Training method and device for insurance policy recognition model, insurance policy recognition method and device

Abstract

The present application provides a training method and device for a policy identification model, and a policy identification method and device, which belong to the field of artificial intelligence technology. The method includes: obtaining first sample data including positive policy samples and negative policy samples; extracting features from the first sample data to obtain target policy features; constructing sample feature data according to the target policy features; training a preset neural network model according to the sample feature data to obtain classification prediction results; performing performance evaluation on the neural network model according to the classification prediction results to obtain performance data; if the performance data is the first data, screening and processing the positive policy samples and the negative policy samples according to the preset quantiles and performance data to obtain second sample data; inputting the second sample data and the classification prediction results into the neural network model for training until the performance data is transformed from the first data to the second data to obtain the policy identification model. The present application can improve the training effect of the model.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a training method and device for an insurance policy recognition model, and an insurance policy recognition method and device. Background Art

[0002] At present, in the training process of common insurance policy recognition models, random sampling is often used to select different types of sample data for model training. This method has the problem of unbalanced input sample data, which often affects the training effect of the model. Therefore, how to improve the training effect of the model has become a technical problem that needs to be solved urgently. Summary of the invention

[0003] The main purpose of the embodiments of the present application is to propose a training method and device for an insurance policy recognition model, and an insurance policy recognition method and device, aiming to improve the training effect of the model.

[0004] To achieve the above-mentioned purpose, a first aspect of an embodiment of the present application proposes a training method for an insurance policy recognition model, the training method comprising:

[0005] Acquire first sample data, wherein the first sample data includes positive policy samples and negative policy samples marked with complaint labels, and the complaint labels are used to mark complaint categories of the first sample data;

[0006] Performing feature extraction on the first sample data to obtain target policy features;

[0007] Constructing sample characteristic data according to the target policy characteristics;

[0008] Training a preset neural network model according to the sample feature data to obtain a classification prediction result of the neural network model;

[0009] Performing a performance evaluation on the neural network model according to the classification prediction result to obtain performance data of the neural network model;

[0010] If the performance data is the first data, the positive policy sample and the negative policy sample are screened according to the preset quantile and the performance data to obtain second sample data;

[0011] The second sample data and the classification prediction result are input into the neural network model for training until the performance data is transformed from the first data to the second data, thereby obtaining a policy recognition model, which is used to identify the complaint category of the target policy, wherein the first data and the second data are used to characterize whether the neural network model meets the training conditions.

[0012] In some embodiments, the step of extracting features from the first sample data to obtain target policy features includes:

[0013] Performing label probability calculation on the first sample data using a preset function and a preset feature label to obtain the label probability of the first sample data;

[0014] Feature extraction is performed on the first sample data according to the label probability and a preset probability threshold to obtain the target policy feature.

[0015] In some embodiments, the step of constructing sample characteristic data according to the target policy characteristics includes:

[0016] Get the preset calculation dimension and sample range;

[0017] The target policy characteristics are formatted within the calculation dimension and sample range to obtain the sample characteristic data.

[0018] In some embodiments, the step of training a preset neural network model according to the sample feature data to obtain a classification prediction result of the neural network model includes:

[0019] Performing classification prediction on the sample feature data by using a preset algorithm of the neural network model to obtain a classification prediction value;

[0020] The classification prediction result is obtained according to the classification prediction value and a preset classification threshold.

[0021] In some embodiments, if the performance data is the first data, the step of screening the positive policy sample and the negative policy sample according to the preset quantile and the performance data to obtain the second sample data includes:

[0022] If the performance data is first data, the first sample data is sorted according to the classification prediction result to obtain a sample data sequence;

[0023] Dividing the sample data sequence into a plurality of sample intervals according to a preset quantile point;

[0024] The positive policy samples and the negative policy samples in the sample interval are screened according to the performance data to obtain the second sample data.

[0025] To achieve the above-mentioned purpose, a second aspect of an embodiment of the present application proposes a policy identification method, the method comprising:

[0026] Obtaining the target insurance policy to be identified;

[0027] The target insurance policy is input into the insurance policy recognition model for recognition processing to obtain a classification probability value of the target insurance policy, wherein the classification probability value is used to characterize the possibility that the target insurance policy contains complaint content; the insurance policy recognition model is trained according to the training method of the insurance policy recognition model as described in the first aspect;

[0028] The target insurance policy is categorized according to the classification probability value to obtain a category label.

