance clause text. It uses a dot product attention mechanism to calculate the difference between the special liability set and the liability clause set, as well as the intersection between the special liability set and the exclusion clause set. This allows it to accurately identify the extended liability scope and the exclusion liability scope corresponding to the special agreement text. By abstracting the semantic problem into a mathematical problem and processing the liability definition in a set-based manner, compared with traditional semantic comparison reasoning, it not only has a clear working mechanism and is easy to backtrack and explain, but also improves the accuracy and efficiency of identification. By determining the risk level of the special liability clauses included in each liability type according to the liability type of the extended liability scope and the exclusion liability scope, and determining the prompt content of the special liability clauses according to the risk level, the business process becomes more scientific and reasonable. It can achieve hierarchical control of risks and improve the efficiency and effectiveness of business processing.

[0114] In some optional implementations of this embodiment, the extraction module 402 may include a template acquisition submodule, a first matching submodule, and a second matching submodule. Wherein:

[0115] The template acquisition submodule is used to acquire a preset responsibility description template. The responsibility description template includes responsibility tags related to the responsibility content. The responsibility tags are set according to preset responsibility set extraction rules. The responsibility tags include special responsibility tags, responsibility scope tags, and exemption scope tags.

[0116] The first matching submodule is used to extract the smallest semantic unit that matches the special responsibility tag from the special text using the responsibility description template, as a special responsibility set;

[0117] The second matching submodule is used to extract the smallest semantic unit that matches the liability scope label from the insurance clause text as a liability clause set, and to extract the smallest semantic unit that matches the exemption scope label from the insurance clause text as an exemption clause set.

[0118] Specifically, the template acquisition submodule is used to obtain a preset liability description template. This template may include liability tags related to the liability content. These tags are set according to preset liability set extraction rules, which may include rules for extracting special liabilities, liability clauses, and exclusion clauses. The liability tags may include special liability tags, liability scope tags, and exclusion scope tags. Each liability tag corresponds to specific liability content, such as the liable party (insured, policyholder), the liable event (accidental injury, diagnosis of illness), the compensation amount, the compensation ratio, and the compensation conditions.

[0119] Specifically, the first matching submodule is used to apply the obtained responsibility description template to the special text, split the special text into multiple smallest semantic units, and compare the special responsibility tags in the responsibility description template with each of the split smallest semantic units one by one, and extract the smallest semantic units that match the special responsibility tags as the special responsibility set.

[0120] Specifically, the second matching submodule applies the obtained liability description template to the insurance clause text, splitting the insurance clause text into multiple smallest semantic units. It then compares the liability scope tags in the liability description template with each of the split smallest semantic units, extracting the smallest semantic units that match each liability scope tag as a set of liability clauses. Similarly, it compares the exemption scope tags in the liability description template with each of the split smallest semantic units, extracting the smallest semantic units that match each exemption scope tag as a set of exemption clauses.

[0121] The extraction module 402 in this embodiment sets a liability description template according to the liability set extraction rules, compares the liability tags and various text contents in the liability description template, and can accurately identify the smallest semantic unit that matches the liability tags. It can efficiently extract key liability information from the special text and insurance clause text, avoiding semantic misunderstandings and omissions.

[0122] In some optional implementations of this embodiment, the calculation module 403 may include a feature extraction submodule, a first calculation submodule, and a second calculation submodule. Wherein:

[0123] The feature extraction submodule is used to input the set of special liabilities, the set of liability clauses, and the set of exemption clauses into the pre-trained semantic understanding model, and to extract features from the set of special liabilities, the set of liability clauses, and the set of exemption clauses respectively to obtain feature vectors of special liabilities, liability clauses, and exemption clauses.

[0124] The first calculation submodule is used to calculate the first dot product between the special liability feature vector and the liability clause feature vector using a dot product attention mechanism, determine the first similarity based on the first dot product, select the special liability feature vector with the first similarity less than a first preset threshold, and determine the difference between the special liability set and the liability clause set.

[0125] The second calculation submodule is used to calculate the second dot product between the special liability feature vector and the disclaimer feature vector using a dot product attention mechanism, determine the second similarity based on the second dot product, select the special liability feature vector with the second similarity greater than the second preset threshold, and determine the intersection of the special liability set and the disclaimer set.

