Insurance scene verification method and device based on large language model and related equipment

Abstract

The embodiment of the invention provides an insurance scene verification method and device based on a large language model and related equipment, and belongs to the technical field of artificial intelligence. The method comprises the following steps: acquiring a target insurance scene image and a pre-trained large language model; inputting the target insurance scene image into a large language model, so that the large language model performs service classification on the target insurance scene image based on the service mapping constraint to obtain service type information, and inputting the target insurance scene image into a corresponding service association sub-model based on the service type information to perform service verification to obtain service verification data; and verifying the target insurance scene image based on the business verification data to obtain corresponding scene verification data. According to the embodiment of the invention, scene verification in a complex insurance scene can be realized through the large language model formed by the plurality of pre-trained large language sub-models, for example, personnel dressing and work cards in the scene and identification definition and vehicle information of vehicles in the scene are verified at the same time.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to an insurance scenario verification method, apparatus and related equipment based on a large language model. Background Art

[0002] At present, deep learning is widely used in the field of scene verification. Through deep learning models, target objects in the scene can be detected and the attribute characteristics of the target objects can be analyzed, thereby realizing compliance verification of the target objects in the scene, such as checking whether the clothing of people in the scene is compliant, whether the vehicle identification is compliant, checking the mortality of pigs or estimating the weight, etc.

[0003] Common algorithms include feature matching-based methods, deep learning-based methods, and target detection-based methods. However, these methods have some problems. In complex scenes with rich attributes, these models are often difficult to complete these tasks alone. Therefore, how to improve scene verification in complex scenes has become a technical problem that needs to be solved urgently. Summary of the invention

[0004] The main purpose of the embodiments of the present application is to propose an insurance scenario verification method, device and related equipment based on a large language model, aiming to realize scenario verification in complex scenarios.

[0005] To achieve the above-mentioned purpose, a first aspect of an embodiment of the present application proposes an insurance scenario verification method based on a large language model, the method comprising:

[0006] Acquire a target insurance scenario image and a pre-trained large language model; wherein the large language model includes a plurality of large language sub-models, the plurality of large language sub-models of the large language model include business association sub-models with business mapping constraints between each other, and each of the large language sub-models corresponds to a business capability;

[0007] Inputting the target insurance scene image into the large language model, so that the large language model performs business classification on the target insurance scene image based on the business mapping constraint to obtain business type information, and inputting the target insurance scene image into the corresponding business association sub-model for business verification based on the business type information to obtain business verification data;

[0008] The target insurance scene image is verified based on the business verification data to obtain scene verification data corresponding to the target insurance scene image.

[0009] In some embodiments, before acquiring the target scene image and the pre-trained large language model, a pre-training process for the large language model is also included, specifically including:

[0010] Acquire multiple original models and a first training data set; wherein the training scene images in the first training data set and the target insurance scene images belong to the same target business scene;

[0011] Performing classification and annotation processing on the first training data set according to the target business scenario to obtain a classified and annotated data set;

[0012] Based on the classified and labeled data set, the multiple original models are trained to obtain multiple pre-trained sub-models;

[0013] For each of the pre-trained sub-models, a corresponding prompt word is generated, and the prompt word is adapted to the corresponding pre-trained sub-model to obtain a plurality of pre-trained sub-models adapted to the prompt word;

[0014] The pre-trained sub-models adapted to the multiple prompt words are cascaded based on the business mapping constraints to obtain the pre-trained large language model.

[0015] In some embodiments, the training of the multiple original models based on the classified annotation dataset to obtain multiple pre-trained sub-models includes:

[0016] Based on the classified and labeled data set, a first original model is trained to obtain a first pre-trained sub-model; wherein the first original model is used to determine whether there is an object to be inspected in the scene image and whether it meets the verification requirements;

[0017] Based on the classified and labeled data set, a second original model is trained to obtain a second pre-trained sub-model; wherein the second original model is used to determine whether the inspection object in the scene image meets the preset verification standard;

[0018] According to the target business scenario, a business mapping constraint relationship is established between the first pre-trained sub-model and the second pre-trained sub-model.

[0019] In some embodiments, the pre-trained sub-models adapted to the multiple prompt words are cascaded based on the business mapping constraints to obtain the pre-trained large language model, including:

[0020] According to the business mapping constraint, determining the cascade order between the pre-trained sub-models adapted by the multiple prompt words, and constructing a cascade structure of the pre-trained sub-models;

[0021] The input and output of the pre-trained sub-models are adapted based on the business mapping constraints so that the input and output formats of adjacent pre-trained sub-models match.

[0022] In some embodiments, after the pre-trained sub-models adapted to the multiple prompt words are cascaded based on the business mapping constraint to obtain the pre-trained large language model, the method further includes:

[0023] Obtain a second training data set;

[0024] Inputting the second training data set into the large language model to obtain training prediction data;

[0025] Compare the predicted labels in the training prediction data with the real labels in the second training data set to obtain predicted labeled data;

[0026] The large language model is fine-tuned based on the predicted and labeled data to obtain a fine-tuned large language model.

[0027] In some embodiments, verifying the target insurance scene image based on the business verification data to obtain scene verification data corresponding to the target insurance scene image includes:

[0028] Binarizing each of the business verification data to obtain corresponding binary verification data;

[0029] Input each of the business verification data into a preset linear regression model of the corresponding binary verification data to obtain linear regression data;

[0030] When the linear regression data satisfies a preset verification condition, it is determined that the target insurance scene image has passed the verification, and scene verification data indicating that the verification has passed is obtained;

[0031] When the linear regression data does not satisfy a preset verification condition, it is determined that the target insurance scene image has not passed the verification, and scene verification data indicating that the verification has not passed is obtained.