[0029] To achieve the above-mentioned purpose, a third aspect of an embodiment of the present application provides a training device for an insurance policy recognition model, the training device comprising:

[0030] A sample data acquisition module, used to acquire first sample data, wherein the first sample data includes positive policy samples and negative policy samples marked with complaint labels, and the complaint labels are used to mark complaint categories of the first sample data;

[0031] A feature extraction module, used to extract features from the first sample data to obtain target policy features;

[0032] A construction module, used to construct sample characteristic data according to the target policy characteristics;

[0033] A training module, used to train a preset neural network model according to the sample feature data to obtain a classification prediction result of the neural network model;

[0034] An evaluation module, used to perform performance evaluation on the neural network model according to the classification prediction result to obtain performance data of the neural network model;

[0035] A screening module, for screening the positive policy sample and the negative policy sample according to a preset quantile and the performance data to obtain second sample data if the performance data is the first data;

[0036] A loop module is used to input the second sample data and the classification prediction result into the neural network model for training until the performance data is transformed from the first data to the second data, thereby obtaining a policy recognition model, wherein the policy recognition model is used to identify the complaint category of the target policy, wherein the first data and the second data are used to characterize whether the neural network model meets the training conditions.

[0037] To achieve the above-mentioned purpose, a fourth aspect of the embodiments of the present application provides a policy identification device, the device comprising:

[0038] The insurance policy acquisition module is used to acquire the target insurance policy to be identified;

[0039] An insurance policy identification module, used for inputting the target insurance policy into an insurance policy identification model for identification processing to obtain a classification probability value of the target insurance policy, wherein the classification probability value is used to characterize the possibility that the target insurance policy contains complaint content; the insurance policy identification model is trained according to the training method of the insurance policy identification model as described in the first aspect;

[0040] The labeling module is used to label the target insurance policy according to the classification probability value to obtain a category label.

[0041] To achieve the above-mentioned purpose, the fifth aspect of an embodiment of the present application proposes an electronic device, which includes a memory, a processor, a program stored in the memory and executable on the processor, and a data bus for realizing connection and communication between the processor and the memory. When the program is executed by the processor, the method described in the first aspect or the second aspect is realized.

[0042] To achieve the above-mentioned purpose, the sixth aspect of an embodiment of the present application proposes a storage medium, which is a computer-readable storage medium used for computer-readable storage, and the storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the method described in the first aspect or the second aspect above.

[0043] The training method and device of the insurance policy identification model, and the insurance policy identification method and device proposed in the present application obtain the first sample data, wherein the first sample data includes positive policy samples and negative policy samples marked with complaint labels, and the complaint labels are used to mark the complaint categories of the first sample data, so that the first sample data can contain multiple types of sample data and improve sample diversity. Furthermore, feature extraction is performed on the first sample data to obtain target policy features, and sample feature data is constructed based on the target policy features, so that feature data used for model training can be easily extracted to reduce the amount of model calculation. Then, the preset neural network model is trained according to the sample feature data to obtain the classification prediction result of the neural network model; and the performance of the neural network model is evaluated according to the classification prediction result to obtain the performance data of the neural network model. This method can more conveniently determine whether the prediction effect of the neural network model meets the training conditions. If the performance data is the first data, the positive policy samples and negative policy samples are screened and processed according to the preset quantiles and performance data to obtain the second sample data. The screening process can effectively reduce the total number of samples and change the proportion of different samples in the total sample data; the second sample data and the classification prediction result are input into the neural network model for training until the performance data is changed from the first data to the second data, and the policy identification model is obtained. The policy identification model is used to identify the complaint category of the target policy, wherein the first data and the second data are used to characterize whether the neural network model meets the training conditions. This method adopts a cyclic iterative model training strategy, which can improve the proportion of positive policy samples in the entire training sample set as a whole by training the neural network model for multiple rounds, thereby solving the problem of sample imbalance in the corresponding business scenario and effectively improving the training effect of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 is a flow chart of a training method for an insurance policy recognition model provided in an embodiment of the present application;

[0045] Figure 2 yes Figure 1 Flow chart of step S102 in FIG.

[0046] Figure 3 yes Figure 1 Flow chart of step S103 in FIG.

[0047] Figure 4 yes Figure 1 Flow chart of step S104 in FIG.

[0048] Figure 5 yes Figure 1 Flow chart of step S106 in FIG.

[0049] Figure 6is a flow chart of the insurance policy identification method provided in an embodiment of the present application;

[0050] Figure 7 It is a structural schematic diagram of a training device for an insurance policy recognition model provided in an embodiment of the present application;

[0051] Figure 8 It is a structural schematic diagram of the insurance policy identification device provided in an embodiment of the present application;

[0052] Fig. 9 It is a schematic diagram of the hardware structure of the electronic device provided in the embodiment of the present application. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0054] It should be noted that, although the functional modules are divided in the device schematic diagram and the logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first", "second", etc. in the specification, claims and the above drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0055] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0056] First, some nouns involved in this application are analyzed:

[0057] Artificial intelligence (AI) is a new technical science that studies and develops theories, methods, technologies and application systems for simulating, extending and expanding human intelligence. AI is a branch of computer science that attempts to understand the essence of intelligence and produce a new type of intelligent machine that can respond in a similar way to human intelligence. Research in this field includes robotics, speech recognition, image recognition, natural language processing and expert systems. AI can simulate the information process of human consciousness and thinking. AI is also a theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0058] Natural language processing (NLP): NLP uses computers to process, understand and apply human languages ​​(such as Chinese, English, etc.). NLP is a branch of artificial intelligence and an interdisciplinary subject between computer science and linguistics. It is often referred to as computational linguistics. Natural language processing includes grammatical analysis, semantic analysis, and text understanding. Natural language processing is often used in technical fields such as 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. It involves data mining related to language processing, machine learning, knowledge acquisition, knowledge engineering, artificial intelligence research, and linguistic research related to language computing.

[0059] Information Extraction: A text processing technology that extracts specified types of entity, relationship, event and other factual information from natural language text and forms structured data output. Information extraction is a technology that extracts specific information from text data. Text data is composed of some specific units, such as sentences, paragraphs, and chapters. Text information is composed of some small specific units, such as characters, words, phrases, sentences, paragraphs, or a combination of these specific units. Extracting noun phrases, names, place names, etc. from text data are all text information extraction. Of course, the information extracted by text information extraction technology can be various types of information.

[0060] Quantile: Also known as quantile, it refers to the numerical point that divides the probability distribution range of a random variable into several equal parts. Commonly used ones are median (ie, bisector), quartile, percentile, etc.

[0061] Wide table: literally means a database table with many fields. Usually refers to a database table that associates indicators, dimensions, and attributes related to a business topic.

[0062] At present, in the training process of common insurance policy recognition models, random sampling is often used to select different types of sample data for model training. This method has the problem of unbalanced input sample data, which often affects the training effect of the model. Therefore, how to improve the training effect of the model has become a technical problem that needs to be solved urgently.

[0063] Based on this, the embodiments of the present application provide a training method and device for an insurance policy recognition model, and an insurance policy recognition method and device, aiming to improve the training effect of the model.

[0064] The training method and device of the insurance policy recognition model, and the insurance policy recognition method and device provided in the embodiments of the present application are specifically illustrated through the following embodiments. First, the training method of the insurance policy recognition model in the embodiments of the present application is described.

[0065] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0066] AI basic technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. AI software technologies mainly include computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0067] The training method of the insurance policy recognition model provided in the embodiment of the present application relates to the field of artificial intelligence technology. The training method of the insurance policy recognition model provided in the embodiment of the present application can be applied to the terminal, can also be applied to the server side, and can also be software running in the terminal or the server side. In some embodiments, the terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, etc.; the server side can be configured as an independent physical server, or it can be configured as a server cluster or a distributed system composed of multiple physical servers, and can also be configured as a cloud server that provides cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms and other basic cloud computing services; the software can be an application of the training method for implementing the insurance policy recognition model, etc., but is not limited to the above forms.

[0068] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0069] Figure 1 is an optional flow chart of the training method of the insurance policy recognition model provided in the embodiment of the present application. Figure 1 The method may include but is not limited to steps S101 to S107.

[0070] Step S101, obtaining first sample data, wherein the first sample data includes positive policy samples and negative policy samples marked with complaint labels, and the complaint labels are used to mark the complaint categories of the first sample data;

[0071] Step S102, extracting features from the first sample data to obtain target policy features;

[0072] Step S103, constructing sample feature data according to the target policy features;

[0073] Step S104, training a preset neural network model according to the sample feature data to obtain a classification prediction result of the neural network model;

[0074] Step S105, performing performance evaluation on the neural network model according to the classification prediction result to obtain performance data of the neural network model;

[0075] Step S106: if the performance data is the first data, the positive policy samples and the negative policy samples are screened according to the preset quantiles and the performance data to obtain the second sample data;

[0076] Step S107, input the second sample data and the classification prediction results into the neural network model for training until the performance data is transformed from the first data to the second data, and obtain the policy recognition model, which is used to identify the complaint category of the target policy, wherein the first data and the second data are used to characterize whether the neural network model meets the training conditions.

[0077] Steps S101 to S107 shown in the embodiment of the present application are obtained by obtaining the first sample data, wherein the first sample data includes positive policy samples and negative policy samples marked with complaint labels, and the complaint labels are used to mark the complaint categories of the first sample data, so that the first sample data can contain multiple types of sample data and improve sample diversity. Feature extraction is performed on the first sample data to obtain target policy features, and sample feature data is constructed based on the target policy features, so that feature data for model training can be easily extracted to reduce the amount of model calculation. The preset neural network model is trained according to the sample feature data to obtain the classification prediction results of the neural network model; and the performance of the neural network model is evaluated according to the classification prediction results to obtain the performance data of the neural network model. This method can more conveniently determine whether the prediction effect of the neural network model meets the training conditions. If the performance data is the first data, the positive policy samples and negative policy samples are screened and processed according to the preset quantiles and performance data to obtain the second sample data. The screening process can effectively reduce the total number of samples and change the proportion of different samples in the total sample data; the second sample data and the classification prediction results are input into the neural network model for training until the performance data is changed from the first data to the second data, and the policy recognition model is obtained. This method adopts a cyclic iterative model training strategy, which can improve the proportion of positive policy samples in the entire training sample set as a whole by training the neural network model for multiple rounds, thereby solving the problem of sample imbalance in the corresponding business scenario and effectively improving the training effect of the model.

