Claim settlement prediction method, claim settlement prediction device, electronic device, and storage medium

Abstract

The application provides a claim settlement prediction method, a claim settlement prediction device, an electronic device and a storage medium, and belongs to the technical field of artificial intelligence. The method comprises the following steps: obtaining first case features of historical claim settlement cases and second case features of new claim settlement cases; obtaining preliminary case features according to the first case features and the second case features; inputting the preliminary case features into a neural network model for embedding processing and semantic extraction to obtain historical case representation vectors; predicting labels of the historical claim settlement cases based on the historical case representation vectors to obtain first claim settlement prediction labels; training the neural network model based on the first claim settlement prediction labels to obtain a preliminary claim settlement prediction model; optimizing the preliminary claim settlement prediction model based on the second case features to obtain a target claim settlement prediction model; and predicting claim settlement of target case target claim settlement data based on the target claim settlement prediction model to obtain a predicted claim settlement category of the target case, thereby improving the accuracy of claim settlement judgment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, in particular to a claim settlement prediction method, a claim settlement prediction device, an electronic device and a storage medium. BACKGROUND

[0002] At present, in the claim settlement business, it is often necessary to rely on a neural network model to judge the claim settlement situation of a claim settlement case. Model training often requires a large amount of high-quality sample data, and the collection of high-quality sample data is often difficult, which affects the training effect of the model and can cause the model to have low accuracy in judging the claim settlement situation. Therefore, how to improve the accuracy of the model in judging the claim settlement situation has become a technical problem to be solved. SUMMARY

[0003] The main purpose of the embodiments of the present application is to propose a claim settlement prediction method, a claim settlement prediction device, an electronic device and a storage medium, which aims to improve the accuracy of the model in judging the claim settlement situation.

[0004] To achieve the above-mentioned purpose, a first aspect of the embodiments of the present application proposes a claim settlement prediction method, which comprises:

[0005] obtaining first claim settlement data and second claim settlement data, wherein the first claim settlement data comprises first case features of historical claim settlement cases, and the second claim settlement data comprises second case features of new claim settlement cases;

[0006] obtaining preliminary case features according to the first case features and the second case features;

[0007] inputting the preliminary case features into a preset neural network model, wherein the neural network model comprises an embedding layer, an encoder and a prediction layer;

[0008] performing embedding processing on the preliminary case features based on the embedding layer to obtain a preliminary case feature embedding vector;

[0009] performing semantic extraction on the preliminary case feature embedding vector based on the encoder to obtain a historical case representation vector;

[0010] performing label prediction on the historical claim settlement cases based on the prediction layer and the historical case representation vector to obtain a first claim settlement prediction label of the historical claim settlement case, wherein the first claim settlement prediction label is used to represent a predicted claim settlement category of the historical claim settlement case;

[0011] performing model training on the neural network model based on the first claim settlement prediction label to obtain a preliminary claim settlement prediction model;

[0012] perform parameter optimization on the preliminary claim prediction model based on the second case feature, to obtain a target claim prediction model;

[0013] obtain target claim data of a target case;

[0014] input the target claim data into the target claim prediction model to perform claim prediction, to obtain a predicted claim category of the target case, wherein the predicted claim category is used to indicate that the target case is a normal claim case or to indicate that the target case is an abnormal claim case.

[0015] In some embodiments, the obtaining of the preliminary case feature according to the first case feature and the second case feature comprises:

[0016] identifying a feature state of the first case feature;

[0017] if the feature state is a normal state, determining the first case feature as the preliminary case feature; wherein the normal state is used to indicate that the first case feature does not exist feature error;

[0018] if the feature state is an abnormal state, determining the second case feature as the preliminary case feature; wherein the abnormal state is used to indicate that the first case feature exists feature error.

[0019] In some embodiments, the preliminary case feature embedding vector comprises a discrete feature embedding vector and a discrete feature value embedding vector, and the embedding processing of the preliminary case feature by the embedding layer to obtain the preliminary case feature embedding vector comprises:

[0020] performing feature extraction on the preliminary case feature to obtain a discrete case feature;

[0021] performing embedding processing on the discrete case feature by the embedding layer to obtain a discrete feature embedding vector;

[0022] performing feature value query on the discrete feature embedding vector based on a preset first correspondence table, to obtain a discrete feature value embedding vector corresponding to the discrete feature embedding vector;

[0023] obtaining the preliminary case feature embedding vector based on the discrete feature value embedding vector and the discrete feature embedding vector.

[0024] In some embodiments, the preliminary case feature embedding vector comprises a numerical feature embedding vector and a numerical feature value embedding vector, and the embedding processing of the preliminary case feature by the embedding layer to obtain the preliminary case feature embedding vector comprises:

[0025] performing feature extraction on the preliminary case feature to obtain a numerical case feature;

[0026] embedding processing on the numerical case feature based on the embedding layer to obtain a numerical feature embedding vector;

[0027] feature bucketing on the numerical case feature to obtain a plurality of case feature buckets;

[0028] discretization processing on the case feature buckets to obtain discrete feature buckets;

[0029] value feature querying on the discrete feature buckets based on a preset second correspondence table to obtain a numerical feature value embedding vector corresponding to the numerical feature embedding vector;

[0030] obtaining the preliminary case feature embedding vector based on the numerical feature embedding vector and the numerical feature value embedding vector.

[0031] In some embodiments, the encoder includes a transformer layer and a pooling layer, and the semantic extraction on the preliminary case feature embedding vector based on the encoder to obtain a historical case representation vector includes:

[0032] semantic encoding on the preliminary case feature embedding vector based on the transformer layer to obtain a case semantic encoding feature;

[0033] average pooling processing on the case semantic encoding feature based on the pooling layer to obtain the historical case representation vector.

[0034] In some embodiments, the first claim prediction label includes a first label and a second label, and the label prediction on the historical claim case based on the prediction layer and the historical case representation vector to obtain the first claim prediction label of the historical claim case includes:

[0035] claim scoring on the historical case representation vector based on a preset function to obtain a case claim score;

[0036] if the case claim score is greater than a preset threshold, determining the first claim prediction label of the historical claim case as the first label, wherein the first label is used to indicate that the historical claim case is an abnormal payment case;

[0037] if the case claim score is less than or equal to the preset threshold, determining the first claim prediction label of the historical claim case as the second label, wherein the second label is used to indicate that the historical claim case is a normal payment case.

[0038] In some embodiments, the second claim data further comprises a true claim label of the new claim case, the parameter optimization of the preliminary claim prediction model based on the second case feature obtains a target claim prediction model, comprising:

[0039] The embedding layer is used for embedding processing on the second case feature to obtain an intermediate case feature embedding vector;

[0040] The encoder is used for semantic extraction on the intermediate case feature embedding vector to obtain a new case representation vector;

[0041] The prediction layer and the new case representation vector are used for label prediction on the new claim case to obtain a second claim prediction label of the new claim case, and the second claim prediction label is used for representing a predicted claim category of the new claim case;

[0042] The second claim prediction label and the true claim label are used for loss calculation to obtain a claim loss value;

[0043] The preliminary claim prediction model is parameter-optimized based on the claim loss value to obtain the target claim prediction model.