[0032] In some embodiments, before inputting the target insurance scene image into the large language model, the method further includes:

[0033] Acquire the image size and image clarity of the target insurance scene image;

[0034] When the image size does not meet the preset size condition, resizing the target insurance scene image according to a preset target size; wherein the preset target size is the input size of the large language model;

[0035] When the clarity of the target insurance scene image is lower than a preset clarity threshold, a preset super-resolution model is used to perform super-resolution reconstruction on the target insurance scene image.

[0036] To achieve the above-mentioned purpose, a second aspect of an embodiment of the present application proposes a scene verification device based on a large language model, the device comprising:

[0037] An acquisition module, used to acquire a target insurance scenario image and a pre-trained large language model; wherein the large language model includes a plurality of large language sub-models, wherein the plurality of large language sub-models of the large language model include business association sub-models with business mapping constraints between each other, and each of the large language sub-models corresponds to a business capability;

[0038] A business verification module, used for inputting the target insurance scene image into the large language model, so that the large language model performs business classification on the target insurance scene image based on the business mapping constraint to obtain business type information, and inputting the target insurance scene image into the corresponding business association sub-model for business verification based on the business type information to obtain business verification data;

[0039] A scene verification module is used to verify the target insurance scene image based on the business verification data to obtain scene verification data corresponding to the target insurance scene image.

[0040] To achieve the above objectives, a third aspect of an embodiment of the present application proposes an electronic device, which includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the method described in the first aspect is implemented.

[0041] To achieve the above-mentioned purpose, the fourth aspect of an embodiment of the present application proposes a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method described in the first aspect is implemented.

[0042] Insurance scenario verification method based on large language model proposed in this application 、A device, an electronic device and a storage medium, which obtain a target insurance scene image and a pre-trained large language model; wherein the large language model includes multiple large language sub-models, and the multiple large language sub-models of the large language model include business association sub-models with business mapping constraints between each other, and each large language sub-model corresponds to a business capability; the target insurance scene image is input into the large language model so that the large language model classifies the target insurance scene image based on the business mapping constraints to obtain business type information, and based on the business type information, the target insurance scene image is input into the corresponding business association sub-model for business verification to obtain business verification data; the target insurance scene image is verified based on the business verification data to obtain scene verification data corresponding to the target insurance scene image. The present application utilizes a pre-trained large language model for insurance scene verification. Compared with traditional rules or single model methods, the large language model can better understand the scene semantics and extract key information, thereby improving verification accuracy. At the same time, insurance scenarios are complex and diverse, and a single verification method is difficult to meet the needs of all scenarios, and the cascade engineering of traditional models is very complicated. Therefore, this application divides the large language model into multiple sub-models, each sub-model corresponds to a business capability, and establishes associations between sub-models through business mapping constraints, so that the verification process can be parallelized and can flexibly adapt to complex business scenarios. Therefore, this application makes full use of the semantic understanding and multi-tasking capabilities of the large language model, and through the division of labor and cooperation of sub-models, it realizes efficient scenario verification in complex insurance scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 It is a flow chart of an insurance scenario verification method based on a large language model provided in an embodiment of the present application;

[0044] Figure 2 yes Figure 1 The flowchart before step S101 in FIG.

[0045] Figure 3 yes Figure 1 Flow chart of step S203 in FIG.

[0046] Figure 4 yes Figure 3 Flow chart of step S205 in FIG.

[0047] Figure 5 yes Figure 3 The flowchart after step S205 in FIG.

[0048] Figure 6 yes Figure 5 Flow chart of step S103 in FIG.

[0049] Figure 7 yes Figure 1 The flowchart before step S101 in FIG.

[0050] Figure 8 It is a structural diagram of a scene verification device based on a large language model provided in an embodiment of the present application;

[0051] Fig. 9 It is a schematic diagram of the hardware structure of the electronic device provided in the embodiment of the present application. DETAILED DESCRIPTION

[0052] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0053] It should be noted that, although the functional modules are divided in the device schematic diagram and the logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first", "second", etc. in the specification, claims and the above drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0054] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0055] First, some nouns involved in this application are analyzed:

[0056] Artificial intelligence (AI) is a new technical science that studies and develops theories, methods, technologies and application systems for simulating, extending and expanding human intelligence. AI is a branch of computer science. AI attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a similar way to human intelligence. Research in this field includes robots, language recognition, image recognition, natural language processing and expert systems. AI can simulate the information process of human consciousness and thinking. AI is also a 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 data.

[0057] Large Language Model (LLM): It is a language model in the field of natural language processing (NLP). It uses deep learning methods to train on massive text data to learn the statistical laws and knowledge of language, and has multiple natural language processing capabilities such as text generation, question answering, summarization, translation, and reasoning. Compared with traditional language models, large language models have the characteristics of large number of parameters (hundreds of millions to hundreds of billions), rich training corpus (hundreds of GB to hundreds of TB), and wide range of adaptability tasks. The emergence of large language models has greatly promoted the progress of artificial intelligence in natural language understanding and generation, and represents the cutting-edge level of current natural language processing technology. Through training on massive text data, the model learns the high-dimensional semantic representation of language and establishes the connection between words, phrases, and sentences. This semantic representation can be mapped to various downstream natural language processing tasks, making the large language model have a wide range of adaptability. The current mainstream large language models include OpenAI's GPT series, Google's BERT, T5, etc. They have made breakthrough progress in tasks such as question and answer, dialogue, text classification, and named entity recognition, and even reached a level comparable to human experts in some scenarios.