[0078] In step S101 of some embodiments, a web crawler can be written, and after the data source is set, data can be crawled in a targeted manner to obtain the first sample data. The first sample data can also be obtained in other ways, not limited to this. Among them, the data source can be a database storing insurance policies or other business platforms, etc., without limitation. The first sample data includes positive insurance policy samples and negative insurance policy samples marked with complaint labels. The complaint label is used to mark the complaint category of the first sample data. The complaint label of the positive insurance policy sample contains the complaint content, and the complaint label of the negative insurance policy sample does not contain the complaint content.

[0079] Furthermore, in order to improve the accuracy of model training, the first sample data within a preset time period can be extracted, where the preset time period can be set according to specific business needs. For example, the first sample data within the past 90 days is obtained, and the policy samples containing complaint content within these 90 days are used as positive policy samples, and the policy samples without complaint content are used as negative policy samples.

[0080] See also Figure 2 In some embodiments, step S102 may include but is not limited to steps S201 to S202:

[0081] Step S201, calculating the label probability of the first sample data by using a preset function and a preset feature label to obtain the label probability of the first sample data;

[0082] Step S202, extracting features from the first sample data according to the label probability and a preset probability threshold to obtain target policy features.

[0083] In step S201 of some embodiments, the preset function may be a softmax function, etc., and the preset feature labels include a variety of different feature types, for example, the preset feature labels include user information changes, user complaint records, the status of the policy agent (employed or resigned), etc. Specifically, the probability of the data content of the first sample data is calculated by the softmax function, and the probability distribution of the first sample data on each preset feature label is created. If the data content of the first sample data contains a record of user information changes, the probability of the current first sample data being user information changes in the preset feature label is high. Therefore, by creating the probability distribution of the first sample data on each preset feature label, the label probability of each first sample data can be more clearly reflected.

[0084] In step S202 of some embodiments, the probability threshold can be set according to actual business needs. In order to increase the amount of data in the training set data of the model, the preset probability threshold can be 0.02, and the first sample data with a label probability greater than or equal to 0.02 is used as the training set data. Further, the label features of the first sample data with a label probability greater than or equal to 0.02 are extracted to obtain the target policy features.

[0085] In addition, in some other embodiments, the first sample data can also be screened based on the experience of business experts to obtain target policy features. For example, sample data containing complaint history records and information change records are selected from the first sample data based on expert experience, and the sample features of this series of sample data are used as target policy features.

[0086] See also Figure 3 In some embodiments, step S103 may include but is not limited to steps S301 to S302:

[0087] Step S301, obtaining a preset calculation dimension and sample range;

[0088] Step S302, convert the format of the target policy characteristics within the calculation dimension and sample range to obtain sample characteristic data.

[0089] In step S301 of some embodiments, the preset calculation dimension may include a policy dimension or a user dimension, etc. The sample range may be determined according to the data source for obtaining the first sample data, the time period for collecting the first sample data, the number of users, etc., without limitation.

[0090] In step S302 of some embodiments, the calculation dimension is selected as the user dimension, the sample range is determined based on the number of users, and the sample range is selected as 100 people. Then, based on the preset judgment basis of positive policy samples and negative policy samples, the users within the sample range are marked as A1, A2, ..., A100 in sequence, and the policy content of the first sample data submitted by each user within a preset time period (for example, within the past 90 days) is analyzed, that is, the target policy features of each user in the past 90 days are formatted and written into a preset table. Based on the target policy features of each user, the user and its corresponding target policy features are marked as positive policy samples or negative policy samples. For example, if at least one of the target policy features c1, c2, ..., c100 of a user A1 contains complaint content, the corresponding position of the user in the "Whether to complain" column is marked as 1, and recorded as a positive policy sample; if the target policy features c1, c2, ..., c100 of a user A2 do not contain complaint content, the corresponding position of the user in the "Whether to complain" column is marked as 0, and recorded as a negative policy sample. In this way, it is relatively convenient to obtain sample feature data based on the target policy features, and the sample feature data can be presented in a table form (such as a wide table), and the sample feature data can be used as training data for the policy recognition model.