[0126] Specifically, the feature extraction submodule is used to input the set of contractual liabilities, the set of liability clauses, and the set of disclaimers into a pre-trained semantic understanding model. Optionally, before inputting the data into the pre-trained semantic understanding model, the set of contractual liabilities, the set of liability clauses, and the set of disclaimers can be preprocessed to obtain preprocessed sets of contractual liabilities, liability clauses, and disclaimers. Using the pre-trained semantic understanding model, features are extracted from the preprocessed sets of contractual liabilities, liability clauses, and disclaimers, respectively, to obtain feature vectors for contractual liabilities, liability clauses, and disclaimers.

[0127] Specifically, the first calculation submodule is used in the model to calculate the first dot product between the special liability feature vector and the liability clause feature vector using a dot product attention mechanism. Based on the calculated first dot product, a first similarity is calculated through normalization or a softmax function. A preset first threshold is obtained, and the first similarity is compared with the first preset threshold. Special liability feature vectors with a first similarity less than the first preset threshold are selected, and the selected special liability feature vectors are determined as the difference between the special liability set and the liability clause set.

[0128] Specifically, the second calculation submodule is used in this model to calculate a second dot product between the special liability feature vector and the disclaimer feature vector using a dot product attention mechanism. Based on the calculated second dot product, a second similarity is calculated through normalization or a softmax function. A preset second threshold is obtained, and the second similarity is compared with the second preset threshold. Special liability feature vectors with a second similarity greater than the second preset threshold are selected, and the selected special liability feature vectors are determined as the intersection of the special liability set and the disclaimer set.

[0129] The calculation module 403 in this embodiment can efficiently extract the feature vectors of the special liability set, the liability clause set, and the disclaimer set by inputting them into a pre-trained semantic understanding model. By using a dot product attention mechanism to calculate the dot product between feature vectors, it can capture subtle differences between feature vectors, thereby achieving accurate measurement of similarity and significantly improving computational efficiency and accuracy. By setting a preset threshold and selecting feature vectors based on the magnitude of similarity, it can accurately determine the difference between the special liability set and the liability clause set, as well as the intersection between the special liability set and the disclaimer set, avoiding semantic misunderstandings and omissions.

[0130] In some optional implementations of this embodiment, the policy risk identification device 400 may further include a model training module. The model training module is used to acquire a pre-collected training dataset, which includes a historical set of special liabilities, a historical set of liability clauses, and a historical set of exclusion clauses.

[0131] The training dataset is input into the initial semantic understanding model, and features are extracted from the historical special liability set, the historical liability clause set, and the historical disclaimer set to obtain the historical special liability feature vector, the historical liability clause feature vector, and the historical disclaimer feature vector. The initial semantic understanding model is constructed using a bidirectional encoder to represent the BERT model.

[0132] The third dot product between the historical special liability feature vector and the historical liability clause feature vector is calculated using a dot product attention mechanism. The third similarity is determined based on the third dot product. Historical special liability feature vectors with a third similarity less than a third preset threshold are selected to determine the training difference set between the historical special liability set and the historical liability clause set.

[0133] The fourth dot product between the historical special liability feature vector and the historical disclaimer feature vector is calculated using a dot product attention mechanism. The fourth similarity is determined based on the fourth dot product. Historical special liability feature vectors with a fourth similarity greater than a fourth preset threshold are selected to determine the training intersection of the historical special liability set and the historical disclaimer set.

[0134] Using a preset loss function, calculate the first loss value between the historical difference set labels and the training difference set, and the second loss value between the historical intersection set labels and the training intersection set;

[0135] Based on the first loss value and the second loss value, the model parameters of the initial semantic understanding model are updated. When the model meets the convergence condition, the pre-trained semantic understanding model is obtained.

[0136] Specifically, the initial semantic understanding model was built upon the bidirectional encoder BERT model, which incorporates a dot product attention mechanism. This mechanism calculates the dot product between feature vectors at different locations, enabling precise computation of the third dot product between the historical special liability feature vector and the historical liability clause feature vector, as well as the fourth dot product between the historical special liability feature vector and the historical disclaimer feature vector.

[0137] More specifically, the process of training the initial semantic understanding model is as follows: First, a pre-collected training dataset is acquired, which includes a set of historical special liabilities, a set of historical liability clauses, and a set of historical disclaimers. Then, the pre-collected training dataset is input into the initial semantic understanding model, which extracts features from the historical special liabilities, historical liability clauses, and historical disclaimers respectively, obtaining feature vectors for historical special liabilities, historical liability clauses, and historical disclaimers. Next, a dot product attention mechanism is used to calculate the third dot product between the historical special liabilities feature vector and the historical liability clause feature vector. A third similarity is determined based on the calculated third dot product, and a pre-set third threshold is obtained. The third similarity is compared with the third threshold, and historical special liabilities feature vectors with a third similarity less than the third threshold are selected, thus determining the training difference set between the historical special liabilities set and the historical liability clause set.

