Abnormal insurance data identification method and apparatus, and computer device

Abstract

The invention discloses an abnormal insurance data identification method and device and computer equipment, relates to the technical field of artificial intelligence and data processing, and mainly aims at solving the problem that existing insurance data are abnormal. Comprising the steps of obtaining to-be-identified insurance data, and extracting target abnormal insurance data in the insurance data according to a data category and an abnormal object library; and performing identification processing on the target abnormal insurance data based on an abnormal identification model which completes model training to obtain an identification result of the target abnormal insurance data, the abnormal identification model being obtained by performing transfer learning training on a plurality of deep semantic residual networks based on an abnormal insurance sample, the abnormal insurance sample comprises an insurance sample with an abnormal label; and determining an abnormal identification result of the insurance data based on the identification result, and reporting the abnormal identification result to adjust the target abnormal insurance data.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence and data processing technology, and in particular to a method and device for identifying abnormal insurance data, and a computer device. Background Art

[0002] With the rapid development of big data and artificial intelligence technology, insurers will make predictions about insurance data to screen out valid and accurate insurance data for subsequent processing. For example, it is necessary to identify abnormal behaviors such as multiple entries of the same information in the insurance data in order to delete such data and ensure the validity and security of insurance data.

[0003] At present, the identification of existing abnormal insurance data is usually carried out by directly comparing and screening according to the data threshold. However, direct comparison based on the data threshold alone cannot identify abnormal data within the data threshold, increasing the risk of abnormal data affecting insurance execution, thereby affecting the effectiveness of insurance execution. Therefore, an abnormal insurance data identification method is urgently needed to solve the above problem. Summary of the invention

[0004] In view of this, the present invention provides a method and device for identifying abnormal insurance data, and a computer device, the main purpose of which is to solve the problem of poor accuracy in the existing identification of abnormal insurance data.

[0005] According to one aspect of the present invention, a method for identifying abnormal insurance data is provided, comprising:

[0006] Obtaining insurance data to be identified, and extracting target abnormal insurance data from the insurance data according to data categories and an abnormal object library;

[0007] Based on the abnormal recognition model that has completed model training, the target abnormal insurance data is recognized and processed to obtain a recognition result of the target abnormal insurance data, wherein the abnormal recognition model is obtained by performing transfer learning training on multiple deep semantic residual networks based on abnormal insurance samples, and the abnormal insurance samples include insurance samples with abnormal labels;

[0008] An abnormal identification result of the insurance data is determined based on the identification result and reported to adjust the target abnormal insurance data.

[0009] Furthermore, before the abnormal recognition model based on the completed model training performs recognition processing on the target abnormal insurance data to obtain the recognition result of the target abnormal insurance data, the method also includes:

[0010] Acquire abnormal insurance samples and construct multiple deep semantic residual networks, wherein the abnormal labels include a first abnormal label and a second abnormal label, wherein the first abnormal label is used to pre-train the multiple deep semantic residual networks, and the second abnormal label is used to perform transfer learning on the deep semantic residual network;

[0011] Determining a loss function of the deep semantic residual network, and pre-training the deep semantic residual network based on the abnormal insurance sample with the first abnormal label and the loss function;

[0012] Determine the transfer learning loss corresponding to the second abnormal label, and perform transfer learning on the parameters of the pre-trained deep semantic residual network based on the transfer learning loss and the abnormal insurance sample with the second abnormal label to obtain the abnormal recognition model that completes the model training.

[0013] Further, the insurance data is text data, the abnormal insurance sample is abnormal text, and the determining of the loss function of the deep semantic residual network and pre-training the deep semantic residual network based on the abnormal insurance sample with the first abnormal label and the loss function includes:

[0014] Determine an abnormal text with the first abnormal label based on the abnormal label description event, and determine a normal text corresponding to the abnormal text;

[0015] Based on the normal text and the abnormal text, a plurality of loss functions of the deep semantic residual network are determined, and the deep semantic residual network is pre-trained based on the abnormal text, the normal sample and the loss function.

[0016] Further, determining the transfer learning loss corresponding to the second abnormal label includes:

[0017] Identifying the abnormal insurance sample with the second abnormal label based on the pre-trained deep semantic residual network to obtain a feature vector, and determining the transfer learning weight of the deep semantic residual network based on the feature vector;

[0018] A transfer cost function is determined based on the transfer learning network, and global aggregation is performed based on the transfer learning weight and the transfer cost function to obtain the transfer learning loss.

[0019] Further, the determining the transfer learning weight of the deep semantic residual network based on the feature vector includes:

[0020] Determine the predicted value of the abnormal insurance sample of the second abnormal label obtained in the deep semantic residual network based on the feature vector, and obtain the prediction threshold corresponding to the deep semantic residual network;

[0021] The transfer learning weight of the deep semantic residual network is determined based on a comparison result between the predicted value and the prediction threshold.

[0022] Furthermore, determining the transfer cost function based on the transfer learning network includes:

[0023] Based on the pre-trained deep semantic residual network, text semantic recognition processing is performed on the insurance text matching the second abnormal label to obtain a first text semantic vector;

[0024] Performing text semantic recognition processing on the insurance text matching the second abnormal label based on the transfer learning network to obtain a second text semantic vector;

[0025] A migration cost function is determined according to the first text semantic vector and the second text semantic vector.