[0091] See also Figure 4 In some embodiments, step S104 may include but is not limited to steps S401 to S402:

[0092] Step S401, classifying and predicting the sample feature data using a preset algorithm of the neural network model to obtain a classification prediction value;

[0093] Step S402, obtaining a classification prediction result according to the classification prediction value and a preset classification threshold.

[0094] In step S401 of some embodiments, the preset algorithm of the neural network model may include a decision tree algorithm, a logistic regression algorithm, an ensemble tree algorithm, a deep learning algorithm, etc. Taking the deep learning algorithm as an example, the neural network model is constructed by a deep neural network, and the neural network model includes an input layer, a hidden layer, and an output layer. The input layer of the neural network model is used to extract features of the sample feature data to obtain a sample feature vector, and the hidden layer is used to separate features of the sample feature vector, and features that can be used for classification prediction are separated from the sample feature vector to obtain a classification sample vector. Finally, the activation function of the output layer is used to classify and predict the classification sample vector to obtain a classification prediction value, wherein the activation function includes a Sigmoid function, a Tanh function, a Relu function, etc.

[0095] In step S402 of some embodiments, the classification prediction value can characterize the possibility that the neural network model predicts that the sample feature data belongs to a positive policy sample and a negative policy sample. The preset classification threshold can be set according to actual business needs. For example, the preset classification threshold is 0.5. When the classification prediction value is greater than 0.5, it indicates that the sample feature data is more likely to belong to a positive policy sample. Therefore, the classification prediction result is that the first sample data corresponding to the sample feature data is a positive policy sample. Similarly, when the classification prediction value is less than or equal to 0.5, it indicates that the sample feature data is more likely to belong to a negative policy sample. Therefore, the classification prediction result is that the first sample data corresponding to the sample feature data is a negative policy sample.

[0096] In step S105 of some embodiments, the performance of the neural network model is evaluated according to the classification prediction results to obtain the performance data of the neural network model, wherein the performance data includes precision, cumulative recall and cumulative improvement. The precision mainly refers to the cumulative proportion of positive policy samples. The number of samples predicted to be positive policy samples and actually positive policy samples is recorded as a, the number of samples predicted to be positive policy samples is recorded as m, and the total number of samples actually positive policy samples is recorded as n. Then the precision = a / m, the cumulative recall = a / n. It should be noted that the higher the precision, the better the prediction effect of the model. Therefore, when the precision is less than the preset threshold, it indicates that the prediction effect of the neural network model has not reached the training condition, and the currently obtained performance data is recorded as the first data. When the precision is greater than or equal to the preset threshold, it indicates that the prediction effect of the neural network model has reached the training condition, and the currently obtained performance data is recorded as the second data.

[0097] See also Figure 5 In some embodiments, step S106 may include but is not limited to steps S501 to S503:

[0098] Step S501, if the performance data is first data, sorting the first sample data according to the classification prediction result to obtain a sample data sequence;

[0099] Step S502, dividing the sample data sequence into a plurality of sample intervals according to a preset quantile point;

[0100] Step S503, screening the positive policy samples and negative policy samples in the sample interval according to the performance data to obtain second sample data.

[0101] In step S501 of some embodiments, if the performance data is the first data, it means that the prediction effect of the neural network model does not meet the training conditions, and the first sample data is sorted according to the classification prediction result. Specifically, according to the size of the classification prediction value of the first sample data, the first sample data is arranged in order from high to low according to the classification prediction value to obtain a sample data sequence.

[0102] In step S502 of some embodiments, the preset quantile can be set according to the actual situation. In order to improve the accuracy of sample screening, the preset quantile can be set to 2%. For example, if the collected first sample data is 1 million, the 1 million first sample data are sorted from large to small according to the classification prediction value to obtain a sample data sequence, and then divided into 1 class according to 2% samples, and the 1 million first sample data are divided into 50 equal parts to obtain multiple sample intervals, wherein the quantiles of the sample intervals are 2%, 4%, ..., 44%, ..., 100% respectively.

[0103] In step S503 of some embodiments, the precision, cumulative recall and cumulative improvement of each sample interval relative to the total sample data are calculated respectively according to the first sample data in different sample intervals. For example, taking 1 million first sample data as an example, for the first sample interval (quantile is 2%), the precision of the sample interval is the proportion of positive policy samples in the 20,000 first sample data in the sample interval to the total sample data, and the cumulative recall refers to the proportion of positive policy samples in the 20,000 first sample data in the sample interval to the total number of positive policy samples in the 1 million first sample data. And so on, the precision, cumulative recall and cumulative improvement of each sample interval relative to the total sample data are obtained.

