Insurance claim settlement user intention identification method and device, equipment and storage medium

Abstract

The invention discloses a user intention recognition method, device and equipment for insurance claim settlement and a storage medium, and belongs to the technical field of artificial intelligence. Firstly, an initial data set is used for training a preliminary user intention recognition model; the initial data set is then extended using a large language model. And performing intention labeling on the extended data set. And performing intention prediction on the extended data set by using the preliminary model, and performing first iterative training on the model according to the labeled intention label and a preliminary prediction result. And determining a parameter updating strategy of the model by analyzing the difference between the user intention label and the prediction result, and performing model iteration training again by using a reinforcement learning method based on the updating strategy. And a second user intention recognition model with better performance is obtained after multiple times of iterative optimization. The invention also relates to the technical field of block chains, and the claim settlement data set is stored on the block chain node. According to the invention, the accuracy and generalization ability of user intention identification in an insurance claim settlement scene are significantly improved.

Description

Technical Field

[0001] The present application belongs to the field of artificial intelligence technology, and specifically relates to a method, device, equipment and storage medium for identifying user intentions for insurance claims. Background Art

[0002] With the development of the insurance industry, insurance companies have accumulated a large amount of claims data. These data contain detailed information such as customer personal information, insurance product information, reasons for claims, and amounts of claims. In-depth mining of these data can help insurance companies better understand customer needs, optimize product design, improve service quality, and reduce operating costs.

[0003] However, traditional claims user intention mining methods often rely on manual analysis and judgment, which is inefficient and has limited accuracy. Existing intention mining methods based on large models often require relatively comprehensive claims sample data. However, the number of claims sample data in certain scenarios is small, so the large model cannot be fully trained. The user intention results mined by the large model will be broad, with corresponding prediction errors, resulting in the inability to generate targeted specific user intentions. For example, in rare disease claims cases, due to the scarcity of relevant samples, the large model may not be able to accurately distinguish the disease from other diseases with similar symptoms, thereby mistakenly classifying the predicted intention as a more common disease claim. This not only affects the user experience, but may also bring unnecessary payment risks to the insurance company, resulting in corresponding economic losses. Summary of the invention

[0004] The purpose of the embodiments of the present application is to propose a method, device, computer equipment and storage medium for identifying user intentions for insurance claims, so as to solve the technical problem that the existing intention mining based on large models has corresponding prediction errors, resulting in the inability to generate targeted specific user intentions.

[0005] In order to solve the above technical problems, the embodiment of the present application provides a method for identifying user intentions for insurance claims, which adopts the following technical solutions:

[0006] A method for identifying user intentions for insurance claims, comprising:

[0007] Obtaining a first claims data set, and using the first claims data set to train a first user intent recognition model;

[0008] Using a preset large language model to expand the first claims data set, to obtain a second claims data set;

[0009] Perform intent labeling on the second claims data set to obtain user intent labels;

[0010] Use the first user intention recognition model to perform intention prediction on the second claims data set to obtain a first intention prediction result;

[0011] Performing a first iterative training on the first user intent recognition model using the annotated second claims data set, and obtaining a second intent prediction result during the iterative training process;

[0012] Determine a parameter update strategy according to the user intent label, the first intent prediction result, and the second intent prediction result;

[0013] Performing a second iterative training on the first user intention recognition model based on the parameter update strategy until the model is fitted, thereby obtaining a second user intention recognition model;

[0014] The claim data to be identified is received, and the claim data to be identified is input into a second user intention identification model to obtain a user intention identification result.

[0015] Furthermore, the first claim data set includes a plurality of first claim cases, each of which corresponds to a first user intention label. The steps of obtaining the first claim data set and using the first claim data set to train the first user intention recognition model specifically include:

[0016] Input each first claim case into the preset initial recognition model in turn;

[0017] The initial recognition model is enabled to sequentially learn the claim data features corresponding to each first claim case, and output the user intention recognition result corresponding to each first claim case, to obtain a first user intention recognition result;

[0018] Compare the first user intention recognition result with the first user intention label, and calculate a first prediction error;

[0019] The initial recognition model is iteratively updated based on the first prediction error to obtain a first user intention recognition model.

[0020] Further, the step of using the preset large language model to expand the first claims data set to obtain the second claims data set specifically includes:

[0021] Extract the entity of the claims data corresponding to each first claims case to obtain the claims case entity;

[0022] Fill in claim data based on the claim case entity to generate a new claim case;

[0023] Combine the new claims cases to obtain the second claims data set.

[0024] Further, the step of using the preset large language model to expand the first claims data set to obtain the second claims data set specifically includes:

[0025] Performing entity recognition on the claims data corresponding to each first claims case to determine the claims case entity of each first claims case;

[0026] Perform entity screening on the claim case entity of each first claim case to obtain the scenario entity of the claim case;

[0027] Obtain a new scenario entity from a preset scenario entity, and use the new scenario entity to replace the scenario entity of the claim case to obtain a new claim case;

[0028] Combine the new claims cases to obtain the second claims data set.

[0029] Further, the step of determining the parameter update strategy according to the user intention label, the first intention prediction result and the second intention prediction result specifically includes:

[0030] Calculate the matching degree between the user intention label and the first intention prediction result to obtain the first intention matching degree;

[0031] Calculate the matching degree between the user intention label and the second intention prediction result to obtain the second intention matching degree;

[0032] A parameter update strategy is determined according to the first intent matching degree and the second intent matching degree.