[0026] Further, extracting target abnormal insurance data from the insurance data according to the data category and the abnormal object library includes:

[0027] If the data category is a text category, then calling an abnormal object library corresponding to the text category, wherein the abnormal object library stores a plurality of abnormal text objects;

[0028] Target abnormal text is extracted from the insurance data according to the plurality of abnormal text objects in the abnormal object library and determined as the target abnormal insurance data.

[0029] Furthermore, after determining the abnormal identification result of the insurance data based on the identification result and reporting it, the method further includes:

[0030] Retrieving a trained processing model, wherein the processing model is obtained by training a long short-term memory model based on a processing behavior sample manually processed on a key database, wherein the key database stores insurance data of different abnormal types;

[0031] The abnormal identification result is processed based on the processing model to obtain an abnormal insurance data processing result, and the abnormal insurance data processing result is sent to the audit end.

[0032] According to another aspect of the present invention, there is provided a device for identifying abnormal insurance data, comprising:

[0033] An acquisition module, used to acquire the insurance data to be identified, and extract target abnormal insurance data from the insurance data according to the data category and the abnormal object library;

[0034] A processing module, used to perform recognition processing on the target abnormal insurance data based on the abnormal recognition model that has completed model training, to obtain a recognition result of the target abnormal insurance data, wherein the abnormal recognition model is obtained by performing transfer learning training on multiple deep semantic residual networks based on abnormal insurance samples, and the abnormal insurance samples include insurance samples with abnormal labels;

[0035] A determination module is used to determine the abnormal identification result of the insurance data based on the identification result, and report it to adjust the target abnormal insurance data.

[0036] Further,

[0037] The acquisition module is further used to acquire abnormal insurance samples and construct multiple deep semantic residual networks, wherein the abnormal labels include a first abnormal label and a second abnormal label, wherein the first abnormal label is used to pre-train the multiple deep semantic residual networks, and the second abnormal label is used to perform transfer learning on the deep semantic residual networks;

[0038] The determination module is further used to determine a loss function of the deep semantic residual network, and pre-train the deep semantic residual network based on the abnormal insurance sample with the first abnormal label and the loss function;

[0039] The determination module is also used to determine the transfer learning loss corresponding to the second abnormal label, and perform transfer learning on the parameters of the pre-trained deep semantic residual network based on the transfer learning loss and the abnormal insurance samples with the second abnormal label to obtain the abnormal recognition model that completes the model training.

[0040] Furthermore, the insurance data is text data, and the abnormal insurance sample is abnormal text.

[0041] The determination module is specifically used to determine the abnormal text with the first abnormal label based on the abnormal label description event, and determine the normal text corresponding to the abnormal text; determine the loss functions of multiple deep semantic residual networks based on the normal text and the abnormal text, and pre-train the deep semantic residual network based on the abnormal text, the normal sample and the loss function.

[0042] Furthermore, the determination module is specifically used to identify abnormal insurance samples with the second abnormal label based on the pre-trained deep semantic residual network to obtain a feature vector, and determine the transfer learning weight of the deep semantic residual network based on the feature vector; determine the transfer cost function based on the transfer learning network, and perform global aggregation based on the transfer learning weight and the transfer cost function to obtain the transfer learning loss.

[0043] Furthermore, the determination module is specifically used to determine the predicted value of the abnormal insurance sample with the second abnormal label obtained in the deep semantic residual network based on the feature vector, and obtain the prediction threshold corresponding to the deep semantic residual network; determine the transfer learning weight of the deep semantic residual network based on the comparison result between the predicted value and the prediction threshold.

[0044] Furthermore, the determination module is specifically used to perform text semantic recognition processing on the insurance text matching the second exception label based on the pre-trained deep semantic residual network to obtain a first text semantic vector; perform text semantic recognition processing on the insurance text matching the second exception label based on the transfer learning network to obtain a second text semantic vector; and determine the transfer cost function based on the first text semantic vector and the second text semantic vector.

[0045] Furthermore, the acquisition module is specifically used to retrieve the exception object library corresponding to the text category if the data category is a text category, and the exception object library stores multiple exception text objects; extract the target exception text from the insurance data according to the multiple exception text objects in the exception object library, and determine it as the target exception insurance data.

[0046] Furthermore, the device also includes:

[0047] A retrieval module, used to retrieve a trained processing model, wherein the processing model is obtained by training a long short-term memory model based on a processing behavior sample manually processed on a key database, wherein the key database stores insurance data of different abnormal types;

[0048] The sending module is used to process the abnormal identification result based on the processing model to obtain the abnormal insurance data processing result, and send the abnormal insurance data processing result to the review end.

[0049] According to another aspect of the present invention, a storage medium is provided, wherein at least one executable instruction is stored in the storage medium, and the executable instruction enables a processor to execute operations corresponding to the above-mentioned abnormal insurance data identification method.

[0050] According to another aspect of the present invention, there is provided a computer device, comprising: a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus;

[0051] The memory is used to store at least one executable instruction, and the executable instruction enables the processor to execute operations corresponding to the above-mentioned abnormal insurance data identification method.