[0104] Furthermore, in the insurance business scenario, there are relatively few positive policy samples (i.e., policy samples containing complaint content). Therefore, in order to make the sample data more balanced, the number of positive policy samples for model training should be expanded, and positive policy samples and negative policy samples in the sample interval with relatively high accuracy should be selected as the second sample data according to the performance data. For example, among the 1 million first sample data mentioned above, positive policy samples and negative policy samples in the sample interval before 44% are selected as the second sample data, including the first sample data in the sample interval with quantiles of 2%, 4%, ..., 44%, and the total number of samples of the second sample data is 440,000.

[0105] Through the above steps S501 to S503, the proportion of positive policy samples in the entire sample data can be increased with a smaller sample loss, so that the number of positive policy samples and negative policy samples in the second sample data is more balanced, thereby improving the training effect of the model.

[0106] In step S107 of some embodiments, the second sample data and the classification prediction results are input into the neural network model for training, and the above-mentioned processing of step S104 and step S105 is repeated until the performance data is transformed from the first data to the second data, so that the accuracy of the neural network model meets the training requirements, and the insurance policy recognition model is obtained, and the insurance policy recognition model is used to identify the complaint category of the target insurance policy. This method adopts a cyclic iterative model training strategy, which can improve the overall proportion of positive insurance policy samples in the entire training sample set by training the neural network model for multiple rounds, thereby solving the problem of sample imbalance in the corresponding business scenario and effectively improving the training effect of the model.

[0107] The training method of the insurance policy recognition model of the embodiment of the present application obtains the first sample data, wherein the first sample data includes positive insurance policy samples and negative insurance policy samples marked with complaint labels, and the complaint labels are used to mark the complaint categories of the first sample data, so that the first sample data can contain multiple types of sample data and improve sample diversity. Furthermore, feature extraction is performed on the first sample data to obtain target insurance policy features, and sample feature data is constructed based on the target insurance policy features, which can easily extract feature data for model training and reduce the amount of model calculation. Then, the preset neural network model is trained according to the sample feature data to obtain the classification prediction result of the neural network model; and the performance of the neural network model is evaluated according to the classification prediction result to obtain the performance data of the neural network model. This method can more conveniently determine whether the prediction effect of the neural network model meets the training conditions. If the performance data is the first data, the positive policy samples and negative policy samples are screened and processed according to the preset quantiles and performance data to obtain the second sample data. The screening process can effectively reduce the total number of samples and change the proportion of different samples in the total sample data; the second sample data and the classification prediction result are input into the neural network model for training until the performance data is changed from the first data to the second data, and the policy identification model is obtained. The policy identification model is used to identify the complaint category of the target policy, wherein the first data and the second data are used to characterize whether the neural network model meets the training conditions. This method adopts a cyclic iterative model training strategy, which can improve the proportion of positive policy samples in the training sample set as a whole by training the neural network model for multiple rounds, thereby solving the problem of sample imbalance in the corresponding business scenario and effectively improving the training effect of the model.

[0108] See also Figure 6 The embodiment of the present application also provides a policy identification method, which includes but is not limited to steps S601 to S603:

[0109] Step S601, obtaining a target insurance policy to be identified;

[0110] Step S602: input the target insurance policy into the insurance policy recognition model for recognition processing to obtain a classification probability value of the target insurance policy, wherein the classification probability value is used to characterize the possibility that the target insurance policy contains complaint content; the insurance policy recognition model is trained according to the training method of the insurance policy recognition model of the first aspect;

[0111] Step S603, classify the target insurance policy according to the classification probability value to obtain a class label.

[0112] In step S601 of some embodiments, a web crawler may be written to crawl data in a targeted manner after setting a data source to obtain a target insurance policy to be identified. The target insurance policy to be identified may also be obtained in other ways, without limitation. The data source may be a data platform for handling insurance business, etc., without limitation.

[0113] In step S602 of some embodiments, the target policy is input into the policy identification model for identification processing to obtain a classification probability value of the target policy, which can characterize the possibility that the target policy contains complaint content. When the classification probability value of the target policy is greater than a preset probability threshold, it indicates that the target policy is more likely to contain a complaint. When the classification probability value of the target policy is less than or equal to the preset probability threshold, it indicates that the target policy is less likely to contain a complaint.

[0114] In step S603 of some embodiments, when the classification probability value of the target policy is greater than the preset probability threshold, it indicates that the target policy is more likely to contain a complaint, and the target policy is marked as a complaint policy; when the classification probability value of the target policy is less than or equal to the preset probability threshold, it indicates that the target policy is less likely to contain a complaint, and the target policy is marked as a normal policy. In this way, complaint policies can be screened out from a large number of target policies, and business personnel can be reminded to handle the complaint policies, effectively improving the accuracy and efficiency of policy identification, while also improving the efficiency of handling complaint policies.