[0033] Furthermore, the parameter update strategy includes a parameter maintenance strategy and a parameter replacement strategy. The step of determining the parameter update strategy according to the first intention matching degree and the second intention matching degree specifically includes:

[0034] When the first intention matching degree is greater than the second intention matching degree, determining the parameter updating strategy to be the parameter maintaining strategy;

[0035] When the first intention matching degree is less than the second intention matching degree, the parameter updating strategy is determined to be a parameter replacement strategy.

[0036] Further, the first user intention recognition model is subjected to a second iterative training based on the parameter update strategy until the model is fitted to obtain a second user intention recognition model, specifically comprising:

[0037] Construct a reward function for iterative model training;

[0038] Taking the reward income as the update target, the first user intention recognition model is trained for the second iteration based on the parameter update strategy;

[0039] During the second iteration of training, the reward function is used to calculate the reward gain and identify the maximum reward gain;

[0040] The model corresponding to the maximum reward benefit is used as the second user intention recognition model.

[0041] In order to solve the above technical problems, the embodiment of the present application further provides a user intention recognition device for insurance claims, which adopts the following technical solution:

[0042] A user intention recognition device for insurance claims, comprising:

[0043] A pre-training module, used to obtain a first claims data set and train a first user intent recognition model using the first claims data set;

[0044] A data expansion module, used to expand the first claims data set using a preset large language model to obtain a second claims data set;

[0045] An intent labeling module, used to label the second claims data set with intent to obtain a user intent label;

[0046] an intention prediction module, configured to use the first user intention recognition model to perform intention prediction on the second claims data set to obtain a first intention prediction result;

[0047] A first iteration module is used to perform a first iteration training on the first user intention recognition model using the annotated second claims data set, and obtain a second intention prediction result during the iterative training process;

[0048] An update strategy module, used to determine a parameter update strategy according to the user intention label, the first intention prediction result and the second intention prediction result;

[0049] A second iteration module is used to perform a second iteration training on the first user intention recognition model based on a parameter update strategy until the model is fitted to obtain a second user intention recognition model;

[0050] The intention recognition module is used to receive the claims data to be recognized, and input the claims data to be recognized into the second user intention recognition model to obtain the user intention recognition result.

[0051] In order to solve the above technical problems, the embodiment of the present application further provides a computer device, which adopts the following technical solution:

[0052] A computer device comprises a memory and a processor, wherein the memory stores computer-readable instructions, and when the processor executes the computer-readable instructions, the steps of the method for identifying user intention for insurance claims as described in any one of the above are implemented.

[0053] In order to solve the above technical problems, the embodiment of the present application further provides a computer-readable storage medium, which adopts the following technical solution:

[0054] A computer-readable storage medium having computer-readable instructions stored thereon, wherein the computer-readable instructions, when executed by a processor, implement the steps of the method for identifying user intent for insurance claims as described in any one of the above.

[0055] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0056] The present application discloses a method, device, equipment and storage medium for identifying user intentions in insurance claims, and belongs to the field of artificial intelligence technology. The present application significantly improves the accuracy and generalization of user intention identification in insurance claims scenarios by introducing a large language model for data expansion and multiple rounds of model iteration training based on a reinforcement learning strategy. By generating more diverse claims cases, the training data range of the model is expanded, and at the same time, the model can be continuously optimized by combining intent annotation and reinforcement learning strategies. Ultimately, the user intent recognition model obtained has strong automated processing capabilities, and can quickly and accurately identify user intentions in the claims process, thereby improving the efficiency of insurance companies in handling claims applications, inquiries and other businesses, reducing manual intervention, and enhancing the level of intelligence. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the scheme in the present application, a brief introduction is given below to the drawings required for use in the description of the embodiments of the present application. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0058] Figure 1 An exemplary system architecture diagram is shown in which the present application can be applied;

[0059] Figure 2 A flow chart showing an embodiment of a method for identifying user intention for insurance claims according to the present application;

[0060] Figure 3 A schematic structural diagram of an embodiment of a user intention recognition device for insurance claims according to the present application is shown;

[0061] Figure 4 A schematic structural diagram of an embodiment of a computer device according to the present application is shown. DETAILED DESCRIPTION

[0062] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of the present application; the terms used in the specification of the application herein are only for the purpose of describing specific embodiments and are not intended to limit the present application; the terms "including" and "having" and any variations thereof in the specification and claims of the present application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of the present application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0063] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0064] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings.

[0065] like Figure 1 As shown, the system architecture 100 may include a terminal device 101, a network 102 and a server 103. The terminal device 101 may be a laptop 1011, a tablet computer 1012 or a mobile phone 1013. The network 102 is used to provide a medium for a communication link between the terminal device 101 and the server 103. The network 102 may include various connection types, such as wired, wireless communication links or optical fiber cables.

[0066] The user can use the terminal device 101 to interact with the server 103 through the network 102 to receive or send messages, etc. Various communication client applications can be installed on the terminal device 101, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.

[0067] The terminal device 101 can be any electronic device with a display screen and supporting web browsing. In addition to a laptop computer 1011, a tablet computer 1012 or a mobile phone 1013, the terminal device 101 can also be an e-book reader, an MP3 player (Moving Picture Experts Group Audio Layer III), an MP4 (Moving Picture Experts Group Audio Layer IV), a laptop computer, a desktop computer, etc.

[0068] The server 103 may be a server that provides various services, such as a background server that provides support for a web page displayed on the terminal device 101 .