[0044] To achieve the above object, a second aspect of the embodiment of the present application provides a claim prediction device, which comprises:

[0045] A training data acquisition module is configured to acquire first claim data and second claim data, wherein the first claim data comprises first case features of historical claim cases, and the second claim data comprises second case features of new claim cases;

[0046] A feature acquisition module is configured to obtain preliminary case features according to the first case features and the second case features;

[0047] An input module is configured to input the preliminary case features into a preset neural network model, wherein the neural network model comprises an embedding layer, an encoder and a prediction layer;

[0048] An embedding module is configured to perform embedding processing on the preliminary case features based on the embedding layer to obtain a preliminary case feature embedding vector;

[0049] A semantic extraction module is configured to perform semantic extraction on the preliminary case feature embedding vector based on the encoder to obtain a historical case representation vector;

[0050] a label prediction module, configured to perform label prediction on the historical claim case based on the prediction layer and the historical case representation vector, to obtain a first claim prediction label of the historical claim case, the first claim prediction label being used to represent a predicted claim category of the historical claim case;

[0051] a model training module, configured to perform model training on the neural network model based on the first claim prediction label, to obtain a preliminary claim prediction model;

[0052] a parameter optimization module, configured to perform parameter optimization on the preliminary claim prediction model based on the second case feature, to obtain a target claim prediction model;

[0053] a target data acquisition module, configured to acquire target claim data of a target case;

[0054] a claim prediction module, configured to input the target claim data into the target claim prediction model to perform claim prediction, to obtain a predicted claim category of the target case, wherein the predicted claim category is used to represent that the target case is a normal claim case or represent that the target case is an abnormal claim case. To achieve the above object, a third aspect of embodiments of the present application provides an electronic device, which comprises a memory and a processor, the memory stores a computer program, and the processor implements the method of the first aspect when executing the computer program.

[0055] To achieve the above object, a fourth aspect of embodiments of the present application provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the method of the first aspect.

[0056] The claim settlement prediction method, the claim settlement prediction device, the electronic device, and the computer readable storage medium provided in the application obtain first claim settlement data and second claim settlement data, wherein the first claim settlement data comprises first case features of historical claim settlement cases, and the second claim settlement data comprises second case features of newly added claim settlement cases; preliminary case features are obtained according to the first case features and the second case features, and feature data used for model training can be conveniently obtained. Further, the preliminary case features are input into a preset neural network model, wherein the neural network model comprises an embedding layer, an encoder, and a prediction layer; the embedding layer is used to perform embedding processing on the preliminary case features to obtain a preliminary case feature embedding vector; the encoder is used to perform semantic extraction on the preliminary case feature embedding vector to obtain a historical case representation vector; the prediction layer and the historical case representation vector are used to perform label prediction on the historical claim settlement cases to obtain first claim settlement prediction labels of the historical claim settlement cases, and the first claim settlement prediction labels are used to represent predicted claim settlement categories of the historical claim settlement cases, so that case content information of the historical claim settlement cases can be conveniently extracted, and the claim settlement prediction labels of the historical claim settlement cases can be predicted. The neural network model is trained based on the first claim settlement prediction labels to obtain a preliminary claim settlement prediction model, and the preliminary claim settlement prediction model is parameter-optimized based on the second case features to obtain a target claim settlement prediction model. Finally, target claim settlement data of a target case is obtained; the target claim settlement data is input into the target claim settlement prediction model to perform claim settlement prediction, and a predicted claim settlement category of the target case is obtained, wherein the predicted claim settlement category is used to represent that the target case is a normal claim settlement case or that the target case is an abnormal claim settlement case. This method can train and optimize the model by combining historical data and newly added data, and is beneficial to improving the prediction accuracy of the model on the claim settlement situation of the claim settlement case. BRIEF DESCRIPTION OF DRAWINGS

[0057] Figure 1 is a flowchart of the claim settlement prediction method provided in the application;

[0058] Figure 2 is a flowchart of step S102 in Figure 1

[0059] is a flowchart of step S104 in Figure 3 Figure 1 is a flowchart of step S104 in

[0060] Figure 4 Figure 1 is another flowchart of step S104 in

[0061] Figure 5 is a flowchart of step S105 in Figure 1

[0062] is a flowchart of step S105 in Figure 6 Figure 1 ​​​the flowchart of step S106 in

[0063] Figure 7 is Figure 1 the flowchart of step S108 in

[0064] Figure 8 is a module schematic diagram of the claim settlement prediction device provided by the embodiment of the present application;

[0065] Figure 9 is a hardware structure schematic diagram of the electronic device provided by the embodiment of the present application. DETAILED DESCRIPTION

[0066] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application is further described in detail below in combination with the 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.

[0067] 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 can be executed in a manner different from the module division in the device or the order in the flowchart. The terms "first", "second", and the like in the specification and claims and the above-described drawings are used to distinguish similar objects, and do not necessarily describe a specific order or sequence.

[0068] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs. The terms used herein are only for the purpose of describing the embodiments of the present application, and are not intended to limit the present application.

[0069] First, the terms involved in the present application are analyzed:

[0070] Artificial intelligence (AI): is a new technical science of researching and developing a theory, method, technology and application system for simulating, extending and expanding human intelligence; artificial intelligence is a branch of computer science, and artificial intelligence attempts to understand the essence of intelligence and produce a new intelligent machine that can react in a similar way to human intelligence. The research in this field includes robots, language recognition, image recognition, natural language processing and expert systems. Artificial intelligence can simulate the information process of human consciousness and thinking. Artificial intelligence is also the theory, method, technology and application system of using digital computers or digital computer controlled machines to simulate, extend and expand human intelligence, to perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0071] Natural language processing (NLP): NLP uses computers to process, understand and use human languages (such as Chinese, English, etc.), and NLP is a branch of artificial intelligence, which is an interdisciplinary subject of computer science and linguistics, and is also commonly referred to as computational linguistics. Natural language processing includes syntax analysis, semantic analysis, discourse understanding, etc. Natural language processing is commonly used in machine translation, handwritten and printed character recognition, speech recognition and text-to-speech conversion, information intent recognition, information extraction and filtering, text classification and clustering, public opinion analysis and opinion mining, etc. It involves data mining, machine learning, knowledge acquisition, knowledge engineering, artificial intelligence research related to language processing, and linguistic research related to language computing.

[0072] Information extraction (Information Extraction, NER): A text processing technology that extracts specified types of entities, relationships, events, etc. from natural language text, and forms structured data output. Information extraction is a technology for extracting specific information from text data. Text data is composed of specific units, such as sentences, paragraphs, and chapters, and text information is composed of small specific units, such as words, phrases, sentences, paragraphs, or combinations of these specific units. Extracting noun phrases, names, and places from text data is text information extraction, and of course, the information extracted by the text information extraction technology can be various types of information.