[0058] Based on this, the embodiments of the present application provide an insurance scenario verification method, apparatus and related equipment based on a large language model, aiming to realize scenario verification in complex insurance scenarios.

[0059] The insurance scenario verification method, device and related equipment based on a large language model provided in the embodiments of the present application are specifically illustrated through the following embodiments. First, the insurance scenario verification method based on a large language model in the embodiments of the present application is described.

[0060] 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 data.

[0061] 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.

[0062] The insurance scenario verification method based on a large language model provided in the embodiment of the present application relates to the field of artificial intelligence technology. The insurance scenario verification method based on a large language model provided in the embodiment of the present application can be applied to a terminal, can also be applied to a server side, and can also be software running in a terminal or a server side. In some embodiments, the terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, etc.; the server side can be configured as an independent physical server, or a server cluster or a distributed system composed of multiple physical servers, and can also be configured as a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application that implements an insurance scenario verification method based on a large language model, etc., but is not limited to the above forms.

[0063] 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 electronics, 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.

[0064] It should be noted that in each specific implementation of the present application, when it is necessary to perform relevant processing based on data related to user identity or characteristics such as user information, user behavior data, user historical data, and user location information, the user's permission or consent will be obtained first, and the collection, use, and processing of these data will comply with relevant laws, regulations, and standards. In addition, when the embodiment of the present application needs to obtain the user's sensitive personal information, the user's separate permission or consent will be obtained through a pop-up window or by jumping to a confirmation page. After clearly obtaining the user's separate permission or consent, the necessary user-related data for the normal operation of the embodiment of the present application will be obtained.

[0065] Figure 1 is an optional flowchart of the insurance scenario verification method based on a large language model provided in an embodiment of the present application. Figure 1The method may include but is not limited to steps S101 to S103.

[0066] Step S101, obtaining a target insurance scene image and a pre-trained large language model;

[0067] Step S102, inputting the target insurance scene image into the large language model, so that the large language model performs business classification on the target insurance scene image based on the business mapping constraint to obtain business type information, and inputting the target insurance scene image into the corresponding business association sub-model for business verification based on the business type information to obtain business verification data;

[0068] Step S103, verifying the target insurance scene image based on the business verification data to obtain scene verification data corresponding to the target insurance scene image.

[0069] Steps S101 to S103 shown in the embodiment of the present application are performed by acquiring a target insurance scene image and a pre-trained large language model; wherein the large language model includes multiple large language sub-models, and the multiple large language sub-models of the large language model include business association sub-models with business mapping constraints between each other, and each large language sub-model corresponds to a business capability; the target insurance scene image is input into the large language model so that the large language model classifies the target insurance scene image based on the business mapping constraints to obtain business type information, and based on the business type information, the target insurance scene image is input into the corresponding business association sub-model for business verification to obtain business verification data; the target insurance scene image is verified based on the business verification data to obtain scene verification data corresponding to the target insurance scene image. The present application utilizes a pre-trained large language model for insurance scene verification. Compared with traditional rules or single model methods, the large language model can better understand the scene semantics and extract key information, thereby improving verification accuracy. At the same time, insurance scenarios are complex and diverse, and a single verification method is difficult to meet the needs of all scenarios, and the cascade engineering of traditional models is very complicated. Therefore, this application divides the large language model into multiple sub-models, each sub-model corresponds to a business capability, and establishes associations between sub-models through business mapping constraints, so that the verification process can be parallelized and can flexibly adapt to complex business scenarios. Therefore, this application makes full use of the semantic understanding and multi-tasking capabilities of the large language model, and through the division of labor and cooperation of sub-models, it realizes efficient scenario verification in complex insurance scenarios.

[0070] In step S101 of some embodiments, it is first necessary to collect or receive insurance scene images that need to be verified. These images may come from on-site shooting, surveillance cameras or other image acquisition devices. A pre-trained large language model is previously trained with a large amount of data and can understand and process complex business logic and scene information. This large language model consists of multiple sub-models, each of which focuses on processing specific business capabilities, and these sub-models are interconnected through business mapping constraints, ensuring the collaborative work and efficient processing of the entire model.

[0071] In step S102 of some embodiments, the language model uses its internal business mapping constraint mechanism to perform business classification on the input image. This means that the model will identify the content in the image and classify it into specific business scenarios, such as vehicle insurance, personal insurance, or other types of insurance. Once the business type is determined, the model will further input the image into the corresponding business-related sub-model based on this information for more in-depth business verification. This verification process may include checking whether specific elements in the image meet insurance requirements, such as the degree of damage to the vehicle, the safety equipment of the personnel, etc. Through this step, key data on whether the image meets specific business requirements, namely business verification data, can be collected.

[0072] In step S103 of some embodiments, the business verification data obtained from the previous step can be used to conduct a comprehensive evaluation of the image. This may involve further analysis of the verification data, such as determining whether specific conditions in the image meet the insurance terms through a specific algorithm or model. Ultimately, scene verification data will be generated, which clearly indicates whether the image has passed the verification and the specific conditions or standards on which the verification results are based. In this way, the insurance scene verification method can not only provide verification results, but also provide detailed basis for verification, providing strong support for insurance decision-making.