[0052] By means of the above technical solution, the technical solution provided by the embodiment of the present invention has at least the following advantages:

[0053] The present invention provides a method and device for identifying abnormal insurance data, and a computer device. Compared with the prior art, the embodiments of the present invention obtain insurance data to be identified, and extract target abnormal insurance data from the insurance data according to data categories and abnormal object libraries; identify and process the target abnormal insurance data based on an abnormal identification model that has completed model training to obtain an identification result of the target abnormal insurance data, wherein the abnormal identification model is obtained by performing transfer learning training on multiple deep semantic residual networks based on abnormal insurance samples, and the abnormal insurance samples include insurance samples with abnormal labels; determine the abnormal identification result of the insurance data based on the identification result, and report it to adjust and process the target abnormal insurance data, so as to achieve the purpose of effectively identifying abnormalities of different insurance data based on intelligent artificial algorithms, and make the intelligent artificial model lightweight based on transfer learning of the intelligent artificial model, thereby improving the accuracy and credibility of artificial intelligence model recognition, thereby improving the purpose of identifying abnormal data in insurance data.

[0054] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented according to the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0056] Figure 1 A flow chart of a method for identifying abnormal insurance data provided by an embodiment of the present invention is shown;

[0057] Figure 2 A block diagram showing the composition of an abnormal insurance data identification device provided by an embodiment of the present invention;

[0058] Figure 3 A schematic structural diagram of a computer device provided by an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0059] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0060] The embodiment of the present invention provides a method for identifying abnormal insurance data. Figure 1 As shown, the method includes:

[0061] 101. Obtain insurance data to be identified, and extract target abnormal insurance data from the insurance data according to data category and abnormal object library.

[0062] In the embodiment of the present invention, the current execution end, as the execution subject for identifying insurance data, can be a computer device server or a cloud server. The insurance data is the data content for executing different insurance businesses, including but not limited to insurance data entered offline, insurance data generated by Internet insurance products, etc. At this time, the current execution end, as the main end for executing insurance business, directly retrieves all insurance data in the process of executing insurance business, which is not specifically limited in the embodiment of the present invention. Among them, the data categories include numerical data categories, text data categories, image data categories, etc., and the abnormal object library stores abnormal objects pre-selected for different insurance contents, so as to pre-screen the target abnormal insurance data that needs to be further identified from the insurance data based on the abnormal object library, that is, the target abnormal insurance data is abnormal and not directly judged as abnormal insurance data, which is not specifically limited in the embodiment of the present invention.

[0063] 102. Perform identification processing on the target abnormal insurance data based on the abnormal identification model that has completed model training to obtain an identification result of the target abnormal insurance data.

[0064] In an embodiment of the present invention, after the current execution end determines the target abnormal insurance data, it is input into the abnormal recognition model that has completed model training for recognition. At this time, the abnormal recognition model is obtained by transfer learning training of multiple deep semantic residual networks based on abnormal insurance samples, and the abnormal insurance samples include insurance samples with abnormal labels. Among them, the deep residual network (ResNet) effectively solves the gradient vanishing and degradation problems that occur in deep neural network training by introducing residual connections, and more accurately identifies valid data from abnormal insurance data, thereby achieving a deeper network structure and higher accuracy. In addition, in order to enable the abnormal recognition model to effectively transfer abnormal features during training, thereby increasing the recognition of abnormal data, the current execution end performs transfer learning on multiple deep semantic residual networks when training the abnormal recognition model, so as to combine insurance samples with abnormal labels to achieve a more lightweight neural network learning purpose.

[0065] 103. Determine an abnormal identification result of the insurance data based on the identification result, and report it to adjust the target abnormal insurance data.

[0066] In the embodiment of the present invention, the recognition result is the result of marking the insurance data with abnormal labels after the target abnormal insurance data is recognized. At this time, due to the different data types of the insurance data and the number of data, the labels of each insurance data in the recognition result can be counted, including normal labels and abnormal labels, so as to obtain the abnormal recognition result of the insurance data. For example, the target abnormal insurance data includes 100 insurance data to be recognized. After the recognition of the abnormal recognition model, there are 3 insurance data with abnormal labels marked. Then the corresponding abnormal recognition result can be directly outputting these 3 insurance data, or it can be based on the statistics of 3 insurance data and 97 insurance data to obtain the abnormal recognition result of whether the 100 insurance data are available, etc., which is not specifically limited in the embodiment of the present invention. Furthermore, after obtaining the abnormal recognition result, in order to effectively realize the automatic reporting of the target abnormal insurance data in the insurance data, the current execution end will report the target abnormal insurance data with abnormal conditions to the manual review computer device or the review server, so as to adjust and process the target abnormal insurance data.

[0067] It should be noted that after the current execution end reports, in order to better utilize the insurance data for subsequent insurance applications and improve the efficiency and credibility of the use of insurance data, the target abnormal insurance data can also be adjusted and processed at the current end. At this time, it includes fully automated abnormal data modification and semi-automated abnormal data modification. Among them, the fully automated abnormal data modification can adjust all target abnormal data through a deep learning model that has completed model training to obtain new data. At this time, the deep learning model is obtained by model training based on the abnormal data adjustment training set, and the adjusted new data samples in the abnormal data adjustment training set can be manually adjusted. The embodiment of the present invention does not make specific limitations. Semi-automated abnormal data modification can be to adjust the target abnormal data through a deep learning model to obtain reference adjustment data for manual modification, and then send it to the manual to determine which specific insurance data needs to be modified and adjusted to achieve the purpose of semi-automatic adjustment.