[0069] It should be noted that the user intention identification method for insurance claims provided in the embodiment of the present application is generally executed by a server / terminal device, and accordingly, the user intention identification device for insurance claims is generally set in the server / terminal device.

[0070] It should be understood that Figure 1 The numbers of terminal devices, networks and servers in the embodiment are merely illustrative. The above system may have any number of terminal devices, networks and servers according to implementation requirements.

[0071] Continue to refer Figure 2 , shows a flow chart of an embodiment of a method for identifying user intentions for insurance claims according to the present application. The method for identifying user intentions for insurance claims comprises the following steps:

[0072] S201, obtaining a first claims data set, and using the first claims data set to train a first user intent recognition model.

[0073] Specifically, the first claim data set contains historical insurance claim applications, inquiries and other related data. These data usually include text descriptions submitted by customers, such as the reason for the insurance claim application, loss situation, etc., combined with the intent label of the data (such as "apply for claims", "query progress", etc.), to form the initial training set, that is, the first claim data set. Then, use these labeled data to train a basic user intent recognition model. The initial goal of this model is to be able to accurately identify the user's intent based on the input text, such as whether the user is applying for a claim or inquiring about the progress.

[0074] In a specific embodiment, a context encoder may be introduced to integrate text information in the first claims data set so that the system can understand the coherence of the conversation and improve the accuracy of intent recognition.

[0075] S202: Use a preset large language model to expand the first claims data set to obtain a second claims data set.

[0076] Specifically, the preset large language model, such as GPT, has the ability to generate and expand data sets. By inputting cases in the first claims data set, the large language model can generate new claims cases similar to these cases based on contextual semantics. This extension can cover different scenarios, wording changes, and even the diversity of claim types. For example, the large language model can generate more claims requests for other types of property damage, health insurance, etc. based on existing auto insurance claims cases. Through this extension, a second claims data set is formed, which is larger than the first data set and has a wider coverage. It can provide diverse user requests, help the model further generalize, handle more different types of intent expressions, and improve the training effect of the model.

[0077] S203: Perform intent labeling on the second claim data set to obtain a user intent label.

[0078] Specifically, after data expansion is completed, accurate intent annotation is required for the second claims data set. This step is usually performed manually or semi-automatically to ensure that the intent of each new claim request is correctly marked. For example, the system or annotator will assign appropriate labels to each claim case, such as "apply for claim", "check progress", "complaint", "cancel claim", etc. The accuracy of intent annotation directly affects the training effect of subsequent models. The annotation process needs to be analyzed and classified in detail in combination with business scenarios.

[0079] S204: Use the first user intention recognition model to perform intention prediction on the second claims data set to obtain a first intention prediction result.

[0080] Specifically, after the labeling is completed, the second claims data set is input into the first user intent recognition model. The first user intent recognition model will make intent predictions for each case in the second claims data set based on the knowledge it has learned, and output the prediction results, namely the first intent prediction results, which represent the user intent judged by the model based on the input text. At this point, since the model has only received preliminary training, the prediction results may have certain deviations or inaccuracies. The main purpose of this process is to test the performance of the model when encountering more diverse claims cases, so that by comparing with the annotated intent labels later, the system can identify in which scenarios the model performs poorly, thereby providing direction for parameter updates.

[0081] S205, performing a first iterative training on the first user intention recognition model using the labeled second claims data set, and obtaining a second intention prediction result during the iterative training process.

[0082] Specifically, in this step, the system uses the second claims dataset that has been annotated to further iterate the training of the first user intent recognition model. By inputting the second claims dataset that contains a wider range of scenarios, the model can learn in a more diverse context and gradually improve its prediction ability. At the end of each round of training, the system generates a second intent prediction result, which represents the model's prediction of the same dataset after iterative training. These results will be more accurate than the first intent prediction results because the model has been optimized on new data. This process is equivalent to the self-correction of the model, gradually improving its understanding of complex scenarios and various claims expressions.

[0083] S206, determining a parameter update strategy according to the user intention label, the first intention prediction result and the second intention prediction result.

[0084] Specifically, the focus of this step is to analyze the performance and gaps of the model by comparing the user intent label, the first intent prediction results, and the second intent prediction results. Through the comparison operation, the system can identify which types of intent prediction errors are higher, and in which scenarios the model's performance is still lacking. Based on these comparison results, the system will select a matching parameter update strategy, such as adjusting the model's weights, optimizing the learning rate, or introducing new features to improve the model's accuracy. This step can be regarded as the feedback link of the model. By continuously adjusting and optimizing the model parameters, the performance of the model can be further improved in subsequent training to ensure that the model better adapts to the complexity and diversity of the data.

[0085] S207, performing a second iterative training on the first user intention recognition model based on the parameter update strategy until the model is fitted, thereby obtaining a second user intention recognition model.

[0086] Specifically, after obtaining the parameter update strategies, the system will further iterate the training of the first user intent recognition model based on these strategies. This process will continue until the gap between the model's output and the actual annotation results is minimized, that is, the model reaches a fitting state. At this time, the model can make more accurate intent recognition for various claims scenarios, and finally generate an optimized second user intent recognition model, which has significant improvements in accuracy and robustness. Multiple iterative training and parameter adjustments ensure that the model can provide more intelligent and efficient intent recognition capabilities in complex claims scenarios, thereby realizing the automation of the claims process.

[0087] In a specific embodiment of the present application, the second iterative training combines the Self-Instruct reinforcement learning idea and PPO-Clip (Proximal Policy Optimization-Clip) dynamic policy optimization to achieve iterative training and model fitting.