[0073] At present, in the claim settlement business, it is often necessary to rely on a neural network model to judge the claim settlement situation of a claim settlement case, and model training often requires a large amount of high-quality sample data, and the collection of high-quality sample data is often difficult, which affects the training effect of the model and can cause the model to have low accuracy in judging the claim settlement situation. Therefore, how to improve the accuracy of the model in judging the claim settlement situation has become a technical problem to be solved.

[0074] Based on this, the embodiments of the present application provide a claim settlement prediction method, a claim settlement prediction device, an electronic device and a storage medium, which aim to improve the accuracy of the model in judging the claim settlement situation.

[0075] The claim settlement prediction method, the claim settlement prediction device, the electronic device and the storage medium provided by the embodiments of the present application are specifically explained by the following embodiments. First, the claim settlement prediction method in the embodiments of the present application is described.

[0076] The embodiments of the present application can acquire and process related data based on artificial intelligence technology. The artificial intelligence (AI) is a theory, method, technology and application system for simulating, extending and expanding human intelligence by using a digital computer or a machine controlled by a digital computer, perceiving an environment, acquiring knowledge and using the knowledge to obtain optimal results.

[0077] The artificial intelligence basic technology generally includes technologies such as sensors, special artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction system, mechatronics, etc. The artificial intelligence software technology mainly includes computer vision technology, robot technology, biometric technology, speech processing technology, natural language processing technology, and machine learning / deep learning, etc.

[0078] The claim settlement prediction method provided by the embodiments of the present application relates to the field of artificial intelligence. The claim settlement prediction method provided by the embodiments of the present application can be applied in a terminal, can also be applied in a server end, and can also be software running in a terminal or a server end. In some embodiments, the terminal can be a smart phone, a tablet computer, a notebook computer, a desktop computer, etc.; the server end can be configured as an independent physical server, can also be configured as a server cluster or a distributed system composed of multiple physical servers, can also be configured as a cloud server providing basic cloud computing services such as cloud service, cloud database, cloud computing, cloud function, cloud storage, network service, cloud communication, middleware service, domain name service, security service, CDN, and big data and artificial intelligence platform; and the software can be an application for implementing the claim settlement prediction method, etc., but is not limited to the above forms.

[0079] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld devices or portable devices, tablet devices, multi-processor 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 a distributed computing environment in which tasks are performed by remote processing devices connected by a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media, including storage devices.

[0080] It should be noted that in various specific embodiments of the present application, when relevant processing needs to be performed on data related to the identity or characteristics of a user, such as user information, user behavior data, user case information data and case description data, user history data, and user location information, the user's permission or consent will be obtained first, and the collection, use and processing of such data will comply with relevant laws, regulations and standards of the country or region. In addition, when the embodiments of the present application need to obtain sensitive personal information of the user, the separate permission or separate consent of the user will be obtained through a pop-up window or a jump to a confirmation page, and after obtaining the separate permission or separate consent of the user, the necessary user-related data for the normal operation of the embodiments of the present application will be obtained.

[0081] Figure 1 is an optional flowchart of the claim prediction method provided by the embodiments of the present application, Figure 1 The method in the method can include but is not limited to steps S101-S110.

[0082] Step S101, obtaining first claim data and second claim data, wherein the first claim data includes first case features of historical claim cases, and the second claim data includes second case features of new claim cases;

[0083] Step S102, obtaining preliminary case features according to the first case features and the second case features;

[0084] Step S103, inputting the preliminary case features into a preset neural network model, wherein the neural network model includes an embedding layer, an encoder, and a prediction layer;

[0085] Step S104, performing embedding processing on the preliminary case features based on the embedding layer to obtain a preliminary case feature embedding vector;

[0086] Step S105, performing semantic extraction on the preliminary case feature embedding vector based on the encoder to obtain a historical case representation vector;

[0087] Step S106, performing label prediction on the historical claim cases based on the prediction layer and the historical case representation vector to obtain a first claim prediction label of the historical claim cases, the first claim prediction label being used to represent a predicted claim category of the historical claim cases;

[0088] Step S107, performing model training on the neural network model based on the first claim prediction label to obtain a preliminary claim prediction model;

[0089] Step S108, performing parameter optimization on the preliminary claim prediction model based on the second case features to obtain a target claim prediction model;

[0090] Step S109, obtaining target claim data of the target case;

[0091] Step S110, inputting the target claim data into the target claim prediction model for claim prediction to obtain a predicted claim category of the target case, wherein the predicted claim category is used to represent that the target case is a normal compensation case or that the target case is an abnormal compensation case.

[0092] The steps S101 to S110 shown in the embodiments of the present application, by obtaining the first claim data and the second claim data, wherein the first claim data includes the first case features of the historical claim cases, and the second claim data includes the second case features of the new claim cases; according to the first case features and the second case features, the preliminary case features are obtained, which can more conveniently obtain the feature data for model training. Further, the preliminary case features are input into a preset neural network model, wherein the neural network model includes an embedding layer, an encoder and a prediction layer; the embedding layer is used to perform embedding processing on the preliminary case features to obtain a preliminary case feature embedding vector; the encoder is used to perform semantic extraction on the preliminary case feature embedding vector to obtain a historical case representation vector; the prediction layer and the historical case representation vector are used to perform label prediction on the historical claim cases to obtain a first claim prediction label of the historical claim cases, and the first claim prediction label is used to represent a predicted claim category of the historical claim cases, which can more conveniently extract the case content information of the historical claim cases and predict the claim prediction label of the historical claim cases. The neural network model is trained based on the first claim prediction label to obtain a preliminary claim prediction model, and the preliminary claim prediction model is parameter-optimized based on the second case features to obtain a target claim prediction model. Finally, the target claim data of the target case is obtained; the target claim data is input into the target claim prediction model for claim prediction to obtain a predicted claim category of the target case, wherein the predicted claim category is used to represent that the target case is a normal compensation case or that the target case is an abnormal compensation case. This way can combine the historical data and the new data to train and optimize the model, which is beneficial to improve the prediction accuracy of the model on the claim situation of the claim case.

[0093] In step S101 of some embodiments, the first claim data and the second claim data can be obtained by writing a web crawler to crawl data from the data source, or by other means. The first claim data includes first case features of historical claim cases and real claim labels of the historical claim cases, where the real claim labels are used to indicate that the historical claim cases are normal payment cases or non-normal payment cases. The second claim data includes second case features of new claim cases and real claim labels of the new claim cases, where the real claim labels are used to indicate that the new claim cases are normal payment cases or non-normal payment cases.

[0094] For example, the first claim data includes diagnosis disease codes of historical claim cases, insurance type codes, date differences between accident times and insurance effective dates, insurance amounts, disease names, disease symptoms, and the like. The second claim data includes diagnosis disease codes of new claim cases, insurance type codes, date differences between accident times and insurance effective dates, insurance amounts, disease names, disease symptoms, and the like.

[0095] Further, the first case features and the second case features can be represented as table data, and the first case features and the second case features include two types of features, i.e., discrete features and numerical features. For example, the first case features include disease names, disease alert lines, claim amounts, and treatment properties, where the disease names, the disease alert lines, and the treatment properties are discrete features, and the claim amounts are numerical features.