[0068] In another embodiment of the present invention, for further definition and explanation, the step is to perform identification processing on the target abnormal insurance data based on the abnormal identification model that has completed model training, and before obtaining the identification result of the target abnormal insurance data, the method further includes:

[0069] Obtain abnormal insurance samples and construct multiple deep semantic residual networks;

[0070] Determining a loss function of the deep semantic residual network, and pre-training the deep semantic residual network based on the abnormal insurance sample with the first abnormal label and the loss function;

[0071] Determine the transfer learning loss corresponding to the second abnormal label, and perform transfer learning on the parameters of the pre-trained deep semantic residual network based on the transfer learning loss and the abnormal insurance sample with the second abnormal label to obtain the abnormal recognition model that completes the model training.

[0072] In order to improve the accuracy and timeliness of identifying abnormal insurance data, the current execution end first obtains abnormal insurance samples and constructs multiple deep semantic residual networks before identifying based on the abnormal recognition model. The abnormal insurance sample data includes insurance samples with abnormal labels. The insurance samples include but are not limited to auto insurance claim texts, bank product evaluation texts, etc. for different insurance scenarios, which are not specifically limited in the embodiment of the present invention. Among them, the abnormal label includes a first abnormal label and a second abnormal label. The first abnormal label is used to pre-train multiple deep semantic residual networks, and the second abnormal label is used to transfer learning to the deep semantic residual network. At the same time, the current execution end constructs multiple deep semantic residual networks and determines the loss function of the deep semantic residual network so as to pre-train the deep semantic residual network based on the abnormal insurance samples with the first abnormal label and the loss function. Among them, the loss function maps the value of a random event or its related random variable to a non-negative real number to represent the "risk" or "loss" of the random event. A 0-1 loss function, a hinge loss function, etc. can be used, which is not specifically limited in the embodiment of the present invention.

[0073] It should be noted that in the embodiment of the present invention, the sampling transfer learning method is used for model training. Therefore, the current execution end first determines the transfer learning loss corresponding to the second abnormal label, and based on this transfer learning loss and the abnormal insurance samples with the second abnormal label, transfer learning is performed on the parameters of the pre-trained deep semantic residual network to obtain the abnormal recognition model that completes the model training. Wherein, when performing transfer learning, the insurance samples corresponding to the first abnormal label are used as source domain samples, and multiple constructed deep semantic residual networks are pre-trained as a lightweight transfer learning network relative to the deep semantic residual network. The insurance samples corresponding to the second abnormal label are used as target domain samples for transfer learning of the deep semantic residual network, and the abnormal features learned from the source domain are applied to the target domain for re-learning. The embodiment of the present invention does not make specific limitations.

[0074] In another embodiment of the present invention, for further definition and explanation, the step of determining the loss function of the deep semantic residual network, and pre-training the deep semantic residual network based on the abnormal insurance sample with the first abnormal label and the loss function includes:

[0075] Determine an abnormal text with the first abnormal label based on the abnormal label description event, and determine a normal text corresponding to the abnormal text;

[0076] Based on the normal text and the abnormal text, a plurality of loss functions of the deep semantic residual network are determined, and the deep semantic residual network is pre-trained based on the abnormal text, the normal sample and the loss function.

[0077] In order to improve the accuracy of identifying abnormal data and realize pre-training of the source domain model as transfer learning, the current execution end determines the loss function. Specifically, firstly, based on the abnormal label description event, the abnormal text with the first abnormal label is determined. Among them, the abnormal label description event is used to indicate the trigger time for marking the abnormal data for detecting abnormal content, and the abnormal content can be configured based on different insurance contents. At the same time, when the abnormal text is determined, the remaining text can be determined as normal text. Furthermore, the current execution end determines the loss functions of multiple deep semantic residual networks based on normal text and abnormal text. At this time, a suitable loss function can be pre-selected based on the number of normal text and abnormal text corresponding to the abnormal text, preferably a Loss function, which is not specifically limited in the embodiment of the present invention. After determining the loss function, the current execution end pre-trains the deep semantic residual network based on the abnormal text, normal samples and loss function, thereby pre-training the deep semantic residual network in the source domain sample to achieve the purpose of learning transfer in the target domain.

[0078] In another embodiment of the present invention, for further definition and explanation, the step of determining the transfer learning loss corresponding to the second abnormal label includes:

[0079] Identifying the abnormal insurance sample with the second abnormal label based on the pre-trained deep semantic residual network to obtain a feature vector, and determining the transfer learning weight of the deep semantic residual network based on the feature vector;

[0080] A transfer cost function is determined based on the transfer learning network, and global aggregation is performed based on the transfer learning weight and the transfer cost function to obtain the transfer learning loss.