[0138] Furthermore, a fourth dot product is calculated between the historical special liability feature vector and the historical disclaimer feature vector using a dot product attention mechanism. The fourth similarity is determined based on the calculated fourth dot product. A pre-set fourth threshold is obtained. The fourth similarity is compared with the fourth threshold. Historical special liability feature vectors with a fourth similarity greater than the fourth threshold are selected. The training intersection of the historical special liability set and the historical disclaimer set is determined.

[0139] Next, obtain the preset loss function, historical difference labels, and historical intersection labels. Using the preset loss function, calculate the first loss value between the historical difference labels and the training difference set, and calculate the second loss value between the historical intersection labels and the training intersection set. Based on the calculated first and second loss values, update the model parameters of the initial semantic understanding model, and iterate the model training process until the model meets the convergence condition. Then, stop the model training process to obtain the pre-trained semantic understanding model.

[0140] In some optional implementations of this embodiment, the hierarchical module 405 may include a type determination submodule, a rule acquisition submodule, and a risk assessment submodule. Wherein:

[0141] The type determination submodule is used to determine the liability type of each special liability clause included in the extended liability scope and the exemption liability scope, respectively;

[0142] The rule acquisition submodule is used to acquire preset risk assessment rules based on the liability type;

[0143] The risk assessment submodule is used to perform a risk assessment on the special liability clauses using the risk assessment rules, and to obtain the risk level corresponding to the special liability clauses.

[0144] Specifically, the expanded scope of liability and the scope of exemption from liability may include one or more special liability clauses. The type determination submodule is used to determine the type of liability for each special liability clause included in the expanded scope of liability and the scope of exemption from liability, based on the characteristics of each type of liability. The type of liability may include, but is not limited to, personal injury liability, property damage liability, medical expense liability, legal liability, and other specific liabilities.

[0145] Specifically, the rule acquisition submodule is used to pre-set different risk assessment rules according to different liability types and store these risk assessment rules in a preset risk assessment rule base. The various special liability clauses included in the extended liability scope and the exemption liability scope are categorized and aggregated according to liability type to obtain the special liability clauses included in each liability type. For each special liability clause included in a liability type, the corresponding risk assessment rule is retrieved from the preset risk assessment rule base based on that liability type.

[0146] Specifically, the risk assessment submodule is used to assess the risk level of each special liability clause under each liability type using the risk assessment rules corresponding to that liability type. This risk level can be divided into two levels: low risk and high risk, or more precisely, into multiple levels such as low risk, lower risk, medium risk, higher risk, and high risk.

[0147] The hierarchical module 405 in this embodiment of the application can accurately identify key risk points in the special liability clauses by determining the liability types of each special liability clause included in the extended liability scope and the exemption liability scope. These risk points may involve multiple aspects such as personal injury, property damage, medical expenses, and legal liability. Obtaining preset risk assessment rules based on the liability type and using these rules to assess the risks of the special liability clauses can greatly improve the accuracy and objectivity of risk assessment.

[0148] In some optional implementations of this embodiment, the risk assessment submodule may include a quantity acquisition unit, a first determination unit, and a second determination unit. Wherein:

[0149] The quantity acquisition unit is used to acquire the number of liability clauses included in the target liability type according to the risk assessment rules.

[0150] The first determining unit is used to determine the risk level of the special liability clauses included in the target liability type as a high-risk level if the number of liability clauses is greater than a preset number threshold.

[0151] The second determining unit is used to determine the risk level of the special liability clauses included in the target liability type as low risk level if the number of liability clauses is less than or equal to a preset number threshold.

[0152] Specifically, the quantity acquisition unit is used to determine the target responsibility type and, according to the risk assessment rule corresponding to the target responsibility type, obtain the number of responsibility clauses included in the target responsibility type. The risk assessment rule can be a quantity-based assessment rule.

[0153] Specifically, the first determining unit is used to extract a preset quantity threshold from the quantity-based assessment rule and compare the obtained number of liability clauses with the preset quantity threshold. If the number of liability clauses is greater than the preset quantity threshold, it indicates that the special liability clauses under the target liability type are relatively complex and may involve higher compensation risks, and the risk level of the special liability clauses included in the target liability type is determined to be a high-risk level. The second determining unit is used to determine the risk level of the special liability clauses included in the target liability type as a low-risk level if the number of liability clauses is less than or equal to the preset quantity threshold, indicating that the special liability clauses under the target liability type are relatively simple and the compensation risks are relatively low.