[0096] Please refer to Figure 2 In some embodiments, step S102 can include, but is not limited to, steps S201 to S203.

[0097] Step S201, identifying a feature state of the first case features;

[0098] Step S202, if the feature state is a normal state, determining the first case features as preliminary case features; where the normal state is used to indicate that the first case features do not have feature errors.

[0099] Step S203, if the feature state is an abnormal state, determining the second case features as the preliminary case features; where the abnormal state is used to indicate that the first case features have feature errors.

[0100] In step S201 of some embodiments, in order to improve the quality of model training, the sample data used for training the model can be optimized in advance, that is, first, the feature state of the first case feature is identified, it is judged whether the first case feature has feature errors and the like, and the feature state of the first case feature is divided into normal state and abnormal state according to whether the first case feature has feature errors, wherein the normal state is used to indicate that the first case feature does not have feature errors, and the abnormal state is used to indicate that the first case feature has feature errors.

[0101] When identifying whether the feature state of the first case feature is normal or abnormal, the feature state of the first case feature can be judged based on the preset total amount of features and feature length and the like. For example, the preset total amount of features is 4, and the first case feature of the historical claim case should include five types of features such as disease name, disease alert line, claim amount, and treatment nature. Then, it is judged whether the first case feature includes these five specific features. If the first case feature includes the disease name, the disease alert line, and the claim amount, the first case feature has feature errors, and the feature state is abnormal.

[0102] In step S202 of some embodiments, if the feature state is normal, it indicates that the first case feature does not have feature errors, and the first case feature can be directly determined as the preliminary case feature.

[0103] In step S203 of some embodiments, if the feature state is abnormal, it indicates that the first case feature has feature errors, and the second case feature is determined as the preliminary case feature. For example, after determining the abnormal part in the first case feature, the corresponding feature part in the second case feature is extracted, and the extracted feature part replaces the abnormal part in the first case feature.

[0104] For example, the first case feature includes the disease name of diabetes, the disease alert line of yellow, and the claim amount of 3 million, and the treatment nature is missing in the description of the first case feature. When the disease name of diabetes is extracted from the second case feature, the treatment nature is "accident", and the treatment nature in the first case feature is supplemented and described as "accident".

[0105] It should be noted that the preliminary case feature is used for subsequent model training, that is, the initial case feature can be understood as sample data used for training the model.

[0106] Through the above steps S201 to S203, the first case feature can be conveniently optimized, which is beneficial to improve the data quality of the feature data used for model training, and can make the trained model have better claim prediction ability.

[0107] In step S103 of some embodiments, the preliminary case features can be input into a preset neural network model according to a preset code program, wherein the neural network model comprises an embedding layer, an encoder and a prediction layer, the embedding layer is mainly used for converting the input feature data from a text form into a vector form; the encoder is mainly used for feature coding of the feature data in the vector form to extract semantic content information of the feature data and obtain case feature data; and the prediction layer is mainly used for predicting the case according to the case feature data to determine whether the input feature data corresponds to a normal compensation case.

[0108] Referring to Figure 3 In some embodiments, the preliminary case feature embedding vector comprises a discrete feature embedding vector and a discrete feature value embedding vector, and step S104 can comprise but is not limited to steps S301 to S304:

[0109] In step S301, the preliminary case features are extracted to obtain discrete case features.

[0110] In step S302, the discrete case features are embedded based on the embedding layer to obtain a discrete feature embedding vector.

[0111] In step S303, the discrete feature embedding vector is queried for a feature value based on a preset first comparison table to obtain a discrete feature value embedding vector corresponding to the discrete feature embedding vector.

[0112] In step S304, the preliminary case feature embedding vector is obtained based on the discrete feature value embedding vector and the discrete feature embedding vector.

[0113] In step S301 of some embodiments, since the preliminary case features comprise discrete features and numerical features, the preliminary case features can be first extracted to obtain specific content information of the preliminary case features, and based on the obtained specific content information, it is determined whether the preliminary case features belong to discrete features or numerical features, so that the preliminary case features of the discrete feature type are extracted to obtain discrete case features.

[0114] In step S302 of some embodiments, the discrete case features are mapped from a data space to a vector space based on the embedding layer to obtain a discrete feature embedding vector corresponding to the discrete case features, and the discrete feature embedding vector is used to represent the feature type of the discrete case features.

[0115] The discrete feature embedding vector can be represented as:

[0116] wherein, represents the discrete feature embedding vector of the i-th historical claim case.

[0117] In step S303 of some embodiments, since corresponding feature values are preset for different types of feature data, based on the feature data and the feature values, a first comparison table can be constructed in advance, through which the specific feature type of the discrete feature embedding vector representation is queried, and the feature value of this feature type is represented in the form of a vector, and the discrete feature value embedding vector corresponding to the discrete feature embedding vector is obtained.

[0118] The discrete feature value embedding vector can be represented as:

[0119] Wherein, represents the discrete feature value embedding vector of the i-th historical claim case.

[0120] In step S304 of some embodiments, after obtaining the preliminary case feature embedding vector based on the discrete feature value embedding vector and the discrete feature embedding vector, the discrete feature value embedding vector and the discrete feature embedding vector are spliced to obtain the preliminary case feature embedding vector.

[0121] The preliminary case feature embedding vector can be represented as: E=[e0,e1,…,e i ],

[0122] Through the above steps S301 to S304, the first case feature can be embedded and processed more conveniently, the key discrete feature information in the first case feature can be extracted, so that the discrete case feature information can be used for model training in subsequent model training, thereby improving the judgment accuracy of the model on the claim settlement situation.

[0123] Please refer to Figure 4 In some embodiments, the preliminary case feature embedding vector includes a numerical feature embedding vector and a numerical feature value embedding vector, and step S104 can further include but is not limited to steps S401 to S406:

[0124] Step S401, feature extraction is performed on the preliminary case feature to obtain a numerical case feature;

[0125] Step S402, embedding processing is performed on the numerical case feature based on an embedding layer to obtain a numerical feature embedding vector;

[0126] Step S403, feature bucketing is performed on the numerical case feature to obtain a plurality of case feature buckets;

[0127] Step S404, the case feature buckets are discretized to obtain discrete feature buckets;

[0128] Step S405, value feature query is carried out on the discrete feature bucket based on the preset second correspondence table, and a numerical feature value embedding vector corresponding to the numerical feature embedding vector is obtained.

[0129] Step S406, based on the numerical feature embedding vector and the numerical feature value embedding vector, a preliminary case feature embedding vector is obtained.

[0130] In step S401 of some embodiments, since the preliminary case features include both discrete features and numerical features, the specific content information of the preliminary case features can be obtained by feature extraction, and based on the obtained specific content information, it is determined whether the preliminary case feature belongs to a discrete feature or a numerical feature, so that the preliminary case feature with a numerical feature type is extracted, and a numerical case feature is obtained.

[0131] In step S402 of some embodiments, the data mapping is carried out on the numerical case feature based on the embedding layer, the numerical case feature is mapped from the data space to the vector space, and a numerical feature embedding vector corresponding to the numerical case feature is obtained, which is used to represent the feature type of the numerical case feature.