[0081] In order to achieve the transfer of content learned from the source domain to the target domain, thereby improving the accuracy of identifying abnormal data, the current execution end, when determining the transfer learning loss, specifically, first identifies the abnormal insurance samples with the second abnormal label based on the pre-trained deep semantic residual network to obtain a feature vector. Among them, since the trained deep semantic residual network is trained in the source domain, at this time, the abnormal insurance samples corresponding to the second abnormal label are samples in the target domain, so the corresponding feature vector is the result obtained by identifying the model obtained in the source domain in the target domain, and then, the transfer learning weight is determined based on this feature vector, and the embodiment of the present invention does not make specific restrictions. After determining the transfer learning weight, in order to improve the adaptability of the model in the target domain, the transfer cost function can be determined based on the transfer learning network, and global aggregation is performed based on the transfer learning weight and the transfer cost function to obtain the transfer learning loss, wherein the transfer learning network is the deep semantic residual network that has been trained in the source domain to be debugged, and the embodiment of the present invention does not make specific restrictions.

[0082] In another embodiment of the present invention, for further limitation and explanation, the step of determining the transfer learning weight of the deep semantic residual network based on the feature vector includes:

[0083] Determine the predicted value of the abnormal insurance sample of the second abnormal label obtained in the deep semantic residual network based on the feature vector, and obtain the prediction threshold corresponding to the deep semantic residual network;

[0084] The transfer learning weight of the deep semantic residual network is determined based on a comparison result between the predicted value and the prediction threshold.

[0085] In order to realize the training of multiple deep semantic residual networks based on transfer learning to lightweight the model and meet the anomaly recognition requirements of different insurance data, the current execution end, when determining the transfer learning weight, specifically, determines the predicted value of the abnormal insurance sample with the second abnormal label in the deep semantic residual network based on the feature vector, and obtains the prediction threshold corresponding to the deep semantic residual network, so as to determine the transfer learning weight of the deep semantic residual network based on the comparison result between the predicted value and the prediction threshold. Specifically, in the scenario where the insurance data is an insurance claim text, when the quantitative prediction result corresponding to the u-th deep semantic residual network of the v-th insurance policy text (the eigenvalue obtained by processing the u-th deep semantic residual network in the feature vector) is greater than the quantitative prediction limit of the u-th deep semantic residual network, the transfer learning weight of the u-th deep semantic residual network for the v-th insurance policy debugging text is determined according to the quantitative prediction result and the quantitative prediction limit. When the quantized prediction result corresponding to the vth insurance policy text in the uth deep semantic residual network is not greater than the quantized prediction limit of the uth deep semantic residual network, the first set coefficient is determined as the transfer learning weight of the uth deep semantic residual network for the vth insurance policy debugging text. In this way, the transfer learning contribution coefficient of the corresponding insurance policy debugging text can be accurately determined based on the comparison between the quantized prediction result and the quantized prediction limit, and because the first set coefficient and the quantized prediction limit can be flexibly set, the flexibility of determining the transfer learning weight can be improved, so that the determination scheme of the transfer learning weight can be applicable to different insurance data abnormality recognition scenarios.

[0086] In another embodiment of the present invention, for further definition and explanation, the step of determining the transfer cost function based on the transfer learning network includes:

[0087] Based on the pre-trained deep semantic residual network, text semantic recognition processing is performed on the insurance text matching the second abnormal label to obtain a first text semantic vector;

[0088] Performing text semantic recognition processing on the insurance text matching the second abnormal label based on the transfer learning network to obtain a second text semantic vector;

[0089] A migration cost function is determined according to the first text semantic vector and the second text semantic vector.

[0090] In order to achieve the purpose of AI intelligent recognition of lightweight models, when the current execution end determines the transfer cost function based on the transfer learning network, specifically, firstly, based on the deep semantic residual network that has completed pre-training, the insurance text matching the second abnormal label is processed for text semantic recognition to obtain the first text semantic vector. At the same time, based on the transfer learning network, the insurance text matching the second abnormal label is processed for text semantic recognition to obtain the second text semantic vector. Among them, when determining the transfer cost function based on the first text semantic vector and the second text semantic vector, it can be calculated and determined based on the weighted summation method, and the embodiment of the present invention does not make specific limitations. In the specific implementation scenario where the insurance data is the insurance claim text, the deep semantic residual network that has completed training in each source domain is used to perform text semantic mining on the vth insurance claim text to obtain each first text semantic vector of the vth insurance claim text. The v-th insurance claim text is subjected to text semantic mining through the transfer learning network to obtain the second text semantic vector of the v-th past motor insurance policy debugging text, and the migration cost function of the v-th insurance claim text in each deep semantic residual network is determined according to the second text semantic vector of the v-th insurance claim text and each first text semantic vector of the v-th insurance claim text. In the embodiment of the present invention, text semantic mining can be understood as mining text features of the corresponding migration cost function, the first text semantic vector can be understood as the text feature of the teacher neural network, and the second text semantic vector can be understood as the text feature of the student neural network. On this basis, through different text semantic vectors (text features), the migration cost function of the corresponding migration cost function in each deep semantic residual network can be accurately determined.

[0091] In another embodiment of the present invention, for further definition and explanation, the step of extracting target abnormal insurance data from the insurance data according to the data category and the abnormal object library includes:

[0092] If the data category is a text category, then calling an abnormal object library corresponding to the text category, wherein the abnormal object library stores a plurality of abnormal text objects;

[0093] Target abnormal text is extracted from the insurance data according to the plurality of abnormal text objects in the abnormal object library and determined as the target abnormal insurance data.