[0088] Self-Instruct is a method that combines self-supervised learning and reinforcement learning, aiming to improve the performance of the model through continuous self-guidance and self-optimization. The core idea is that during the iteration of the model, the model continuously generates new data or feedback to improve its understanding of the task, and updates and optimizes the model under the guidance of feedback.

[0089] PPO-Clip is a commonly used reinforcement learning algorithm that is particularly good at dynamic policy optimization. Unlike traditional policy gradient methods, PPO-Clip maintains the stability of policy updates and limits the magnitude of each policy update to avoid drastic changes in model parameters that lead to unstable performance.

[0090] In this solution, by analyzing the results of each prediction, the model can generate feedback and dynamically adjust the prediction strategy. For example, if the model predicts a lot of errors in certain scenarios, the strategy will be dynamically adjusted through PPO-Clip to gradually reduce the error rate in these scenarios. PPO-Clip ensures that each update of the model will not cause drastic changes. Especially in complex claims scenarios, the model can be gradually optimized in small amounts to avoid instability caused by excessive one-time parameter adjustments.

[0091] Iterative training stops when the model performance reaches a certain stable state, that is, the model can accurately identify the user's intention in most claims scenarios. The combination of Self-Instruct and PPO-Clip allows the model to always optimize in the optimal direction during the training process, and find a balance between parameter adjustment and strategy update, and finally achieve model fitting.

[0092] S208, receiving the claim data to be identified, and inputting the claim data to be identified into a second user intention recognition model to obtain a user intention recognition result.

[0093] Specifically, in actual applications, when a new claim request is submitted to the system, the claims data to be identified will be sent as input to the optimized second user intent recognition model. Based on its learning experience in multiple rounds of training, the model will quickly determine the user's intention, such as whether to apply for a claim, check the claim status, or submit a complaint. The user intent recognition result can be directly used for subsequent claims process processing, such as automatic assignment to relevant processing personnel or systems, or triggering corresponding workflows. This step reflects the practicality of the entire technical solution, that is, through the optimized model, the system can greatly reduce manual intervention and improve the efficiency of claims processing and user experience.

[0094] Furthermore, the first claim data set includes a plurality of first claim cases, each of which corresponds to a first user intention label. The steps of obtaining the first claim data set and using the first claim data set to train the first user intention recognition model specifically include:

[0095] Input each first claim case into the preset initial recognition model in turn;

[0096] The initial recognition model is enabled to sequentially learn the claim data features corresponding to each first claim case, and output the user intention recognition result corresponding to each first claim case, to obtain a first user intention recognition result;

[0097] Compare the first user intention recognition result with the first user intention label, and calculate a first prediction error;

[0098] The initial recognition model is iteratively updated based on the first prediction error to obtain a first user intention recognition model.

[0099] In this embodiment, the first claim data set contains multiple specific claim cases, and each claim case has been marked with a corresponding user intent label, such as "apply for claim", "claim progress query", etc. In the step of obtaining the first claim data set, each claim case will be input into a preset initial recognition model in turn. The task of the initial recognition model is to learn the feature data of each claim case. These features include text description, claim type, amount, time and other key elements. By inputting these feature data, the initial recognition model generates the recognition results of the user intent. These results represent the model's preliminary judgment of the user intent of the case, that is, the first user intent recognition result.

[0100] The system then compares the recognition result with the first user intent label corresponding to each claim case, and calculates the difference between the two to obtain the first prediction error. The smaller the error, the higher the accuracy of the model; the larger the error, the lower the accuracy. Based on this error, the system adjusts the parameters of the initial recognition model and iteratively updates it, gradually optimizing the model so that it can more accurately match user intent.

[0101] Through the above steps, an initial intent recognition model can be trained. Through gradual training and error correction, the model can learn and gradually optimize its understanding of the characteristics of claims data, improve its ability to accurately predict user intentions, thereby reducing the error rate in subsequent practical applications and improving automated processing efficiency and user experience.

[0102] Further, the step of using the preset large language model to expand the first claims data set to obtain the second claims data set specifically includes:

[0103] Extract the entity of the claims data corresponding to each first claims case to obtain the claims case entity;

[0104] Fill in claim data based on the claim case entity to generate a new claim case;

[0105] Combine the new claims cases to obtain the second claims data set.

[0106] In this embodiment, first, entity extraction is performed on the claim data in each first claim case. Entity refers to the key information elements in the claim case, such as the identity information of the claimant, the type of insurance, the time of the accident, the amount of loss, the reason for the claim application, etc. These entities are automatically extracted from the text through natural language processing technology or other data processing methods. Next, based on these extracted claim entities, the existing claim data is supplemented and expanded to generate new claim cases. For example, by replacing, expanding or combining the extracted entities, a variety of claim scenarios can be generated to simulate the claim requests of different users, so that the system can generate multiple new claim cases and further enrich the data set. Finally, all the generated new claim cases are combined together to form a second claim data set.

[0107] Through the above steps, the system can generate diverse and representative claims data, further enriching the training data set and improving the model's ability to handle complex claims scenarios.

[0108] Further, the step of using the preset large language model to expand the first claims data set to obtain the second claims data set specifically includes:

[0109] Performing entity recognition on the claims data corresponding to each first claims case to determine the claims case entity of each first claims case;

[0110] Perform entity screening on the claim case entity of each first claim case to obtain the scenario entity of the claim case;

[0111] Obtain a new scenario entity from a preset scenario entity, and use the new scenario entity to replace the scenario entity of the claim case to obtain a new claim case;

[0112] Combine the new claims cases to obtain the second claims data set.