[0073] See also Figure 2 In some embodiments, before step S101, a pre-training process for a large language model is also included, which may specifically include but is not limited to steps S201 to S205:

[0074] Step S201, obtaining multiple original models and a first training data set.

[0075] Step S202: classify and annotate the first training data set according to the target business scenario to obtain a classified annotated data set.

[0076] Step S203: Based on the classified and labeled data set, multiple original models are trained to obtain multiple pre-trained sub-models.

[0077] Step S204: for each pre-trained sub-model, a corresponding prompt word is generated, and the prompt word is adapted to the corresponding pre-trained sub-model to obtain a plurality of prompt word-adapted pre-trained sub-models.

[0078] Step S205 , cascading the pre-trained sub-models adapted to the multiple prompt words based on the business mapping constraints to obtain a pre-trained large language model.

[0079] In step S201 of some embodiments, the original model generally refers to an untrained machine learning model, which can be a deep learning network, such as a convolutional neural network (CNN) or other types of neural networks. These models will be used in subsequent training processes. The first training data set contains a large amount of images and related data, which are related to the target insurance scenario, such as vehicle damage pictures in vehicle insurance scenarios, and pictures of injured people in life insurance scenarios. These data sets are the basis for training models, and their quality and representativeness directly affect the performance of the model.

[0080] In step S202 of some embodiments, after the first training data set is acquired, the next step is to classify and label it according to the target business scenario. This step involves identifying the images in the data set and classifying them according to the type of insurance scenario they represent. For example, if the data set contains vehicle insurance scenarios, then the images may need to be classified into categories such as "vehicle with minor damage" and "vehicle with serious damage". At the same time, specific elements in the image need to be labeled, such as vehicle model, damaged parts, etc. This step is key to ensuring that the model can understand and recognize different insurance scenarios.

[0081] In step S203 of some embodiments, each original model is trained to recognize and process a specific type of insurance scenario. For example, one model may be trained to recognize the extent of damage to a vehicle, and another model may be trained to recognize whether a person's focus meets the standard. Through a large amount of training, the model can learn how to extract useful information from images and make accurate judgments based on this information.

[0082] In step S204 of some embodiments, corresponding prompt words need to be generated for each pre-trained sub-model. Prompt words are a special input that can guide the model to focus on specific parts or features in the image. For example, if a sub-model is trained to recognize vehicle models, the prompt words may contain keywords such as "vehicle model" and "car logo". By adapting these prompt words, the model's recognition and processing capabilities for specific business scenarios can be further improved.

[0083] In step S205 of some embodiments, all pre-trained sub-models adapted by the prompt words are cascaded according to the business mapping constraints. This step involves building a complex model network, in which each sub-model is responsible for processing specific business logic. For example, one sub-model may be responsible for identifying whether there is an object to be inspected in the scene, and another sub-model may be responsible for judging whether the inspected object meets the preset verification standard. In this way, it can be ensured that the entire large language model can efficiently and accurately execute complex business processes when processing insurance scene images. For example, in the scene of pig experience mark, it may be necessary to identify the individual identity and behavior of pigs. During the pre-training process, the original model may be trained to identify the characteristics of pigs, such as body shape, color, etc. Then, the generated prompt words may include "pig body shape", "pig color", etc., to guide the model to pay attention to these characteristics. Finally, these pre-trained sub-models will be cascaded to form a large language model that can comprehensively judge whether the pig body meets the verification standard.

[0084] Through steps S201 to S205, it can be ensured that the large language model can accurately identify and process various complex insurance scenarios in practical applications, thereby improving the level of automation and intelligence in the insurance industry.

[0085] See also Figure 3 In some embodiments, step S203 may also include but is not limited to steps 301 to S303:

[0086] Step S301: Based on the classified and labeled data set, the first original model is trained to obtain a first pre-trained sub-model.

[0087] Step S302: Based on the classified and labeled data set, the second original model is trained to obtain a second pre-trained sub-model.

[0088] Step S303: establishing a business mapping constraint relationship between the first pre-trained sub-model and the second pre-trained sub-model according to the target business scenario.

[0089] In step S301 of some embodiments, the focus is on training the first original model, the purpose of which is to determine whether there are objects to be inspected in the scene image and whether these objects meet the verification requirements. This involves the analysis of a large number of classified and labeled data sets, which contain images of various insurance scenes, each of which is detailedly annotated to determine whether there are objects to be inspected and whether they meet specific verification standards. Through this training, the first pre-trained sub-model can learn to identify and evaluate key elements in the scene, laying the foundation for the subsequent verification process.

[0090] In step S302 of some embodiments, the second original model is trained to determine whether the inspection object in the scene image meets the preset verification standard. This is also based on the classified annotation data set, but the training focuses on evaluating specific attributes of the object, such as the degree of damage, the integrity of safety equipment, etc. The second pre-trained sub-model can accurately evaluate whether the object meets the verification standard in the insurance business by learning these annotated data.

[0091] In step S303 of some embodiments, in steps S301 and S302, two pre-trained sub-models specifically trained for different tasks are obtained respectively. The purpose of step S303 is to ensure that the two sub-models can work together to form a coherent verification process. The business mapping constraint relationship defines the logical connection and data flow between the sub-models, ensuring that the output of the first sub-model can provide the necessary input for the second sub-model, or ensuring that the outputs of the two models can be combined to form a comprehensive verification result. This relationship may include rules, conditions, or priority settings to ensure the accuracy and efficiency of the verification process.