[0132] In step S403 of some embodiments, when the numerical case feature is feature-bucketed, the feature-bucketing can be carried out in the manner of equidistant bucketing, equal frequency bucketing or model bucketing, and a plurality of case feature buckets are obtained. Taking equidistant bucketing as an example, a numerical segment is divided into a plurality of numerical intervals, and the numerical case features belonging to the same numerical interval are divided into a case feature bucket, and the length of each case feature bucket is the same. For example, the divided case feature buckets include [0, 99], [100, 199], [200, 299] and the like.

[0133] In step S404 of some embodiments, when the case feature bucket is discretized, the feature discretization can be carried out in the manner of one-hot encoding and the like, and the case feature bucket is converted from a continuous feature set to a plurality of discrete feature sets, that is, after the discretization processing such as one-hot encoding, a discrete feature bucket corresponding to the case feature bucket is obtained.

[0134] In step S405 of some embodiments, since the feature values corresponding to different types of feature data are preset, the second correspondence table can be constructed in advance based on the feature data and the feature values. Through the second correspondence table, the feature values of the discrete feature bucket are queried, the feature values are represented in the form of a vector, and a numerical feature value embedding vector corresponding to the numerical feature embedding vector is obtained.

[0135] In step S406 of some embodiments, when the preliminary case feature embedding vector is obtained based on the numerical feature embedding vector and the numerical feature value embedding vector, the numerical feature embedding vector and the numerical feature value embedding vector are spliced to obtain the preliminary case feature embedding vector.

[0136] Through the above steps S401 to S406, the first case feature can be embedded and processed more conveniently, the key numerical feature information in the first case feature is extracted, so that the numerical case feature information can be used for model training in subsequent model training, thereby improving the judgment accuracy of the model on the claim settlement situation.

[0137] Please refer to Figure 5 In some embodiments, the encoder includes a transformer layer and a pooling layer, and step S105 can include but is not limited to steps S501 to S502:

[0138] In step S501, the transformer layer is used to perform semantic encoding on the preliminary case feature embedding vector to obtain a case semantic encoding feature.

[0139] In step S502, the pooling layer is used to perform average pooling processing on the case semantic encoding feature to obtain a historical case representation vector.

[0140] In step S501 of some embodiments, when the transformer layer is used to perform semantic encoding on the preliminary case feature embedding vector to obtain a case semantic encoding feature, the semantic feature information in the preliminary case feature embedding vector is obtained to obtain the case semantic encoding feature A. For example, the process of using the transformer layer to perform semantic extraction on the preliminary case feature embedding vector E can be represented as A = transformer(E), and the case semantic encoding feature can be represented as A = [A0, A1, …, A i ,…A L-1 ], where A i is the case semantic encoding feature corresponding to the i-th historical claim case.

[0141] In step S502 of some embodiments, the pooling layer is used to perform average pooling processing on the case semantic encoding feature to extract comprehensive semantic information of all semantic encoding features, and obtain a historical case representation vector S, which can be represented as formula (1):

[0142]

[0143] S = pool(A) (1) iis the historical case representation vector corresponding to the ith historical claim case, the case semantic vector of each historical claim case is calculated based on this manner, all behavior semantic vectors are averaged to obtain the final historical case representation vector S.

[0144] Through the above steps S501 to S502, the case semantic content of the historical claim case can be conveniently extracted, and the historical case representation vector is obtained, so that the historical claim case representation vector of the historical claim case can be used for claim prediction, and the accuracy of claim prediction is improved.

[0145] Please refer to Figure 6 In some embodiments, the first claim prediction label includes a first label and a second label, and the first claim prediction label is used to represent the predicted claim category of the historical claim case. Step S106 includes but is not limited to steps S601 to S603:

[0146] Step S601, performing claim scoring on the historical case representation vector based on a preset function to obtain a case claim score;

[0147] Step S602, if the case claim score is greater than a preset threshold, determining the first claim prediction label of the historical claim case as the first label, wherein the first label is used to represent that the historical claim case is an abnormal payment case;

[0148] Step S603, if the case claim score is less than or equal to the preset threshold, determining the first claim prediction label of the historical claim case as the second label, wherein the second label is used to represent that the historical claim case is a normal payment case.

[0149] In step S601 of some embodiments, the preset function in the output layer can be a softmax function or a sigmoid function, etc., without limitation. Taking the sigmoid function as an example, the sigmoid function and the historical case representation vector are used to perform claim scoring to obtain the case claim score. The claim scoring process can be represented as formula (2):

[0150] p=sigmoid(<S,W>+b) Formula (2)

[0151] Wherein, p is the case claim score, S is the historical case representation vector, W and b are both preset model parameters, W is a vector, and <S,W> is the dot product of the historical case representation vector S and the parameter W.

[0152] In step S602 of some embodiments, since the size of the case claim score can clearly reflect the predicted claim situation of the historical claim case, that is, the higher the case claim score, the greater the possibility that the historical claim case is an abnormal claim case, and the lower the case claim score, the smaller the possibility that the historical claim case is an abnormal claim case. Therefore, whether the historical claim case is an abnormal claim case can be determined according to the size of the case claim score. Specifically, the case claim score is compared with a preset threshold value, wherein the specific value of the preset threshold value can be set according to actual conditions and is not limited, for example, the preset threshold value can be 0.5. If the case claim score is greater than the preset threshold value, it indicates that the possibility that the historical claim case is an abnormal claim case is greater, and therefore the first claim prediction label of the historical claim case is determined as a first label, wherein the first label is used to indicate that the historical claim case is an abnormal claim case.

[0153] In step S603 of some embodiments, if the case claim score is less than or equal to the preset threshold value, it indicates that the possibility that the historical claim case is an abnormal claim case is smaller, and therefore the first claim prediction label of the historical claim case is determined as a second label, wherein the second label is used to indicate that the historical claim case is a normal claim case.

[0154] Through the above steps S601 to S603, it can be clearly determined whether the historical claim case has an abnormal claim situation according to the historical case representation vector, and the possibility that the historical claim case is an abnormal claim case is quantified by the case claim score, which can better improve the prediction accuracy of the claim situation of the case.

[0155] In step S107 of some embodiments, when the neural network model is trained based on the first claim prediction label, the real claim label of the historical claim case in the first claim data is first obtained, the model loss value is calculated according to the first claim prediction label and the real claim label, and the model parameters of the neural network model are adjusted according to the model loss value, so that the model loss value meets the preset training requirement, thereby obtaining a preliminary claim prediction model, wherein the specific process of calculating the model loss value according to the first claim prediction label and the real claim label can be represented as formula (3):

[0156] loss = -y1*log(y2)-(1-y1)*log(1-y2) Formula (3)

[0157] wherein y1 is a true claim label of the historical claim case, and y2 is the first claim prediction label. It should be noted that the above formula calculates the loss in the form of a label value, that is, y1 takes a value of 1 or 0, and when the true claim label indicates that the historical claim case is a normal compensation case, y1 takes 1; when the true claim label indicates that the historical claim case is an abnormal compensation case, y1 takes 0. y2 takes the case claim score described above.