[0113] In this embodiment, entity recognition is first performed on the claims data of each first claim case, and key entity information such as accident type, compensation amount, application reason, etc. are automatically extracted from the claims data. These entity information represents the basic characteristics of the claim case. Then, these extracted entities are screened, focusing on entities closely related to the claim scenario, such as accident location, loss category, etc. These screened entities are called scenario entities. Then, the system obtains new scenario entities from the preset scenario entity library. These new entities can represent different claim situations or scenarios. Next, the system replaces the scenario entities in the original claim case with these new scenario entities to generate a new claim case. For example, the original case may involve a "traffic accident" scenario, and the new scenario entity can be replaced with a "fire accident" to generate a different claim scenario. Finally, all new claim cases are combined together to construct a new and richer second claim dataset.

[0114] Through the above steps, the system generates more claims data with different scenarios, further enriching the scenario diversity of the data set, thereby improving the model's ability to handle different claims scenarios.

[0115] Further, the step of determining the parameter update strategy according to the user intention label, the first intention prediction result and the second intention prediction result specifically includes:

[0116] Calculate the matching degree between the user intention label and the first intention prediction result to obtain the first intention matching degree;

[0117] Calculate the matching degree between the user intention label and the second intention prediction result to obtain the second intention matching degree;

[0118] A parameter update strategy is determined according to the first intent matching degree and the second intent matching degree.

[0119] In this embodiment, the matching degree between the user intention label and the first intention prediction result generated by the model is first calculated, that is, the first intention matching degree. The user intention label is the real user intention corresponding to the manually annotated claim scenario, and the first intention prediction result is the intention automatically recognized by the model based on the input data in the initial stage. The matching degree measures the similarity between the model recognition result and the real intention, which can usually be evaluated by indicators such as accuracy, recall or F1 value.

[0120] Next, after preliminary training, the model will re-predict based on the new data to generate a second intent prediction result. At this point, the system again calculates the match between the user intent label and the second intent prediction result to obtain the second intent match. By comparing the difference between the first intent match and the second intent match, the system can evaluate the progress of the model in iterative training. If the second intent match is significantly improved, it means that the model has become more accurate after one round of training; if the match does not change much, it means that the model parameters need to be further optimized. Finally, the system will formulate a parameter update strategy based on the changes in the two rounds of matching, determine which model parameters need to be adjusted, how much the adjustment should be, and the direction of training.

[0121] Through the above steps, the system can effectively evaluate the performance of the model in each round of training, and intelligently adjust the model parameters based on the matching degree difference to optimize the model performance. This dynamic adjustment strategy based on actual performance ensures that the model can continuously improve the accuracy and robustness of intent recognition in each iteration.

[0122] Furthermore, the parameter update strategy includes a parameter maintenance strategy and a parameter replacement strategy. The step of determining the parameter update strategy according to the first intention matching degree and the second intention matching degree specifically includes:

[0123] When the first intention matching degree is greater than the second intention matching degree, determining the parameter updating strategy to be the parameter maintaining strategy;

[0124] When the first intention matching degree is less than the second intention matching degree, the parameter updating strategy is determined to be a parameter replacement strategy.

[0125] In this embodiment, the system determines the parameter update strategy by comparing the first intent match and the second intent match. When the first intent match is greater than the second intent match, it means that after iterative training, the prediction effect of the model has not been improved, but has declined. At this time, the system selects the parameter maintenance strategy, that is, the model parameters are not updated, and the existing parameter settings are maintained to avoid the degradation of model performance due to over-training or incorrect adjustment. When the first intent match is less than the second intent match, it means that the recognition effect of the model has improved during the iteration process, and the system will select the parameter replacement strategy, that is, optimize and adjust the model parameters according to the training results to further improve the accuracy and generalization ability of the model.

[0126] Through the above steps, the system can automatically select appropriate parameter adjustment strategies based on the model performance, ensuring that the model training process is more stable and effective, and gradually improving the recognition effect.

[0127] Further, the first user intention recognition model is subjected to a second iterative training based on the parameter update strategy until the model is fitted to obtain a second user intention recognition model, specifically comprising:

[0128] Construct a reward function for iterative model training;

[0129] Taking the reward income as the update target, the first user intention recognition model is trained for the second iteration based on the parameter update strategy;

[0130] During the second iteration of training, the reward function is used to calculate the reward gain and identify the maximum reward gain;

[0131] The model corresponding to the maximum reward benefit is used as the second user intention recognition model.

[0132] In this embodiment, it is first necessary to construct a reward function for iterative model training. This reward function is obtained by quantitatively evaluating the performance of the model in identifying user intent, and usually includes multiple indicators, such as prediction accuracy, recall rate, and F1 value, etc. These indicators will be comprehensively considered to form an overall reward signal to guide the training direction of the model. After the construction is completed, the system will use the reward income as the update target and perform a second round of iterative training on the first user intent recognition model based on the previously defined parameter update strategy. During this process, the model will continue to receive new claims case data and use the reward function to calculate the reward income for each iteration during the training process.

[0133] In a specific embodiment, the reward function J(θ) is expressed as follows:

[0134]

[0135] In the formula, πθ represents the policy network parameters, τ is the trajectory from the initial state to the final state of the iteration, T represents the sum of all states, t represents the identifier of any state, r(s t , a t ) is in state s t Take iterative action a t is the immediate reward obtained when , and γ is the discount factor.