[0115] The insurance policy identification method of the embodiment of the present application can screen out complained insurance policies from a large number of target insurance policies through the insurance policy identification model, and remind business personnel to handle the complained insurance policies, thereby effectively improving the accuracy and efficiency of insurance policy identification, and at the same time can also greatly reduce the cost of manpower investigation and improve the efficiency of business personnel in handling complained insurance policies.

[0116] See also Figure 7 The embodiment of the present application further provides a training device for an insurance policy recognition model, which can implement the training method for the insurance policy recognition model, and the device includes:

[0117] The sample data acquisition module 701 is used to acquire first sample data, wherein the first sample data includes positive policy samples and negative policy samples marked with complaint labels, and the complaint labels are used to mark the complaint categories of the first sample data;

[0118] A feature extraction module 702 is used to extract features from the first sample data to obtain target policy features;

[0119] A construction module 703 is used to construct sample characteristic data according to the target policy characteristics;

[0120] The training module 704 is used to train the preset neural network model according to the sample feature data to obtain the classification prediction result of the neural network model;

[0121] An evaluation module 705 is used to evaluate the performance of the neural network model according to the classification prediction results to obtain performance data of the neural network model;

[0122] A screening module 706 is used for screening the positive policy samples and the negative policy samples according to the preset quantiles and the performance data to obtain second sample data if the performance data is the first data;

[0123] The loop module 707 is used to input the second sample data and the classification prediction results into the neural network model for training until the performance data is transformed from the first data to the second data, and obtain the policy recognition model. The policy recognition model is used to identify the complaint category of the target policy, wherein the first data and the second data are used to characterize whether the neural network model meets the training conditions.

[0124] In some embodiments, the feature extraction module 702 includes:

[0125] A calculation unit, configured to calculate the label probability of the first sample data by using a preset function and a preset feature label to obtain the label probability of the first sample data;

[0126] The extraction unit is used to extract features from the first sample data according to the label probability and a preset probability threshold to obtain target policy features.

[0127] In some embodiments, building block 703 includes:

[0128] An acquisition unit, used to acquire a preset calculation dimension and sample range;

[0129] The conversion unit is used to convert the format of the target policy characteristics within the calculation dimension and sample range to obtain sample characteristic data.

[0130] In some embodiments, the training module 704 includes:

[0131] A prediction unit, used to perform classification prediction on sample feature data through a preset algorithm of a neural network model to obtain a classification prediction value;

[0132] The comparison unit is used to obtain the classification prediction result according to the classification prediction value and the preset classification threshold.

[0133] In some embodiments, the screening module 706 includes:

[0134] a sorting unit, configured to, if the performance data is the first data, sort the first sample data according to the classification prediction result to obtain a sample data sequence;

[0135] A partitioning unit, used to divide the sample data sequence into a plurality of sample intervals according to a preset quantile point;

[0136] The screening unit is used to screen the positive policy samples and the negative policy samples in the sample interval according to the performance data to obtain second sample data.

[0137] The specific implementation of the training device for the insurance policy recognition model is basically the same as the specific implementation of the training method for the above-mentioned insurance policy recognition model, and will not be repeated here.

[0138] See also Figure 8 The embodiment of the present application also provides a policy identification device, which can implement the above-mentioned policy identification method, and the device includes:

[0139] The insurance policy acquisition module 801 is used to acquire the target insurance policy to be identified;

[0140] The insurance policy identification module 802 is used to input the target insurance policy into the insurance policy identification model for identification processing, and obtain a classification probability value of the target insurance policy, wherein the classification probability value is used to characterize the possibility that the target insurance policy contains complaint content; the insurance policy identification model is trained according to the training method of the insurance policy identification model of the first aspect;

[0141] The labeling module 803 is used to label the target insurance policy according to the classification probability value to obtain a category label.

[0142] The specific implementation of the insurance policy identification device is basically the same as the specific implementation of the above-mentioned insurance policy identification method, and will not be repeated here.

[0143] The embodiment of the present application also provides an electronic device, the electronic device comprising: a memory, a processor, a program stored in the memory and executable on the processor, and a data bus for realizing connection and communication between the processor and the memory, wherein when the program is executed by the processor, the training method of the insurance policy recognition model or the insurance policy recognition method described above is realized. The electronic device may be any intelligent terminal including a tablet computer, a vehicle-mounted computer, etc.

[0144] See also Fig. 9 , Fig. 9 The hardware structure of an electronic device of another embodiment is illustrated, and the electronic device includes:

[0145] The processor 901 may be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present application;

[0146] The memory 902 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 902 can store an operating system and other applications. When the technical solution provided in the embodiments of this specification is implemented by software or firmware, the relevant program code is stored in the memory 902, and the processor 901 calls and executes the training method or the policy identification method of the policy identification model of the embodiment of this application;

[0147] Input / output interface 903, used to implement information input and output;

[0148] Communication interface 904, used to realize communication interaction between the device and other devices, which can be realized by wired mode (such as USB, network cable, etc.) or wireless mode (such as mobile network, WIFI, Bluetooth, etc.);

[0149] A bus 905 that transmits information between various components of the device (e.g., the processor 901, the memory 902, the input / output interface 903, and the communication interface 904);

[0150] The processor 901 , the memory 902 , the input / output interface 903 and the communication interface 904 are connected to each other in communication within the device via a bus 905 .