[0092] For example, if the first pre-trained sub-model is responsible for identifying whether there are objects to be inspected in the scene, and the second pre-trained sub-model is responsible for evaluating the properties of these objects, then in step S303, it is necessary to ensure that the output of the first model (i.e., the judgment of the existence of the object) can correctly guide the second model for further evaluation. This business mapping constraint relationship may involve data format conversion, triggering of specific conditions, or priority sorting of model outputs.

[0093] In some embodiments, if the first pre-trained sub-model is trained to identify the presence of pigs, and the second pre-trained sub-model is trained to assess the health of pigs, then in step S303, it is necessary to ensure that when the first model confirms that there are pigs in the scene, the second model can receive this information and perform health assessment accordingly. Such a business mapping constraint relationship ensures the coherence and effectiveness of the entire verification process, allowing the large language model to integrate the outputs of multiple sub-models and make accurate business decisions.

[0094] Through the training process of steps S301 to S303, two pre-trained sub-models specifically optimized for different business needs are obtained. The first pre-trained sub-model focuses on identifying key objects in insurance scenarios, such as the presence of vehicles or personnel, while the second pre-trained sub-model evaluates specific attributes of these objects, such as the degree of damage or dress standards. This division of labor and cooperation enables each sub-model to conduct in-depth learning in its field of expertise, thereby improving the model's processing capabilities and accuracy for specific tasks. The establishment of business mapping constraint relationships is the key to achieving model collaboration. In this way, not only the consistency and logic of the model output are ensured, but also the flexibility and scalability of the large language model are improved. When business needs change or new business scenarios emerge, the new sub-model can be quickly integrated into the large language model by adjusting the business mapping constraint relationship without large-scale reconstruction.

[0095] See also Figure 4 In some embodiments, step S205 may include but is not limited to steps S401 to S402:

[0096] Step S401, according to the business mapping constraint, determine the cascade order between the pre-trained sub-models adapted to the multiple prompt words, and construct the cascade structure of the pre-trained sub-models.

[0097] Step S402, adapting the input and output of the pre-trained sub-models based on the business mapping constraints to match the input and output formats of adjacent pre-trained sub-models.

[0098] In step S401 of some embodiments, according to the business mapping constraints, it is necessary to analyze the functions and business requirements of each sub-model to determine their order in the processing flow. For example, it may be necessary to first identify whether there are objects to be inspected in the scene, and then evaluate whether the attributes of these objects meet the verification criteria. Therefore, the sub-model responsible for identifying the existence of objects may need to be executed before the sub-model responsible for evaluating the attributes. Building such a cascade structure ensures that data can flow between models in the correct order, so that the entire large language model can efficiently and accurately process insurance scenarios.

[0099] For example, in the pig recognition scenario, it is necessary to first identify whether there are pigs in the image (the task of the first pre-trained sub-model), and then evaluate the health status of these pigs (the task of the second pre-trained sub-model). Step S401 is to determine the execution order of these two tasks, that is, first perform pig recognition, and then health assessment. This order is based on the needs of business logic, because only when the existence of pigs is confirmed, it is meaningful to evaluate their health status.

[0100] In step S402 of some embodiments, the input and output of the pre-trained sub-model are adapted to ensure that the data formats and interfaces between adjacent sub-models match. This is because different sub-models may have different requirements for the format and type of input data. For example, one sub-model may output labels in a specific format, while the next sub-model requires these labels as input in a different format. In this case, the output needs to be adapted to ensure that it can be correctly understood and processed by the next sub-model. This adaptation process may include operations such as data format conversion, feature extraction, and information enhancement to ensure that the transfer of data between models is seamless and efficient.

[0101] Through step S401 to step S402, not only the logical order and data flow between sub-models are ensured, but also the flexibility and scalability of the scenario verification model are improved. This sophisticated model integration method enables the large language model to handle complex insurance scenarios more accurately, improves the accuracy and efficiency of verification, and also facilitates possible model expansion or modification in the future.

[0102] See also Figure 5 In some embodiments, after step S205, the following steps may also be performed, including but not limited to steps S501 to S503:

[0103] Step S501, obtaining a second training data set.

[0104] Step S502: input the second training data set into the large language model to obtain training prediction data.

[0105] Step S503, comparing the predicted labels in the training prediction data with the real labels in the second training data set to obtain predicted labeled data.

[0106] Step S505 , fine-tune the large language model based on the predicted annotation data to obtain a fine-tuned large language model.

[0107] In step S501 of some embodiments, a second training data set is collected for further training and optimization of the pre-trained large language model. This data set should contain images in a variety of scenarios, especially those complex scenarios that may be encountered in practical applications. For example, in the scenario of vehicle and personnel verification, this data set may include vehicle and personnel images under various lighting conditions, different angles and different backgrounds. These images need to be accurately annotated to ensure that the model can learn from these data how to accurately identify and evaluate the characteristics of vehicles and personnel.

[0108] In step S502 of some embodiments, the second training data set obtained in step S501 is input into the pre-trained large language model. The model predicts these images and generates predicted labels. These predicted labels represent the model's understanding and judgment of the image content based on its current training state, such as judging whether there are vehicles or people to be inspected in the image, and whether they meet specific verification standards. Since these predictions are generated based on the current training state of the model, they may not be completely accurate. The model may not be able to correctly identify specific attributes of vehicles or people in some cases, or may misjudge complex scenes. Therefore, these prediction results need further verification and adjustment.