[0136] In addition, during the second iterative training process, the system will continuously evaluate the reward gains obtained from each round of training to identify the maximum reward gain. This process usually involves evaluating different model states during the training process to find the best performing model version. Ultimately, the system will select the model corresponding to the maximum reward gain as the second user intent recognition model. Through this reward feedback-based training method, the system can achieve more efficient model optimization, enabling it to achieve better performance levels in the continuous iteration process.

[0137] Through the above steps, the system can dynamically adjust the training strategy and ensure that the model is continuously optimized in each iteration through the reward mechanism, ultimately achieving high accuracy and robustness in user intent recognition.

[0138] In the above embodiment, the present application discloses a method for identifying user intentions in insurance claims, which belongs to the field of artificial intelligence technology. The present application significantly improves the accuracy and generalization of user intention identification in insurance claims scenarios by introducing a large language model for data expansion and multiple rounds of model iterative training based on a reinforcement learning strategy. By generating more diverse claims cases, the training data range of the model is expanded, and the model can be continuously optimized by combining intent annotation and reinforcement learning strategies. Ultimately, the user intent recognition model obtained has strong automated processing capabilities, and can quickly and accurately identify user intentions in the claims process, thereby improving the efficiency of insurance companies in handling claims applications, inquiries and other businesses, reducing manual intervention, and enhancing the level of intelligence.

[0139] In this embodiment, the method for identifying the user intention of insurance claims is executed on an electronic device (eg Figure 1 The server shown in the figure) can receive instructions or obtain data through a wired connection or a wireless connection. It should be noted that the above wireless connection method may include but is not limited to 3G / 4G connection, WiFi connection, Bluetooth connection, WiMAX connection, Zigbee connection, UWB (ultra wideband) connection, and other wireless connection methods currently known or to be developed in the future.

[0140] It should be emphasized that in order to further ensure the privacy and security of the above-mentioned claims data set information, the above-mentioned claims data set information can also be stored in a node of a blockchain.

[0141] The blockchain referred to in this application is a new application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanism, encryption algorithm, etc. Blockchain is essentially a decentralized database, a string of data blocks generated by cryptographic methods. Each data block contains a batch of network transaction information, which is used to verify the validity of its information (anti-counterfeiting) and generate the next block. Blockchain can include the underlying blockchain platform, platform product service layer, and application service layer.

[0142] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0143] AI basic technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. AI software technologies mainly include computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0144] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through computer-readable instructions, and the computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the aforementioned storage medium can be a non-volatile storage medium such as a disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0145] It should be understood that, although the steps in the flowchart of the accompanying drawings are displayed in sequence as indicated by the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least a part of the steps in the flowchart of the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

[0146] Further references Figure 3 , as a response to the above Figure 2 The present application provides an embodiment of a user intention recognition device for insurance claims. Figure 2 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0147] like Figure 3 As shown, the user intention recognition device 300 for insurance claims according to this embodiment includes:

[0148] A pre-training module 301 is used to obtain a first claims data set and use the first claims data set to train a first user intention recognition model;

[0149] A data expansion module 302 is used to expand the first claims data set using a preset large language model to obtain a second claims data set;

[0150] The intention labeling module 303 is used to label the second claim data set with intentions to obtain a user intention label;

[0151] The intention prediction module 304 is used to perform intention prediction on the second claim data set using the first user intention recognition model to obtain a first intention prediction result;

[0152] The first iteration module 305 is used to perform a first iteration training on the first user intention recognition model using the annotated second claim data set, and obtain a second intention prediction result during the iterative training process;

[0153] An update strategy module 306, configured to determine a parameter update strategy according to the user intent label, the first intent prediction result, and the second intent prediction result;

[0154] A second iteration module 307 is used to perform a second iteration training on the first user intention recognition model based on a parameter update strategy until the model is fitted to obtain a second user intention recognition model;

[0155] The intention recognition module 308 is used to receive the claim data to be recognized, and input the claim data to be recognized into the second user intention recognition model to obtain a user intention recognition result.

[0156] Furthermore, the first claim data set includes a plurality of first claim cases, each of which corresponds to a first user intention label, and the pre-training module 301 is specifically used for:

[0157] Input each first claim case into a preset initial recognition model in turn; enable the initial recognition model to learn the claim data features corresponding to each first claim case in turn, and output the user intention recognition result corresponding to each first claim case to obtain a first user intention recognition result; compare the first user intention recognition result with the first user intention label to calculate a first prediction error; iteratively update the initial recognition model based on the first prediction error to obtain a first user intention recognition model.

[0158] Furthermore, the data expansion module 302 is specifically used for:

[0159] Entity extraction is performed on the claim data corresponding to each first claim case to obtain a claim case entity; claim data is filled based on the claim case entity to generate a new claim case; and the new claim cases are combined to obtain a second claim data set.

[0160] Furthermore, the data expansion module 302 is also used for:

[0161] Perform entity recognition on the claim data corresponding to each first claim case to determine the claim case entity of each first claim case; perform entity screening on the claim case entity of each first claim case to obtain the scenario entity of the claim case; obtain a new scenario entity from the preset scenario entity, and use the new scenario entity to replace the scenario entity of the claim case to obtain a new claim case; combine the new claim cases to obtain a second claim data set.

[0162] Furthermore, the update strategy module 306 is specifically used for:

[0163] Calculate the matching degree between the user intention label and the first intention prediction result to obtain the first intention matching degree; calculate the matching degree between the user intention label and the second intention prediction result to obtain the second intention matching degree; determine the parameter update strategy according to the first intention matching degree and the second intention matching degree.