[0158] Please refer to Figure 7 In some embodiments, the second claim data further includes a true claim label of the new claim case, and step S108 can include but is not limited to steps S701 to S705:

[0159] Step S701, embedding processing the second case feature based on an embedding layer to obtain an intermediate case feature embedding vector;

[0160] Step S702, performing semantic extraction on the intermediate case feature embedding vector based on an encoder to obtain a new case representation vector;

[0161] Step S703, performing label prediction on the new claim case based on the prediction layer and the new case representation vector to obtain a second claim prediction label of the new claim case, the second claim prediction label being used to indicate a predicted claim category of the new claim case;

[0162] Step S704, loss calculation based on the second claim prediction label and the true claim label to obtain a claim loss value;

[0163] Step S705, parameter optimization of the preliminary claim prediction model based on the claim loss value to obtain a target claim prediction model.

[0164] In step S701 of some embodiments, when embedding processing the second case feature based on the embedding layer, first, the numerical features and discrete features in the second case feature are extracted according to the specific feature types of the second case feature, and different embedding processing is performed based on the feature types to obtain the intermediate case feature embedding vector. When the extracted features are discrete features, the specific processing process is basically the same as steps S301 to S304 described above; when the extracted features are numerical features, the specific processing process is basically the same as steps S401 to S406 described above, which will not be repeated here.

[0165] In step S702 of some embodiments, when the intermediate case feature embedding vector is semantically extracted based on the encoder to obtain the new case representation vector, the intermediate case feature embedding vector can be semantically encoded based on a transformer layer to obtain a new case semantic coding feature; and the new case semantic coding feature is processed by average pooling based on a pooling layer to obtain the new case representation vector. The process is basically the same as the processes of steps S501 to S502 described above, and will not be repeated here.

[0166] In step S703 of some embodiments, when the new claim case is labeled and predicted based on the prediction layer and the new case representation vector to obtain the second claim prediction label of the new claim case, first, the claim score of the new case representation vector is obtained based on the prediction layer, and the second claim prediction label is determined based on the size relationship between the claim score and the preset threshold. When the claim score is greater than the preset threshold, the second claim prediction label indicates that the new claim case is a normal claim case; and when the claim score is less than or equal to the preset threshold, the second claim prediction label indicates that the new claim case is an abnormal claim case. The specific prediction process is basically the same as steps S601 to S603 described above, and will not be repeated here.

[0167] In step S704 of some embodiments, the loss calculation is performed based on the second claim prediction label and the real claim label to obtain the claim loss value. The calculation process is basically the same as step S107 described above, and will not be repeated here.

[0168] In step S705 of some embodiments, a preset learning rate is obtained. The specific value of the learning rate can be determined according to actual business requirements, without limitation. In order to enable the preliminary claim prediction model to retain the model knowledge obtained during training, the learning rate is usually very small. Therefore, the parameters of the preliminary claim prediction model are optimized based on the learning rate and the claim loss value, and the model parameters of the preliminary claim prediction model are constantly adjusted so that the learning rate and the claim loss value after optimization meet the preset optimization requirements, thereby obtaining the target claim prediction model.

[0169] Through the above steps S701 to S705, the preliminary claim prediction model obtained by training can be conveniently optimized based on the related data of the new claim case, the model performance of the preliminary claim prediction model is improved, and the target claim prediction model obtained after optimization has better claim prediction performance, which can effectively improve the judgment accuracy of the model on the claim case.

[0170] In step S109 of some embodiments, the target claim data of the target case can be obtained by writing a web crawler to crawl the data from the data source, or by other means. The target claim data includes the diagnosis disease code, the insurance type code, the date difference between the accident time and the insurance effective period, the insurance policy amount, the disease name, the disease symptoms, and the like.

[0171] In step S110 of some embodiments, the target claim data is input into the target claim prediction model for claim prediction. The target claim data is embedded by the embedding layer of the target claim prediction model to obtain a target case feature embedding vector. The target case feature embedding vector is semantically extracted by the encoder to obtain a target case representation vector. Finally, the target case representation vector is scored by the prediction layer, and the predicted claim category of the target case is determined according to the claim score result, i.e., whether the target case is a normal claim case. If the target case is a normal claim case, the target case can be directly paid. If the target case is an abnormal claim case, the target case needs to be further audited and confirmed, and whether the target case can be paid is determined according to the audit result. The specific process of step S110 is basically the same as steps S104 to S106 described above, and will not be repeated here.

[0172] The claim settlement prediction method of the embodiment of the present application, by obtaining first claim settlement data and second claim settlement data, wherein the first claim settlement data comprises first case characteristics of historical claim settlement cases, and the second claim settlement data comprises second case characteristics of newly added claim settlement cases; according to the first case characteristics and the second case characteristics, preliminary case characteristics are obtained, and feature data used for model training can be obtained more conveniently. Further, the preliminary case characteristics are input into a preset neural network model, wherein the neural network model comprises an embedding layer, an encoder and a prediction layer; the embedding layer is used for embedding processing on the preliminary case characteristics to obtain a preliminary case characteristic embedding vector; the encoder is used for semantic extraction on the preliminary case characteristic embedding vector to obtain a historical case representation vector; the prediction layer and the historical case representation vector are used for label prediction on the historical claim settlement cases to obtain first claim settlement prediction labels of the historical claim settlement cases, and the first claim settlement prediction labels are used for indicating predicted claim settlement categories of the historical claim settlement cases, case content information of the historical claim settlement cases can be extracted more conveniently, and the claim settlement prediction labels of the historical claim settlement cases are predicted. The neural network model is trained based on the first claim settlement prediction labels to obtain a preliminary claim settlement prediction model, and the preliminary claim settlement prediction model is parameter-optimized based on the second case characteristics to obtain a target claim settlement prediction model, finally, target claim settlement data of a target case is obtained; the target claim settlement data is input into the target claim settlement prediction model for claim settlement prediction to obtain a predicted claim settlement category of the target case, wherein the predicted claim settlement category is used for indicating that the target case is a normal settlement case or indicating that the target case is an abnormal settlement case, and this way can train and optimize the model by combining historical data and newly added data, which can not only take advantage of the related data of the historical claim settlement cases to train the model sufficiently, but also fine-tune the model by taking advantage of the higher accuracy of the related data of the newly added claim settlement cases, thereby improving the prediction accuracy of the model on the claim settlement situation of the claim settlement case.

[0173] Please refer to Figure 8 The embodiment of the present application also provides a claim settlement prediction device, which can realize the above claim settlement prediction method, and the device comprises:

[0174] The training data acquisition module 801 is used for acquiring first claim settlement data and second claim settlement data, wherein the first claim settlement data comprises first case characteristics of historical claim settlement cases, and the second claim settlement data comprises second case characteristics of newly added claim settlement cases;

[0175] The feature acquisition module 802 is used for obtaining preliminary case characteristics according to the first case characteristics and the second case characteristics;

[0176] The input module 803 is used for inputting the preliminary case characteristics into a preset neural network model, wherein the neural network model comprises an embedding layer, an encoder and a prediction layer;

[0177] The embedding module 804 is configured to perform embedding processing on the preliminary case features based on an embedding layer to obtain preliminary case feature embedding vectors;

[0178] The semantic extraction module 805 is configured to perform semantic extraction on the preliminary case feature embedding vectors based on an encoder to obtain historical case representation vectors;

[0179] The label prediction module 806 is configured to perform label prediction on the historical claim cases based on a prediction layer and the historical case representation vectors to obtain first claim prediction labels of the historical claim cases, the first claim prediction labels being used to represent predicted claim categories of the historical claim cases.