[0154] The risk assessment submodule of this application embodiment can perform preliminary risk screening on a large number of liability clauses in a short time by counting the number of special liability clauses under the target liability type and comparing them with a preset number threshold, and quickly classify the liability type into high-risk or low-risk levels, which greatly improves the efficiency of risk assessment.

[0155] In some optional implementations of this embodiment, the hierarchical module 405 may include a first prompt submodule and a second prompt submodule. Wherein:

[0156] The first prompt submodule is used to generate a highlighted prompt message based on the special liability clause corresponding to the low risk level if the risk level is low risk level.

[0157] The second prompt submodule is used to generate a handover review prompt message based on the special liability clauses corresponding to the high-risk level if the risk level is high-risk level.

[0158] Specifically, the first prompt submodule is used to generate low-risk prompt content if the risk level of the special liability clause is low-risk. Based on the special liability clause corresponding to the low-risk level, it generates highlighted prompt information. Highlighting can be achieved through color, bold font, or other visually emphasizing methods to attract the attention of users or reviewers.

[0159] Specifically, the second notification submodule is used to generate high-risk notification content if the risk level of the special liability clause is high-risk. Based on the special liability clause corresponding to the high-risk level, it generates transfer review notification information. This transfer review notification information includes more detailed clause content, potential risk points, recommended review department, review process, or precautions, to ensure that high-risk clauses are assigned to more specialized departments for review.

[0160] The classification module 405 in this embodiment generates different forms of prompts for special liability clauses based on their risk levels, enabling rapid identification of low-risk and high-risk clauses. Highlighting low-risk clauses allows reviewers to quickly browse and confirm their risk controllability, while prompts for transferring high-risk clauses to a review team ensure that these clauses receive in-depth review and evaluation by a professional team. This differentiated approach significantly improves the efficiency of risk identification.

[0161] To address the aforementioned technical problems, embodiments of this application also provide a computer device. Please refer to [link / reference needed]. Figure 4 , Figure 4 This is a basic structural block diagram of the computer device in this embodiment.

[0162] The computer device 6 includes a memory 61, a processor 62, and a network interface 63 that are interconnected via a system bus. It should be noted that only the computer device 6 with memory 61, processor 62, and network interface 63 is shown in the figure; however, it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented alternatively. Those skilled in the art will understand that the computer device described here is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0163] The computer device can be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device can interact with the user via a keyboard, mouse, remote control, touchpad, or voice control.

[0164] The memory 61 includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 61 may be an internal storage unit of the computer device 6, such as the hard disk or memory of the computer device 6. In other embodiments, the memory 61 may also be an external storage device of the computer device 6, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the computer device 6. Of course, the memory 61 may include both the internal storage unit and its external storage device of the computer device 6. In this embodiment, the memory 61 is typically used to store the operating system and various application software installed on the computer device 6, such as computer-readable instructions for policy risk identification methods. In addition, the memory 61 can also be used to temporarily store various types of data that have been output or will be output.

[0165] In some embodiments, the processor 62 may be a central processing unit (CPU), controller, microcontroller, microprocessor, or other data processing chip. The processor 62 is typically used to control the overall operation of the computer device 6. In this embodiment, the processor 62 is used to execute computer-readable instructions stored in the memory 61 or to process data, for example, to execute computer-readable instructions for the policy risk identification method.

[0166] The network interface 63 may include a wireless network interface or a wired network interface, which is typically used to establish communication connections between the computer device 6 and other electronic devices.

[0167] This application also provides another embodiment, namely, providing a computer-readable storage medium storing computer-readable instructions that can be executed by at least one processor to cause the at least one processor to perform the steps of the policy risk identification method described above.

[0168] The computer device, computer-readable storage medium, and computer-readable instructions provided in this application, through processor execution using preset liability set extraction rules, can accurately extract liability sets from the contract text and insurance clause text, including the contract liability set, liability clause set, and exemption clause set. By employing a pre-trained semantic understanding model, it can deeply understand the semantic information in the contract text and insurance clause text, and use a dot product attention mechanism to calculate the difference between the contract liability set and the liability clause set, as well as the intersection between the contract liability set and the exemption clause set. This allows for precise identification of the extended liability scope and the exemption liability scope corresponding to the contract text. By abstracting semantic problems into mathematical problems and processing liability definitions in a set-based manner, compared to traditional semantic comparison reasoning, it not only has a clear working mechanism and is easy to backtrack and explain, but also improves the accuracy and efficiency of identification. By determining the risk level of the contract liability clauses included in each liability type based on the liability type of the extended liability scope and the exemption liability scope, and determining the prompt content of the contract liability clauses based on the risk level, the business process becomes more scientific and reasonable, enabling hierarchical risk control and improving the efficiency and effectiveness of business processing.