[0109] In step S503 of some embodiments, the predicted labels obtained in step S502 are compared with the real labels in the second training data set. This comparison process is key to evaluating the accuracy of model predictions. By comparing the predicted labels and the real labels, it is possible to identify where the model does well and where it needs improvement. For example, if the model performs poorly in identifying a specific type of vehicle or a person in specific clothing, this information will be used to guide subsequent model fine-tuning.

[0110] In step S504 of some embodiments, the large language model is fine-tuned using the predicted annotation data obtained in step S503. The fine-tuning process involves adjusting the parameters of the model to reduce prediction errors and improve the accuracy of the model. This may include adjusting the weights of the model, optimizing the hyperparameters of the algorithm, or modifying the structure of the model. The fine-tuned model should be able to more accurately identify and evaluate the characteristics of vehicles and people, thereby better meeting the needs of practical applications.

[0111] Through steps S501 to S504, not only can the model's processing capabilities for specific tasks be improved, but the model can also be better adapted to complex scenarios in actual applications. In the example of vehicle and personnel verification, this means that the model can more accurately identify the characteristics of vehicles and personnel and more effectively support the verification work of insurance scenarios. This continuous optimization and adjustment is the key to ensuring that the model remains efficient and accurate in actual deployment.

[0112] See also Figure 6 In some embodiments, step S103 includes but is not limited to steps S601 to S604:

[0113] Step S601, binarize each business verification data to obtain corresponding binary verification data.

[0114] Step S602: input each business verification data into a preset linear regression model of the corresponding binary verification data to obtain linear regression data.

[0115] Step S603, when the linear regression data meets the preset verification conditions, it is determined that the target insurance scene image has passed the verification, and scene verification data indicating that the verification has passed is obtained.

[0116] Step S604, when the linear regression data does not meet the preset verification conditions, it is determined that the target insurance scene image has not passed the verification, and scene verification data indicating that the verification has not passed is obtained.

[0117] In step S601 of some embodiments, the business verification data can be converted into binary form, that is, each data point is either 0 (indicating non-compliance) or 1 (indicating compliance). This processing method simplifies the data and makes it easier to be processed by subsequent models. For example, in vehicle verification, if the degree of damage to a vehicle exceeds the insurance claim standard, the relevant verification data will be binarized to 0; if the vehicle is in good condition, it will be binarized to 1.

[0118] In step S602 of some embodiments, the binarized verification data is input into a preset linear regression model. The linear regression model is used to predict and analyze the relationship between data, and is used here to evaluate the contribution of each binarized data point to the final verification result. For example, if multiple attributes of the vehicle are verified, such as the degree of damage, model matching, etc., the linear regression model will analyze the binarized data of these attributes to predict whether the vehicle as a whole meets the insurance claim standards.

[0119] In step S603 of some embodiments, if the linear regression data meets the preset verification conditions, for example, the overall conformity of the model prediction exceeds the set threshold, then the target insurance scene image is determined to have passed the verification, and scene verification data indicating that the verification has passed is generated. This step is the decision point of the verification process, ensuring that only images that meet all business standards are deemed qualified.

[0120] In step S604 of some embodiments, if the linear regression data does not meet the preset verification conditions, it will be determined that the target insurance scene image has not passed the verification, and scene verification data indicating that the verification has not passed will be generated. This may involve recording the specific reasons for non-compliance with the standards, such as the degree of damage to the vehicle exceeds the scope of the claim, or the personnel lack necessary safety equipment, etc. Such data is crucial for subsequent audits and improvement measures.

[0121] Through steps S601 to S604, the insurance scene verification method based on the large language model can systematically process and analyze insurance scene images, ensuring that only images that meet business standards can pass the verification, greatly improving the accuracy of the verification.

[0122] See also Figure 7In some embodiments, before step S101, the following steps may also be performed, but are not limited to, steps S701 to S703:

[0123] Step S701, obtaining the image size and image clarity of the target insurance scene image.

[0124] Step S702: In response to the image size not satisfying a preset size condition, the target insurance scene image is resized according to a preset target size.

[0125] Step S703, in response to the clarity of the target insurance scene image being lower than a preset clarity threshold, super-resolution reconstruction is performed on the target insurance scene image using a preset super-resolution model.

[0126] In step S701 of some embodiments, the size and clarity of the image directly affect the effect of subsequent model processing. For example, in the scene of vehicle and personnel verification, if the image size is too small or the clarity is insufficient, the model may not be able to accurately identify the damage degree of the vehicle or the safety equipment of the personnel.

[0127] In step S702 of some embodiments, if it is detected that the image size does not meet the preset size condition, the image will be adjusted according to the preset target size. This step ensures that all images input to the model have a uniform size, which helps the model learn and recognize features in the image more effectively. For example, if the original image is too narrow or too wide, it may not contain all the vehicle or personnel features that need to be recognized. By adjusting the image size, it can be ensured that the image received by the model is in the most suitable format for analysis.

[0128] In step S703 of some embodiments, if the clarity of the image is lower than a preset clarity threshold, a preset super-resolution model is used to perform super-resolution reconstruction on the image. This step improves the model's ability to recognize subtle features in the image by enhancing the details and clarity of the image. For example, in vehicle verification, super-resolution technology can help the model more clearly identify the vehicle's license plate number or slight scratches, which may be difficult to recognize in the original low-resolution image.