[0164] Furthermore, the parameter update strategy includes a parameter maintenance strategy and a parameter replacement strategy, and the update strategy module 306 is further used to:

[0165] When the first intention matching degree is greater than the second intention matching degree, the parameter update strategy is determined to be a parameter maintenance strategy; when the first intention matching degree is less than the second intention matching degree, the parameter update strategy is determined to be a parameter replacement strategy.

[0166] Furthermore, the second iteration module 307 is specifically used for:

[0167] Construct a reward function for iterative model training; take the reward benefit as the update target and perform a second iterative training on the first user intent recognition model based on the parameter update strategy; during the second iterative training, use the reward function to calculate the reward benefit and identify the maximum reward benefit; use the model corresponding to the maximum reward benefit as the second user intent recognition model.

[0168] In the above embodiment, the present application discloses a user intention recognition device for insurance claims, which belongs to the field of artificial intelligence technology. The present application significantly improves the accuracy and generalization ability of user intention recognition in insurance claims scenarios by introducing a large language model for data expansion and multiple rounds of model iterative training based on reinforcement learning strategies. By generating more diverse claims cases, the training data range of the model is expanded, and at the same time, the model can be continuously optimized by combining intent annotation and reinforcement learning strategies. Ultimately, the user intention recognition model obtained has strong automated processing capabilities, and can quickly and accurately identify user intentions in the claims process, thereby improving the efficiency of insurance companies in handling claims applications, inquiries and other businesses, reducing manual intervention, and enhancing the level of intelligence.

[0169] To solve the above technical problems, the present application also provides a computer device. Figure 4 , Figure 4 This is a basic structural block diagram of the computer device in this embodiment.

[0170] The computer device 4 includes a memory 41, a processor 42, and a network interface 43 that are interconnected through a system bus. It should be noted that the figure only shows a computer device 4 with a memory 41, a processor 42, and a network interface 43, but it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented instead. Among them, those skilled in the art can understand that the computer device here is a device that can automatically perform 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 (Application Specific Integrated Circuit, ASIC), programmable gate arrays (Field-Programmable Gate Array, FPGA), digital processors (Digital Signal Processor, DSP), embedded devices, etc.

[0171] The computer device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The computer device may interact with a user through a keyboard, a mouse, a remote controller, a touch pad, or a voice control device.

[0172] The memory 41 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), 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, disk, optical disk, etc. In some embodiments, the memory 41 can be an internal storage unit of the computer device 4, such as a hard disk or memory of the computer device 4. In other embodiments, the memory 41 can also be an external storage device of the computer device 4, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (FlashCard), etc. equipped on the computer device 4. Of course, the memory 41 can also include both the internal storage unit of the computer device 4 and its external storage device. In this embodiment, the memory 41 is generally used to store the operating system and various application software installed on the computer device 4, such as computer-readable instructions of the user intention recognition method for insurance claims, etc. In addition, the memory 41 can also be used to temporarily store various types of data that have been output or are to be output.

[0173] The processor 42 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip in some embodiments. The processor 42 is generally used to control the overall operation of the computer device 4. In this embodiment, the processor 42 is used to run computer-readable instructions stored in the memory 41 or process data, such as computer-readable instructions for running the user intention recognition method for insurance claims.

[0174] The network interface 43 may include a wireless network interface or a wired network interface. The network interface 43 is generally used to establish a communication connection between the computer device 4 and other electronic devices.

[0175] In the above embodiment, the present application discloses a computer device, which belongs to the field of artificial intelligence technology. The present application significantly improves the accuracy and generalization ability of user intent recognition in insurance claims scenarios by introducing a large language model for data expansion and multiple rounds of model iterative training based on reinforcement learning strategies. By generating more diverse claims cases, the training data range of the model is expanded, and the model can be continuously optimized by combining intent annotation and reinforcement learning strategies. Ultimately, the user intent recognition model obtained has strong automated processing capabilities, and can quickly and accurately identify user intent in the claims process, thereby improving the efficiency of insurance companies in handling claims applications, inquiries and other businesses, reducing manual intervention, and enhancing the level of intelligence.

[0176] The present application also provides another embodiment, namely, providing a computer-readable storage medium, wherein the computer-readable storage medium stores computer-readable instructions, and the computer-readable instructions can be executed by at least one processor to enable the at least one processor to perform the steps of the user intent identification method for insurance claims as described above.

[0177] In the above embodiments, the present application discloses a computer-readable storage medium, which belongs to the field of artificial intelligence technology. The present application significantly improves the accuracy and generalization ability of user intent recognition in insurance claims scenarios by introducing a large language model for data expansion and multiple rounds of model iterative training based on reinforcement learning strategies. By generating more diverse claims cases, the training data range of the model is expanded, and the model can be continuously optimized by combining intent annotation and reinforcement learning strategies. Ultimately, the user intent recognition model obtained has strong automated processing capabilities, and can quickly and accurately identify user intent in the claims process, thereby improving the efficiency of insurance companies in handling claims applications, inquiries and other businesses, reducing manual intervention, and enhancing the level of intelligence.

[0178] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0179] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronic devices, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. The present application can be described in the general context of computer executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0180] Obviously, the embodiments described above are only some embodiments of the present application, rather than all embodiments. The preferred embodiments of the present application are given in the accompanying drawings, but they do not limit the patent scope of the present application. The present application can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive. Although the present application is described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions recorded in the aforementioned specific implementation methods, or to perform equivalent replacement of some of the technical features therein. Any equivalent structure made using the contents of the specification and drawings of this application, directly or indirectly used in other related technical fields, is similarly within the scope of patent protection of this application.