[0169] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0170] Obviously, the embodiments described above are only some embodiments of this application, not all embodiments. The accompanying drawings show preferred embodiments of this application, but do not limit the patent scope of this application. This application can be implemented in many different forms; rather, these embodiments are provided to provide a more thorough and comprehensive understanding of the disclosure of this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this application's specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the scope of patent protection of this application.

[0171] The software tools or components not belonging to our company that appear in the embodiments of this application are merely examples and do not represent actual use.

Claims

1. A method for identifying policy risks, characterized in that, Includes the following steps: Obtain the special agreement text to be identified and the corresponding insurance clause text; Using a preset liability set extraction rule, a special liability set is extracted from the special text, and a liability clause set and an exemption clause set are extracted from the insurance clause text. Each of the special liability set, liability clause set, and exemption clause set is composed of the smallest semantic unit. The set of special liabilities, the set of liability clauses, and the set of exemption clauses are input into a pre-trained semantic understanding model. The dot product attention mechanism is used to calculate the difference between the set of special liabilities and the set of liability clauses, as well as the intersection between the set of special liabilities and the set of exemption clauses. The extended liability scope corresponding to the special text is determined based on the difference set, and the exemption liability scope corresponding to the special text is determined based on the intersection set. Based on the types of liability in the extended liability scope and the exempted liability scope, determine the risk level of the special liability clauses included in each type of liability, and determine the content of the notice for the special liability clauses based on the risk level; The step of inputting the set of special liabilities, the set of liability clauses, and the set of exemption clauses into a pre-trained semantic understanding model, and using a dot product attention mechanism to calculate the difference between the set of special liabilities and the set of liability clauses, and the intersection between the set of special liabilities and the set of exemption clauses, includes: The set of special liabilities, the set of liability clauses, and the set of exemption clauses are input into a pre-trained semantic understanding model, and features are extracted from the set of special liabilities, the set of liability clauses, and the set of exemption clauses respectively to obtain feature vectors of special liabilities, liability clauses, and exemption clauses. The dot product attention mechanism is used to calculate the first dot product between the special liability feature vector and the liability clause feature vector. The first similarity is determined based on the first dot product. The special liability feature vectors with the first similarity less than the first preset threshold are selected to determine the difference between the special liability set and the liability clause set. A second dot product is calculated between the special liability feature vector and the disclaimer feature vector using a dot product attention mechanism. A second similarity is determined based on the second dot product. Special liability feature vectors with a second similarity greater than a second preset threshold are selected to determine the intersection of the special liability set and the disclaimer set.

2. The policy risk identification method according to claim 1, characterized in that, The steps of extracting a set of special liabilities from the special agreement text and extracting a set of liability clauses and a set of exclusion clauses from the insurance clause text using preset liability set extraction rules include: Obtain a preset liability description template, which includes liability tags related to the liability content. The liability tags are set according to preset liability set extraction rules and include special liability tags, liability scope tags, and exemption scope tags. Using the aforementioned responsibility description template, the smallest semantic unit that matches the aforementioned responsibility tag is extracted from the special text as a set of special responsibilities; Using the liability description template, the smallest semantic unit that matches the liability scope label is extracted from the insurance clause text as a liability clause set, and the smallest semantic unit that matches the exemption scope label is extracted from the insurance clause text as an exemption clause set.