[0129] Through these preprocessing steps, the target insurance scene image is optimized to the form that is most suitable for large language model processing. This not only improves the accuracy of verification, but also ensures that the model can extract as much useful information as possible from the image, thereby making more reliable verification decisions. In practical applications, these preprocessing steps are particularly important for processing images from different sources and taken under different conditions. In this way, insurance companies can more effectively use large language models for automated insurance scene verification, improve work efficiency, and reduce verification error rates.

[0130] See also Figure 8 The embodiment of the present application also provides a scenario verification device based on a large language model, which can implement the above-mentioned insurance scenario verification method based on a large language model, and the device includes:

[0131] An acquisition module is used to acquire a target insurance scenario image and a pre-trained large language model; wherein the large language model includes multiple large language sub-models, and the multiple large language sub-models of the large language model include business association sub-models with business mapping constraints between each other, and each large language sub-model corresponds to a business capability.

[0132] The business verification module is used to input the target insurance scene image into the large language model so that the large language model can classify the target insurance scene image based on the business mapping constraints to obtain business type information, and input the target insurance scene image into the corresponding business-related sub-model based on the business type information for business verification to obtain business verification data.

[0133] The scene verification module is used to verify the target insurance scene image based on the business verification data to obtain the scene verification data corresponding to the target insurance scene image.

[0134] The specific implementation of the scenario verification device based on a large language model is basically the same as the specific implementation example of the insurance scenario verification method based on a large language model mentioned above, and will not be repeated here.

[0135] The embodiment of the present application also provides an electronic device, the electronic device includes a memory and a processor, the memory stores a computer program, and the processor implements the above-mentioned insurance scenario verification method based on a large language model when executing the computer program. The electronic device can be any smart terminal including a tablet computer, a car computer, etc.

[0136] See also Fig. 9 , Fig. 9 The hardware structure of an electronic device of another embodiment is illustrated, and the electronic device includes:

[0137] The processor 901 may be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present application;

[0138] The memory 902 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 902 can store an operating system and other applications. When the technical solution provided in the embodiment of this specification is implemented by software or firmware, the relevant program code is stored in the memory 902, and the processor 901 calls and executes the insurance scenario verification method based on a large language model in the embodiment of this application;

[0139] Input / output interface 903, used to implement information input and output;

[0140] Communication interface 904, used to realize communication interaction between the device and other devices, which can be realized through wired mode (such as USB, network cable, etc.) or wireless mode (such as mobile network, WI FI, Bluetooth, etc.);

[0141] A bus 905 that transmits information between the various components of the device (e.g., the processor 901, the memory 902, the input / output interface 903, and the communication interface 904);

[0142] The processor 901 , the memory 902 , the input / output interface 903 and the communication interface 904 are connected to each other in communication within the device via a bus 905 .

[0143] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program, and when the computer program is executed by a processor, it implements the above-mentioned insurance scenario verification method based on a large language model.

[0144] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely disposed relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0145] Insurance scenario verification method based on large language model proposed in this application 、A device, an electronic device and a storage medium, which obtain a target insurance scene image and a pre-trained large language model; wherein the large language model includes multiple large language sub-models, and the multiple large language sub-models of the large language model include business association sub-models with business mapping constraints between each other, and each large language sub-model corresponds to a business capability; the target insurance scene image is input into the large language model so that the large language model classifies the target insurance scene image based on the business mapping constraints to obtain business type information, and based on the business type information, the target insurance scene image is input into the corresponding business association sub-model for business verification to obtain business verification data; the target insurance scene image is verified based on the business verification data to obtain scene verification data corresponding to the target insurance scene image. The present application utilizes a pre-trained large language model for insurance scene verification. Compared with traditional rules or single model methods, the large language model can better understand the scene semantics and extract key information, thereby improving verification accuracy. At the same time, insurance scenarios are complex and diverse, and a single verification method is difficult to meet the needs of all scenarios, and the cascade engineering of traditional models is very complicated. Therefore, this application divides the large language model into multiple sub-models, each sub-model corresponds to a business capability, and establishes associations between sub-models through business mapping constraints, so that the verification process can be parallelized and can flexibly adapt to complex business scenarios. Therefore, this application makes full use of the semantic understanding and multi-tasking capabilities of the large language model, and through the division of labor and cooperation of sub-models, it realizes efficient scenario verification in complex insurance scenarios.

[0146] The embodiments described in the embodiments of the present application are intended to more clearly illustrate the technical solutions of the embodiments of the present application and do not constitute a limitation on the technical solutions provided in the embodiments of the present application. Those skilled in the art will appreciate that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0147] Those skilled in the art will appreciate that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than shown in the figures, or a combination of certain steps, or different steps.

[0148] The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0149] Those skilled in the art will appreciate that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices may be implemented as software, firmware, hardware, or a suitable combination thereof.

[0150] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0151] It should be understood that in the present application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0152] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the above units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0153] The units described above as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0154] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0155] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including multiple instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, referred to as ROM), random access memory (Random Access Memory, referred to as RAM), disk or optical disk and other media that can store programs.

[0156] The preferred embodiments of the present invention are described above with reference to the accompanying drawings, but the scope of the rights of the present invention is not limited thereto. Any modification, equivalent substitution and improvement made by a person skilled in the art without departing from the scope and essence of the present invention should be within the scope of the rights of the present invention.