[0180] The model training module 807 is configured to perform model training on the neural network model based on the first claim prediction labels to obtain a preliminary claim prediction model.

[0181] The parameter optimization module 808 is configured to perform parameter optimization on the preliminary claim prediction model based on second case features to obtain a target claim prediction model.

[0182] The target data acquisition module 809 is configured to acquire target claim data of a target case.

[0183] The claim prediction module 810 is configured to input the target claim data into the target claim prediction model to perform claim prediction, and obtain a predicted claim category of the target case, the predicted claim category being used to represent that the target case is a normal claim case or a non-normal claim case.

[0184] The specific implementation of the claim prediction apparatus is basically the same as that of the above-mentioned specific embodiment of the claim prediction method, and will not be repeated here.

[0185] The embodiments of the present application also provide an electronic device, which comprises a memory, a processor, a program stored in the memory and capable of running on the processor, and a data bus for realizing connection and communication between the processor and the memory, and the program is executed by the processor to realize the above-mentioned claim prediction method. The electronic device can be any intelligent terminal including a tablet computer, a vehicle-mounted computer, etc.

[0186] Please refer to Figure 9 , Figure 9 The hardware structure of the electronic device of another embodiment is illustrated, and the electronic device comprises:

[0187] The processor 901 can be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits, and is configured to execute related programs to implement the technical solutions provided by the embodiments of the present application.

[0188] The memory 902 can be implemented by a ROM (Read Only Memory), a static storage device, a dynamic storage device, or a RAM (Random Access Memory), etc. The memory 902 can store an operating system and other application programs. When the technical solutions provided by the embodiments of the present application are implemented by software or firmware, the related program codes are stored in the memory 902 and are called and executed by the processor 901 to implement the claim settlement prediction method of the embodiments of the present application.

[0189] The input / output interface 903 is configured to implement information input and output.

[0190] The communication interface 904 is configured to implement the communication interaction between the device and other devices. The communication can be implemented by a wired manner (for example, a USB, a network cable, etc.) or a wireless manner (for example, a mobile network, WIFI, Bluetooth, etc.).

[0191] The bus 905 is configured to transmit information between the components (for example, the processor 901, the memory 902, the input / output interface 903, and the communication interface 904) of the device.

[0192] The processor 901, the memory 902, the input / output interface 903, and the communication interface 904 are connected to each other by the bus 905 to realize the communication connection between the components in the device.

[0193] The embodiments of the present application further provide a computer readable storage medium. The computer readable storage medium stores one or more programs. The one or more programs can be executed by one or more processors to implement the above claim settlement prediction method.

[0194] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0195] The claims prediction method, claims prediction device, electronic device, and computer-readable storage medium provided in this application embodiment acquire first claims data and second claims data. The first claims data includes first case features of historical claims, and the second claims data includes second case features of newly added claims. Based on the first and second case features, preliminary case features are obtained, which facilitates the acquisition of feature data for model training. Further, the preliminary case features are input into a preset neural network model, which includes an embedding layer, an encoder, and a prediction layer. The preliminary case features are embedded using the embedding layer to obtain a preliminary case feature embedding vector. Semantic extraction is performed on the preliminary case feature embedding vector using the encoder to obtain a historical case representation vector. Label prediction is performed on the historical claims based on the prediction layer and the historical case representation vector to obtain a first claims prediction label for the historical claims. This first claims prediction label represents the predicted claims category of the historical claims, facilitating the extraction of case content information and the prediction of claims prediction labels for historical claims. The neural network model is trained based on the first claim prediction label to obtain a preliminary claim prediction model. Then, the parameters of the preliminary claim prediction model are optimized based on the second case characteristics to obtain the target claim prediction model. Finally, the target claim data for the target case is obtained. This target claim data is input into the target claim prediction model for claim prediction to obtain the predicted claim category for the target case. The predicted claim category indicates whether the target case is a normal payout case or an abnormal payout case. This method combines historical and new data to train and optimize the model. It leverages the advantage of the large amount of relevant data from historical claims cases during model training, and also utilizes the higher accuracy of relevant data from new claims cases for fine-tuning the model, thus improving the model's accuracy in predicting claim situations.

[0196] The embodiments described in the specification of the present application are intended to more clearly illustrate the technical solutions of the present application, and do not constitute a limitation on the technical solutions provided by the present application. Those skilled in the art can know that, with the evolution of technology and the appearance of new application scenarios, the technical solutions provided by the present application are also applicable to similar technical problems.

[0197] Those skilled in the art can understand that, Figures 1-7 The technical solutions shown in the above description do not constitute a limitation on the embodiments of the present application, and can include more or fewer steps than shown, or combine certain steps, or different steps.

[0198] The device embodiments described above are merely illustrative, and units described as separate components can or can not be physically separated, that is, can be located in one place, or can be distributed on multiple network units. Part or all of the modules can be selected according to actual needs to achieve the purpose of the embodiments of the present application.

[0199] Those skilled in the art can understand that all or some steps in the above disclosed method, the functions of the modules / units in the system and the device can be implemented as software, firmware, hardware and their appropriate combinations.

[0200] The terms "first", "second", "third", "fourth" and the like (if any) in the specification of the present application and the above description are used to distinguish similar objects, and do not necessarily indicate a specific order or chronological sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0201] It should be understood that, in the application, "at least one" refers to one or more, and "multiple" refers to two or more. "And / or" is used to describe the association relationship of the associated objects, which means that there can be three relationships, for example, "A and / or B" can represent three cases of only A, only B, and A and B existing at the same time, wherein A and B can be singular or plural. The character " / " generally represents an "or" relationship between the associated objects before and after it. "At least one of the following" or similar expressions means any combination of these items, including any combination of single or multiple items. For example, at least one of a, b or c can represent a, b, c, "a and b", "a and c", "b and c", or "a and b and c", wherein a, b, and c can be single or multiple.

[0202] In several embodiments provided in the 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 illustrative, for example, the division of the above units is only a logical function division, and actual implementation can have another division manner, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the displayed or discussed units can be indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0203] The units described above as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, that is, they can be located in one place, or they can be distributed on multiple network units. According to actual needs, some or all of the units can be selected to achieve the purpose of the embodiment scheme.

[0204] In addition, the functional units in each embodiment of the application can be integrated in one processing unit, or each unit can be physically present, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit.

[0205] The integrated unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application, essentially or in other words, the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes multiple instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various program storage media.