3. The policy risk identification method according to claim 1, characterized in that, Before the step of inputting the set of special liabilities, the set of liability clauses, and the set of disclaimers into the pre-trained semantic understanding model, the method further includes: Obtain a pre-collected training dataset, which includes a set of historical special liabilities, a set of historical liability clauses, and a set of historical disclaimers; The training dataset is input into the initial semantic understanding model, and features are extracted from the historical special liability set, the historical liability clause set, and the historical disclaimer set to obtain the historical special liability feature vector, the historical liability clause feature vector, and the historical disclaimer feature vector. The initial semantic understanding model is constructed using a bidirectional encoder to represent the BERT model. The third dot product between the historical special liability feature vector and the historical liability clause feature vector is calculated using a dot product attention mechanism. The third similarity is determined based on the third dot product. Historical special liability feature vectors with a third similarity less than a third preset threshold are selected to determine the training difference set between the historical special liability set and the historical liability clause set. The fourth dot product between the historical special liability feature vector and the historical disclaimer feature vector is calculated using a dot product attention mechanism. The fourth similarity is determined based on the fourth dot product. Historical special liability feature vectors with a fourth similarity greater than a fourth preset threshold are selected to determine the training intersection of the historical special liability set and the historical disclaimer set. Using a preset loss function, calculate the first loss value between the historical difference set labels and the training difference set, and the second loss value between the historical intersection set labels and the training intersection set; Based on the first loss value and the second loss value, the model parameters of the initial semantic understanding model are updated. When the model meets the convergence condition, the pre-trained semantic understanding model is obtained.

4. The policy risk identification method according to claim 1, characterized in that, The step of determining the risk level of the special liability clauses included in each type of liability based on the types of liability of the extended liability scope and the exemption liability scope includes: The liability types of each special liability clause included in the extended liability scope and the exemption liability scope shall be determined respectively; Based on the type of liability, obtain the preset risk assessment rules; Using the aforementioned risk assessment rules, a risk assessment is conducted on the special liability clauses to obtain the risk level corresponding to the special liability clauses.

5. The policy risk identification method according to claim 4, characterized in that, The step of using the aforementioned risk assessment rules to conduct a risk assessment on the special liability clauses and obtain the risk level corresponding to the special liability clauses includes: Based on the aforementioned risk assessment rules, obtain the number of liability clauses included in the target liability type; If the number of liability clauses exceeds a preset threshold, the risk level of the special liability clauses included in the target liability type will be determined as a high-risk level. If the number of liability clauses is less than or equal to a preset threshold, the risk level of the special liability clauses included in the target liability type will be determined as low risk.

6. The policy risk identification method according to claim 5, characterized in that, The step of determining the content of the special liability clause based on the risk level includes: If the risk level is low risk, a highlighted prompt message will be generated based on the special liability clauses corresponding to the low risk level. If the risk level is high risk, a handover review prompt message will be generated based on the special liability clauses corresponding to the high risk level.

7. A policy risk identification device, characterized in that, The device includes: The acquisition module is used to acquire the special text to be identified and the insurance clause text corresponding to the special text; The extraction module is used to extract a set of special liabilities from the special text and a set of liability clauses and a set of exemption clauses from the insurance clause text using preset liability set extraction rules. The set of special liabilities, the set of liability clauses and the set of exemption clauses are all composed of the smallest semantic units. The calculation module is used to input the set of special liabilities, the set of liability clauses, and the set of exemption clauses into a pre-trained semantic understanding model, and to use a dot product attention mechanism to calculate the difference between the set of special liabilities and the set of liability clauses, as well as the intersection between the set of special liabilities and the set of exemption clauses. The determining module is used to determine the extended liability scope corresponding to the special text based on the difference set, and to determine the exemption liability scope corresponding to the special text based on the intersection set; The grading module is used to determine the risk level of the special liability clauses included in each liability type based on the liability types of the extended liability scope and the exemption liability scope, and to determine the prompt content of the special liability clauses based on the risk level; The calculation module includes a feature extraction submodule, a first calculation submodule, and a second calculation submodule. The feature extraction submodule is used to input the set of special liabilities, the set of liability clauses, and the set of exemption clauses into a pre-trained semantic understanding model, and to extract features from the set of special liabilities, the set of liability clauses, and the set of exemption clauses respectively, to obtain feature vectors of special liabilities, liability clauses, and exemption clauses. The first calculation submodule is used to calculate the first dot product between the special liability feature vector and the liability clause feature vector using a dot product attention mechanism, determine the first similarity based on the first dot product, select the special liability feature vector with the first similarity less than a first preset threshold, and determine the difference between the special liability set and the liability clause set. The second calculation submodule is used to calculate the second dot product between the special liability feature vector and the disclaimer feature vector using a dot product attention mechanism, determine the second similarity based on the second dot product, select the special liability feature vector with the second similarity greater than the second preset threshold, and determine the intersection of the special liability set and the disclaimer set.

8. A computer device, characterized in that, The system includes a memory and a processor, wherein the memory stores computer-readable instructions, and the processor executes the computer-readable instructions to implement the steps of the policy risk identification method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-readable instructions that, when executed by a processor, implement the steps of the policy risk identification method as described in any one of claims 1 to 6.

Images