[0094] In order to realize multi-scenario application of abnormal data, so as to meet the demand for abnormal data identification of different insurance data, the current execution end first determines the data category of the insurance data when extracting the target abnormal insurance data to be identified from the insurance data. At this time, the data category includes a numerical category, a text category, an image category, etc. In a specific real scenario, if the data category of the insurance data is a text category, the abnormal object library corresponding to the text category is retrieved. At this time, a plurality of abnormal text objects are stored in the abnormal object library, so that the target abnormal insurance data is obtained by searching the insurance data according to the abnormal text objects. For example, the insurance data is an insurance claim text, and the abnormal object library includes a plurality of abnormal behavior text objects, so that the target abnormal text that needs to be identified is searched from the insurance claim text according to the abnormal behavior text object, and is input into the abnormal identification model as the target abnormal insurance data. The embodiment of the present invention does not make specific limitations.

[0095] In another embodiment of the present invention, for further definition and explanation, after determining the abnormal identification result of the insurance data based on the identification result and reporting it, the method further includes:

[0096] Retrieve the trained processing model;

[0097] The abnormal identification result is processed based on the processing model to obtain an abnormal insurance data processing result, and the abnormal insurance data processing result is sent to the audit end.

[0098] In order to effectively process abnormal insurance data, so as to meet the automatic detection requirements of abnormal insurance data in different scenarios, the current execution end calls the processing model that has completed model training, so as to process the abnormal identification results based on this processing model and obtain the processing results. Among them, the processing model is obtained by training the long short-term memory model based on the processing behavior samples of artificial processing of the key database. The processing model is a long short-term memory model (Long Short-Term Memory, LSTM). LSTM is a kind of neural network containing LSTM blocks, which can remember the value of indefinite time length, so as to judge whether the input is important enough to be remembered and whether it can be output through the LSTM block, so as to complete the processing and determine which abnormal insurance data is output according to the insurance data processing results learned in the key database. Among them, the key database stores insurance data of different abnormal types, that is, the purpose of processing abnormal insurance data based on the processing model is to mark according to different abnormal types in the key database, so as to send it to the audit end for manual review. Specifically, the abnormal types include but are not limited to data form abnormal types, data source abnormal types, business scenario abnormal types, etc., so as to meet the diversified processing requirements for abnormal situations in insurance data.

[0099] An embodiment of the present invention provides a method for identifying abnormal insurance data. Compared with the prior art, the embodiment of the present invention obtains insurance data to be identified, and extracts target abnormal insurance data from the insurance data according to data categories and abnormal object libraries; identifies and processes the target abnormal insurance data based on an abnormal identification model that has completed model training to obtain an identification result of the target abnormal insurance data, wherein the abnormal identification model is obtained by performing transfer learning training on multiple deep semantic residual networks based on abnormal insurance samples, and the abnormal insurance samples include insurance samples with abnormal labels; determines the abnormal identification result of the insurance data based on the identification result, thereby achieving the purpose of effectively identifying abnormalities of different insurance data based on intelligent artificial algorithms, and makes the intelligent artificial model lightweight based on transfer learning of the intelligent artificial model, thereby improving the accuracy and credibility of artificial intelligence model recognition, thereby improving the purpose of identifying abnormal data in insurance data.

[0100] Furthermore, as a response to the above Figure 1 The embodiment of the present invention provides a device for identifying abnormal insurance data, such as Figure 2 As shown, the device comprises:

[0101] An acquisition module 21 is used to acquire insurance data to be identified, and extract target abnormal insurance data from the insurance data according to data categories and an abnormal object library;

[0102] A processing module 22 is used to perform recognition processing on the target abnormal insurance data based on the abnormal recognition model that has completed model training, and obtain the recognition result of the target abnormal insurance data, wherein the abnormal recognition model is obtained by performing transfer learning training on multiple deep semantic residual networks based on abnormal insurance samples, and the abnormal insurance samples include insurance samples with abnormal labels;

[0103] The determination module 23 is used to determine the abnormal identification result of the insurance data based on the identification result, and report it to adjust the target abnormal insurance data.

[0104] Further,

[0105] The acquisition module is further used to acquire abnormal insurance samples and construct multiple deep semantic residual networks, wherein the abnormal labels include a first abnormal label and a second abnormal label, wherein the first abnormal label is used to pre-train the multiple deep semantic residual networks, and the second abnormal label is used to perform transfer learning on the deep semantic residual networks;

[0106] The determination module is further used to determine a loss function of the deep semantic residual network, and pre-train the deep semantic residual network based on the abnormal insurance sample with the first abnormal label and the loss function;

[0107] The determination module is also used to determine the transfer learning loss corresponding to the second abnormal label, and perform transfer learning on the parameters of the pre-trained deep semantic residual network based on the transfer learning loss and the abnormal insurance samples with the second abnormal label to obtain the abnormal recognition model that completes the model training.

[0108] Furthermore, the insurance data is text data, and the abnormal insurance sample is abnormal text.