[0206] The preferred embodiments of the embodiments of the present application are described above with reference to the accompanying drawings, and are not limited to the scope of the embodiments of the present application. Any modifications, equivalent replacements and improvements made by those skilled in the art without departing from the scope and essence of the embodiments of the present application shall be within the scope of the embodiments of the present application.

Claims

1. A claim settlement prediction method characterized by, The method comprises: obtaining first claim data and second claim data, wherein the first claim data comprises first case characteristics of historical claim cases, and the second claim data comprises second case characteristics of new claim cases; obtaining preliminary case characteristics according to the first case characteristics and the second case characteristics; inputting the preliminary case characteristics into a preset neural network model, wherein the neural network model comprises an embedding layer, an encoder and a prediction layer; embedding processing the preliminary case characteristics based on the embedding layer to obtain a preliminary case characteristic embedding vector; performing semantic extraction on the preliminary case characteristic embedding vector based on the encoder to obtain a historical case representation vector; performing label prediction on the historical claim cases based on the prediction layer and the historical case representation vector to obtain first claim prediction labels of the historical claim cases, wherein the first claim prediction labels are used to represent predicted claim categories of the historical claim cases; model training the neural network model based on the first claim prediction labels to obtain a preliminary claim prediction model; parameter optimization of the preliminary claim prediction model based on the second case characteristics to obtain a target claim prediction model; obtaining target claim data of a target case; inputting the target claim data into the target claim prediction model for claim prediction to obtain a predicted claim category of the target case, wherein the predicted claim category is used to represent that the target case is a normal claim case or a non-normal claim case.

2. The method of claim 1, wherein, The method comprises: identifying a feature state of the first case characteristics; if the feature state is a normal state, determining the first case characteristics as the preliminary case characteristics; wherein the normal state is used to represent that the first case characteristics do not have feature errors; if the feature state is an abnormal state, determining the second case characteristics as the preliminary case characteristics; wherein the abnormal state is used to represent that the first case characteristics have feature errors.

3. The method of claim 1, wherein, The preliminary case characteristic embedding vector comprises a discrete feature embedding vector and a discrete feature value embedding vector, and the embedding processing the preliminary case characteristics based on the embedding layer to obtain a preliminary case characteristic embedding vector comprises: performing feature extraction on the preliminary case characteristics to obtain discrete case characteristics; embedding processing the discrete case characteristics based on the embedding layer to obtain a discrete feature embedding vector; performing feature value query on the discrete feature embedding vector based on a preset first comparison table to obtain a discrete feature value embedding vector corresponding to the discrete feature embedding vector; obtaining the preliminary case characteristic embedding vector based on the discrete feature value embedding vector and the discrete feature embedding vector.

4. The method of claim 3, wherein, The preliminary case characteristic embedding vector comprises a numerical feature embedding vector and a numerical feature value embedding vector, and the embedding processing the preliminary case characteristics based on the embedding layer to obtain a preliminary case characteristic embedding vector comprises: performing feature extraction on the preliminary case characteristics to obtain numerical case characteristics; embedding processing on the numerical case characteristics based on the embedding layer to obtain a numerical characteristic embedding vector; feature bucketing on the numerical case characteristics to obtain a plurality of case characteristic buckets; discretization processing on the case characteristic buckets to obtain discrete characteristic buckets; value characteristic querying on the discrete characteristic buckets based on a preset second lookup table to obtain a numerical characteristic value embedding vector corresponding to the numerical characteristic embedding vector; obtaining the preliminary case characteristic embedding vector based on the numerical characteristic embedding vector and the numerical characteristic value embedding vector.

5. The method of claim 1, wherein, The encoder includes a transformer layer and a pooling layer, and the semantic extraction on the preliminary case characteristic embedding vector based on the encoder to obtain a historical case representation vector includes: semantic encoding on the preliminary case characteristic embedding vector based on the transformer layer to obtain a case semantic encoding feature; average pooling processing on the case semantic encoding feature based on the pooling layer to obtain the historical case representation vector.

6. The method of claim 1, wherein, The first claim prediction label includes a first label and a second label, and the label prediction on the historical claim case based on the prediction layer and the historical case representation vector to obtain the first claim prediction label of the historical claim case includes: claim scoring on the historical case representation vector based on a preset function to obtain a case claim score; if the case claim score is greater than a preset threshold, the first claim prediction label of the historical claim case is determined as the first label, where the first label is used to indicate that the historical claim case is an abnormal payment case; if the case claim score is less than or equal to the preset threshold, the first claim prediction label of the historical claim case is determined as the second label, where the second label is used to indicate that the historical claim case is a normal payment case.

7. The method according to any one of claims 1 to 6, characterized in that, The second claim data further includes a real claim label of the new claim case, and the parameter optimization on the preliminary claim prediction model based on the second case characteristic to obtain a target claim prediction model includes: embedding processing on the second case characteristic based on the embedding layer to obtain an intermediate case characteristic embedding vector; semantic extraction on the intermediate case characteristic embedding vector based on the encoder to obtain a new case representation vector; label prediction on the new claim case based on the prediction layer and the new case representation vector to obtain a second claim prediction label of the new claim case, the second claim prediction label being used to indicate a predicted claim category of the new claim case; loss calculation based on the second claim prediction label and the real claim label to obtain a claim loss value; parameter optimization on the preliminary claim prediction model based on the claim loss value to obtain the target claim prediction model.

8. A claim settlement prediction apparatus characterized by comprising: The device includes: a training data acquisition module configured to acquire first claim data and second claim data, wherein the first claim data includes first case characteristics of historical claim cases, and the second claim data includes second case characteristics of new claim cases; The feature acquisition module is configured to obtain preliminary case features according to the first case features and the second case features. The input module is configured to input the preliminary case features into a preset neural network model, wherein the neural network model comprises an embedding layer, an encoder, and a prediction layer. The embedding module is configured to perform embedding processing on the preliminary case features based on the embedding layer to obtain preliminary case feature embedding vectors. The semantic extraction module is configured to perform semantic extraction on the preliminary case feature embedding vectors based on the encoder to obtain historical case representation vectors. The label prediction module is configured to perform label prediction on the historical claim cases based on the prediction layer and the historical case representation vectors to obtain first claim prediction labels of the historical claim cases, wherein the first claim prediction labels are used to represent predicted claim categories of the historical claim cases. The model training module is configured to perform model training on the neural network model based on the first claim prediction labels to obtain a preliminary claim prediction model. The parameter optimization module is configured to perform parameter optimization on the preliminary claim prediction model based on the second case features to obtain a target claim prediction model. The target data acquisition module is configured to acquire target claim data of a target case. The claim prediction module is configured to input the target claim data into the target claim prediction model to perform claim prediction and obtain a predicted claim category of the target case, wherein the predicted claim category is used to represent that the target case is a normal claim case or that the target case is an abnormal claim case.

9. An electronic device, comprising: The electronic device comprises a memory and a processor, the memory stores a computer program, and the processor implements the claim prediction method in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium storing a computer program, the computer-readable storage medium comprising: The computer program is executed by the processor to implement the claim prediction method in any one of claims 1 to 7.

Images