Claims

1. A method for identifying user intentions for insurance claims, characterized in that: include: Acquire a first claims data set, and use the first claims data set to train a first user intent recognition model; Using a preset large language model to expand the first claims data set to obtain a second claims data set; Performing intent labeling on the second claim data set to obtain a user intent label; Using the first user intention recognition model to perform intention prediction on the second claims data set to obtain a first intention prediction result; Performing a first iterative training on the first user intention recognition model using the labeled second claim data set, and obtaining a second intention prediction result during the iterative training process; Determine a parameter update strategy according to the user intention label, the first intention prediction result and the second intention prediction result; Performing a second iterative training on the first user intention recognition model based on the parameter update strategy until the model is fitted, thereby obtaining a second user intention recognition model; The claim data to be identified is received, and the claim data to be identified is input into the second user intention identification model to obtain a user intention identification result.

2. The method for identifying user intention for insurance claims according to claim 1, characterized in that: The first claim data set includes a plurality of first claim cases, each of which corresponds to a first user intention label. The step of obtaining the first claim data set and using the first claim data set to train the first user intention recognition model specifically includes: Inputting each of the first claim cases into a preset initial recognition model in sequence; The initial recognition model is enabled to sequentially learn the claim data features corresponding to each of the first claim cases, and output a user intention recognition result corresponding to each of the first claim cases to obtain a first user intention recognition result; Compare the first user intention recognition result with the first user intention label to calculate a first prediction error; The initial recognition model is iteratively updated based on the first prediction error to obtain the first user intention recognition model.

3. The method for identifying user intention for insurance claims according to claim 2, characterized in that: The step of using the preset large language model to expand the first claims data set to obtain the second claims data set specifically includes: Perform entity extraction on the claim data corresponding to each of the first claim cases to obtain a claim case entity; Filling claim data based on the claim case entity to generate a new claim case; The new claim settlement cases are combined to obtain the second claim settlement data set.

4. The method for identifying user intention for insurance claims according to claim 2, characterized in that: The step of using the preset large language model to expand the first claims data set to obtain the second claims data set specifically includes: Performing entity recognition on the claim data corresponding to each of the first claim cases to determine a claim case entity of each of the first claim cases; Perform entity screening on the claim case entity of each of the first claim cases to obtain a scenario entity of the claim case; Obtaining a new scenario entity from a preset scenario entity, and using the new scenario entity to replace the scenario entity of the claim case to obtain a new claim case; The new claim settlement cases are combined to obtain the second claim settlement data set.

5. The method for identifying user intention for insurance claims according to claim 1, characterized in that: The step of determining a parameter update strategy according to the user intent label, the first intent prediction result, and the second intent prediction result specifically includes: Calculating a matching degree between the user intention label and the first intention prediction result to obtain a first intention matching degree; Calculating the matching degree between the user intention label and the second intention prediction result to obtain the second intention matching degree; Determine a parameter update strategy according to the first intent matching degree and the second intent matching degree.

6. The method for identifying user intention for insurance claims according to claim 5, characterized in that: The parameter update strategy includes a parameter maintenance strategy and a parameter replacement strategy. The step of determining the parameter update strategy according to the first intent matching degree and the second intent matching degree specifically includes: When the first intention matching degree is greater than the second intention matching degree, determining that the parameter update strategy is a parameter maintenance strategy; When the first intention matching degree is less than the second intention matching degree, the parameter update strategy is determined to be a parameter replacement strategy.

7. The method for identifying user intention for insurance claims according to claim 6, characterized in that: The step of performing a second iterative training on the first user intent recognition model based on the parameter update strategy until the model is fitted to obtain a second user intent recognition model specifically includes: Construct a reward function for iterative model training; Taking the reward income as the update target, and performing a second iterative training on the first user intention recognition model based on the parameter update strategy; In the second iterative training process, the reward function is used to calculate the reward benefit and identify the maximum reward benefit; The model corresponding to the maximum reward benefit is used as the second user intention recognition model.

8. A user intention recognition device for insurance claims, characterized in that: include: A pre-training module, used to obtain a first claims data set and train a first user intent recognition model using the first claims data set; A data expansion module, configured to use a preset large language model to expand the first claims data set to obtain a second claims data set; An intention labeling module, used to label the second claim data set with intentions to obtain a user intention label; an intention prediction module, configured to use the first user intention recognition model to perform intention prediction on the second claims data set to obtain a first intention prediction result; A first iteration module, configured to perform a first iteration training on the first user intention recognition model using the labeled second claim data set, and obtain a second intention prediction result during the iterative training process; An update strategy module, used to determine a parameter update strategy according to the user intention label, the first intention prediction result and the second intention prediction result; A second iterative module, used for performing a second iterative training on the first user intention recognition model based on the parameter update strategy until the model is fitted to obtain a second user intention recognition model; The intention recognition module is used to receive the claim data to be recognized, and input the claim data to be recognized into the second user intention recognition model to obtain a user intention recognition result.

9. A computer device, characterized in that: It comprises a memory and a processor, wherein the memory stores computer-readable instructions, and when the processor executes the computer-readable instructions, the steps of the method for identifying user intentions for insurance claims according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the method for identifying user intent for insurance claims according to any one of claims 1 to 7.

Images