Claims

1. An insurance scenario verification method based on a large language model, characterized in that: The method comprises: Acquire a target insurance scenario image and a pre-trained large language model; wherein the large language model includes a plurality of large language sub-models, the plurality of large language sub-models of the large language model include business association sub-models with business mapping constraints between each other, and each of the large language sub-models corresponds to a business capability; Inputting the target insurance scene image into the large language model, so that the large language model performs business classification on the target insurance scene image based on the business mapping constraint to obtain business type information, and inputting the target insurance scene image into the corresponding business association sub-model for business verification based on the business type information to obtain business verification data; The target insurance scene image is verified based on the business verification data to obtain scene verification data corresponding to the target insurance scene image.

2. The insurance scenario verification method based on a large language model according to claim 1 is characterized in that: Before obtaining the target scene image and the pre-trained large language model, a pre-training process for the large language model is also included, specifically including: Acquire multiple original models and a first training data set; wherein the training scene images in the first training data set and the target insurance scene images belong to the same target business scene; Performing classification and annotation processing on the first training data set according to the target business scenario to obtain a classified and annotated data set; Based on the classified and labeled data set, the multiple original models are trained to obtain multiple pre-trained sub-models; For each of the pre-trained sub-models, a corresponding prompt word is generated, and the prompt word is adapted to the corresponding pre-trained sub-model to obtain a plurality of pre-trained sub-models adapted to the prompt word; The pre-trained sub-models adapted to the multiple prompt words are cascaded based on the business mapping constraints to obtain the pre-trained large language model.

3. The insurance scenario verification method based on a large language model according to claim 2 is characterized in that: The multiple original models are trained based on the classified and labeled data set to obtain multiple pre-trained sub-models, including: Based on the classified and labeled data set, a first original model is trained to obtain a first pre-trained sub-model; wherein the first original model is used to determine whether there is an object to be inspected in the scene image and whether it meets the verification requirements; Based on the classified and labeled data set, a second original model is trained to obtain a second pre-trained sub-model; wherein the second original model is used to determine whether the inspection object in the scene image meets the preset verification standard; According to the target business scenario, a business mapping constraint relationship is established between the first pre-trained sub-model and the second pre-trained sub-model.

4. The insurance scenario verification method based on a large language model according to claim 2 is characterized in that: The pre-trained sub-models adapted to the multiple prompt words are cascaded based on the business mapping constraints to obtain the pre-trained large language model, including: According to the business mapping constraint, determining the cascade order between the pre-trained sub-models adapted by the multiple prompt words, and constructing a cascade structure of the pre-trained sub-models; The input and output of the pre-trained sub-models are adapted based on the business mapping constraints so that the input and output formats of adjacent pre-trained sub-models match.

5. The method according to claim 2, characterized in that: After the pre-trained sub-models adapted to the multiple prompt words are cascaded based on the business mapping constraint to obtain the pre-trained large language model, the method further includes: Obtain a second training data set; Inputting the second training data set into the large language model to obtain training prediction data; Compare the predicted labels in the training prediction data with the real labels in the second training data set to obtain predicted labeled data; The large language model is fine-tuned based on the predicted and labeled data to obtain a fine-tuned large language model.

6. The insurance scenario verification method based on a large language model according to claim 1 is characterized in that: The verifying the target insurance scene image based on the business verification data to obtain scene verification data corresponding to the target insurance scene image includes: Binarizing each of the business verification data to obtain corresponding binary verification data; Input each of the business verification data into a preset linear regression model of the corresponding binary verification data to obtain linear regression data; When the linear regression data satisfies a preset verification condition, it is determined that the target insurance scene image has passed the verification, and scene verification data indicating that the verification has passed is obtained; When the linear regression data does not satisfy a preset verification condition, it is determined that the target insurance scene image has not passed the verification, and scene verification data indicating that the verification has not passed is obtained.

7. The insurance scenario verification method based on a large language model according to claim 1 is characterized in that: Before inputting the target insurance scene image into the large language model, the method further includes: Acquire the image size and image clarity of the target insurance scene image; In response to the image size not satisfying a preset size condition, resizing the target insurance scene image according to a preset target size; wherein the preset target size is an input size of the large language model; In response to the clarity of the target insurance scene image being lower than a preset clarity threshold, a preset super-resolution model is used to perform super-resolution reconstruction on the target insurance scene image.

8. A scene verification device based on a large language model, characterized in that: The device comprises: An acquisition module, used to acquire a target insurance scenario image and a pre-trained large language model; wherein the large language model includes a plurality of large language sub-models, wherein the plurality of large language sub-models of the large language model include business association sub-models with business mapping constraints between each other, and each of the large language sub-models corresponds to a business capability; A business verification module, used for inputting the target insurance scene image into the large language model, so that the large language model performs business classification on the target insurance scene image based on the business mapping constraint to obtain business type information, and inputting the target insurance scene image into the corresponding business association sub-model for business verification based on the business type information to obtain business verification data; A scene verification module is used to verify the target insurance scene image based on the business verification data to obtain scene verification data corresponding to the target insurance scene image.

9. An electronic device, characterized in that: The electronic device includes a memory and a processor, the memory stores a computer program, and the processor implements the insurance scenario verification method based on a large language model as described in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the insurance scenario verification method based on a large language model as described in any one of claims 1 to 7 is implemented.

Images