[0109] The determination module is specifically used to determine the abnormal text with the first abnormal label based on the abnormal label description event, and determine the normal text corresponding to the abnormal text; determine the loss functions of multiple deep semantic residual networks based on the normal text and the abnormal text, and pre-train the deep semantic residual network based on the abnormal text, the normal sample and the loss function.

[0110] Furthermore, the determination module is specifically used to identify abnormal insurance samples with the second abnormal label based on the pre-trained deep semantic residual network to obtain a feature vector, and determine the transfer learning weight of the deep semantic residual network based on the feature vector; determine the transfer cost function based on the transfer learning network, and perform global aggregation based on the transfer learning weight and the transfer cost function to obtain the transfer learning loss.

[0111] Furthermore, the determination module is specifically used to determine the predicted value of the abnormal insurance sample with the second abnormal label obtained in the deep semantic residual network based on the feature vector, and obtain the prediction threshold corresponding to the deep semantic residual network; determine the transfer learning weight of the deep semantic residual network based on the comparison result between the predicted value and the prediction threshold.

[0112] Furthermore, the determination module is specifically used to perform text semantic recognition processing on the insurance text matching the second exception label based on the pre-trained deep semantic residual network to obtain a first text semantic vector; perform text semantic recognition processing on the insurance text matching the second exception label based on the transfer learning network to obtain a second text semantic vector; and determine the transfer cost function based on the first text semantic vector and the second text semantic vector.

[0113] Furthermore, the acquisition module is specifically used to retrieve the exception object library corresponding to the text category if the data category is a text category, and the exception object library stores multiple exception text objects; extract the target exception text from the insurance data according to the multiple exception text objects in the exception object library, and determine it as the target exception insurance data.

[0114] Furthermore, the device also includes:

[0115] A retrieval module, used to retrieve a trained processing model, wherein the processing model is obtained by training a long short-term memory model based on a processing behavior sample manually processed on a key database, wherein the key database stores insurance data of different abnormal types;

[0116] The sending module is used to process the abnormal identification result based on the processing model to obtain the abnormal insurance data processing result, and send the abnormal insurance data processing result to the review end.

[0117] An embodiment of the present invention provides a device for identifying abnormal insurance data. Compared with the prior art, the embodiment of the present invention obtains insurance data to be identified, and extracts target abnormal insurance data from the insurance data according to data categories and abnormal object libraries; identifies and processes the target abnormal insurance data based on an abnormal identification model that has completed model training to obtain an identification result of the target abnormal insurance data, wherein the abnormal identification model is obtained by transfer learning training of multiple deep semantic residual networks based on abnormal insurance samples, and the abnormal insurance samples include insurance samples with abnormal labels; determines the abnormal identification result of the insurance data based on the identification result, and reports it to adjust and process the target abnormal insurance data, thereby achieving the purpose of effectively identifying abnormalities of different insurance data based on intelligent artificial algorithms, and making the intelligent artificial model lightweight based on transfer learning of the intelligent artificial model, thereby improving the accuracy and credibility of artificial intelligence model recognition, thereby improving the purpose of identifying abnormal data in insurance data.

[0118] According to one embodiment of the present invention, a storage medium is provided, wherein the storage medium stores at least one executable instruction, and the computer executable instruction can execute the method for identifying abnormal insurance data in any of the above method embodiments.

[0119] Figure 3 A schematic diagram of the structure of a computer device provided according to an embodiment of the present invention is shown. The specific embodiment of the present invention does not limit the specific implementation of the computer device.

[0120] like Figure 3As shown, the computer device may include: a processor (processor) 302 , a communication interface (Communications Interface) 304 , a memory (memory) 306 , and a communication bus 308 .

[0121] The processor 302 , the communication interface 304 , and the memory 306 communicate with each other via the communication bus 308 .

[0122] The communication interface 304 is used to communicate with other devices such as clients or other servers.

[0123] The processor 302 is used to execute the program 310, and specifically can execute the relevant steps in the above-mentioned abnormal insurance data identification method embodiment.

[0124] Specifically, the program 310 may include program codes, which include computer operation instructions.

[0125] The processor 302 may be a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present invention. The one or more processors included in the computer device may be processors of the same type, such as one or more CPUs; or processors of different types, such as one or more CPUs and one or more ASICs.

[0126] The memory 306 is used to store the program 310. The memory 306 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0127] The program 310 may be specifically configured to enable the processor 302 to perform the following operations:

[0128] Obtaining insurance data to be identified, and extracting target abnormal insurance data from the insurance data according to data categories and an abnormal object library;

[0129] Based on the abnormal recognition model that has completed model training, the target abnormal insurance data is recognized and processed to obtain a recognition result of the target abnormal insurance data, wherein the abnormal recognition model is obtained by performing transfer learning training on multiple deep semantic residual networks based on abnormal insurance samples, and the abnormal insurance samples include insurance samples with abnormal labels;

[0130] An abnormal identification result of the insurance data is determined based on the identification result and reported to adjust the target abnormal insurance data.

[0131] Obviously, those skilled in the art should understand that the above modules or steps of the present invention can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, and optionally, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order than here, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.

[0132] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for identifying abnormal insurance data, characterized in that: include: Obtaining insurance data to be identified, and extracting target abnormal insurance data from the insurance data according to data categories and an abnormal object library; Based on the abnormal recognition model that has completed model training, the target abnormal insurance data is recognized and processed to obtain a recognition result of the target abnormal insurance data, wherein the abnormal recognition model is obtained by performing transfer learning training on multiple deep semantic residual networks based on abnormal insurance samples, and the abnormal insurance samples include insurance samples with abnormal labels; An abnormal identification result of the insurance data is determined based on the identification result and reported to adjust the target abnormal insurance data.

2. The method according to claim 1, characterized in that Before the abnormal recognition model based on the completed model training performs recognition processing on the target abnormal insurance data to obtain the recognition result of the target abnormal insurance data, the method further includes: Acquire abnormal insurance samples and construct multiple deep semantic residual networks, wherein the abnormal labels include a first abnormal label and a second abnormal label, wherein the first abnormal label is used to pre-train the multiple deep semantic residual networks, and the second abnormal label is used to perform transfer learning on the deep semantic residual network; Determining a loss function of the deep semantic residual network, and pre-training the deep semantic residual network based on the abnormal insurance sample with the first abnormal label and the loss function; Determine the transfer learning loss corresponding to the second abnormal label, and perform transfer learning on the parameters of the pre-trained deep semantic residual network based on the transfer learning loss and the abnormal insurance sample with the second abnormal label to obtain the abnormal recognition model that completes the model training.

3. The method according to claim 2, characterized in that The insurance data is text data, the abnormal insurance sample is abnormal text, and the determining of the loss function of the deep semantic residual network and pre-training the deep semantic residual network based on the abnormal insurance sample with the first abnormal label and the loss function includes: Determine an abnormal text with the first abnormal label based on the abnormal label description event, and determine a normal text corresponding to the abnormal text; Based on the normal text and the abnormal text, a plurality of loss functions of the deep semantic residual network are determined, and the deep semantic residual network is pre-trained based on the abnormal text, the normal sample and the loss function.

4. The method according to claim 3, characterized in that The determining of the transfer learning loss corresponding to the second abnormal label includes: Identifying the abnormal insurance sample with the second abnormal label based on the pre-trained deep semantic residual network to obtain a feature vector, and determining the transfer learning weight of the deep semantic residual network based on the feature vector; A transfer cost function is determined based on the transfer learning network, and global aggregation is performed based on the transfer learning weight and the transfer cost function to obtain the transfer learning loss.

5. The method according to claim 4, characterized in that The step of determining the transfer learning weight of the deep semantic residual network based on the feature vector comprises: Determine the predicted value of the abnormal insurance sample of the second abnormal label obtained in the deep semantic residual network based on the feature vector, and obtain the prediction threshold corresponding to the deep semantic residual network; The transfer learning weight of the deep semantic residual network is determined based on a comparison result between the predicted value and the prediction threshold.

6. The method according to claim 4, characterized in that Determining the transfer cost function based on the transfer learning network includes: Based on the pre-trained deep semantic residual network, text semantic recognition processing is performed on the insurance text matching the second abnormal label to obtain a first text semantic vector; Performing text semantic recognition processing on the insurance text matching the second abnormal label based on the transfer learning network to obtain a second text semantic vector; A migration cost function is determined according to the first text semantic vector and the second text semantic vector.

7. The method according to any one of claims 1 to 6, characterized in that: The extracting target abnormal insurance data from the insurance data according to the data category and the abnormal object library includes: If the data category is a text category, then calling an abnormal object library corresponding to the text category, wherein the abnormal object library stores a plurality of abnormal text objects; Target abnormal text is extracted from the insurance data according to a plurality of abnormal text objects in the abnormal object library and determined as the target abnormal insurance data.

8. The method according to claim 1, characterized in that After determining the abnormal identification result of the insurance data based on the identification result and reporting it, the method further includes: Retrieving a trained processing model, wherein the processing model is obtained by training a long short-term memory model based on a processing behavior sample manually processed on a key database, wherein the key database stores insurance data of different abnormal types; The abnormal identification result is processed based on the processing model to obtain an abnormal insurance data processing result, and the abnormal insurance data processing result is sent to the audit end.

9. A device for identifying abnormal insurance data, characterized in that: include: An acquisition module, used to acquire the insurance data to be identified, and extract target abnormal insurance data from the insurance data according to the data category and the abnormal object library; A processing module, used to perform recognition processing on the target abnormal insurance data based on the abnormal recognition model that has completed model training, to obtain a recognition result of the target abnormal insurance data, wherein the abnormal recognition model is obtained by performing transfer learning training on multiple deep semantic residual networks based on abnormal insurance samples, and the abnormal insurance samples include insurance samples with abnormal labels; A determination module is used to determine the abnormal identification result of the insurance data based on the identification result, and report it to adjust the target abnormal insurance data.

10. A storage medium storing at least one executable instruction, wherein the executable instruction enables a processor to execute an operation corresponding to the method for identifying abnormal insurance data according to any one of claims 1 to 8.

11. A computer device comprising: A processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus; The memory is used to store at least one executable instruction, and the executable instruction enables the processor to execute an operation corresponding to the abnormal insurance data identification method according to any one of claims 1-8.

Images