[0151] An embodiment of the present application also provides a storage medium, which is a computer-readable storage medium used for computer-readable storage. The storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the above-mentioned policy recognition model training method or policy recognition method.

[0152] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely disposed relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0153] The training method of the insurance policy recognition model, the insurance policy recognition method, the training device of the insurance policy recognition model, the insurance policy recognition device, the electronic device and the storage medium provided in the embodiments of the present application obtain the first sample data, wherein the first sample data includes positive insurance policy samples and negative insurance policy samples marked with complaint labels, and the complaint labels are used to mark the complaint categories of the first sample data, so that the first sample data can contain multiple types of sample data and improve the sample diversity. Furthermore, feature extraction is performed on the first sample data to obtain the target insurance policy features, and sample feature data is constructed according to the target insurance policy features, so that feature data used for model training can be easily extracted to reduce the amount of calculation of the model. Then, the preset neural network model is trained according to the sample feature data to obtain the classification prediction result of the neural network model; and the performance of the neural network model is evaluated according to the classification prediction result to obtain the performance data of the neural network model. This method can more conveniently determine whether the prediction effect of the neural network model meets the training conditions. If the performance data is the first data, the positive policy samples and negative policy samples are screened and processed according to the preset quantiles and performance data to obtain the second sample data. The screening process can effectively reduce the total number of samples and change the proportion of different samples in the total sample data; the second sample data and the classification prediction result are input into the neural network model for training until the performance data is changed from the first data to the second data, and the policy identification model is obtained. The policy identification model is used to identify the complaint category of the target policy, wherein the first data and the second data are used to characterize whether the neural network model meets the training conditions. This method adopts a cyclic iterative model training strategy, which can improve the proportion of positive policy samples in the entire training sample set as a whole by training the neural network model for multiple rounds, thereby solving the problem of sample imbalance in the corresponding business scenario and effectively improving the training effect of the model.

[0154] The embodiments described in the embodiments of the present application are intended to more clearly illustrate the technical solutions of the embodiments of the present application and do not constitute a limitation on the technical solutions provided in the embodiments of the present application. Those skilled in the art will appreciate that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0155] It can be understood by those skilled in the art that Figure 1-6 The technical solutions shown in the figure do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than those shown in the figure, or a combination of certain steps, or different steps.

[0156] The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0157] Those skilled in the art will appreciate that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices may be implemented as software, firmware, hardware, or a suitable combination thereof.

[0158] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0159] It should be understood that in the present application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0160] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the above units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0161] The units described above as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0162] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0163] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including multiple instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, referred to as ROM), random access memory (Random Access Memory, referred to as RAM), disk or optical disk and other media that can store programs.

[0164] The preferred embodiments of the present invention are described above with reference to the accompanying drawings, but the scope of the rights of the present invention is not limited thereto. Any modification, equivalent substitution and improvement made by a person skilled in the art without departing from the scope and essence of the present invention should be within the scope of the rights of the present invention.

Claims

1. A training method for an insurance policy recognition model, characterized in that: The training method comprises: Acquire first sample data, wherein the first sample data includes positive policy samples and negative policy samples marked with complaint labels, and the complaint labels are used to mark complaint categories of the first sample data; Performing feature extraction on the first sample data to obtain target policy features; Constructing sample characteristic data according to the target policy characteristics; Training a preset neural network model according to the sample feature data to obtain a classification prediction result of the neural network model; Performing a performance evaluation on the neural network model according to the classification prediction result to obtain performance data of the neural network model; If the performance data is the first data, the positive policy sample and the negative policy sample are screened according to the preset quantile and the performance data to obtain second sample data; Inputting the second sample data and the classification prediction result into the neural network model for training until the performance data is transformed from the first data to the second data, thereby obtaining an insurance policy recognition model, wherein the insurance policy recognition model is used to identify the complaint category of the target insurance policy, wherein the first data and the second data are used to characterize whether the neural network model meets the training conditions; The step of training a preset neural network model according to the sample feature data to obtain a classification prediction result of the neural network model includes: The sample feature data is classified and predicted by the preset algorithm of the neural network model to obtain a classification prediction value; the classification prediction result is obtained according to the classification prediction value and a preset classification threshold; wherein the preset algorithm is a deep learning algorithm, the neural network model includes an input layer, a hidden layer and an output layer, and the sample feature data is classified and predicted by the preset algorithm of the neural network model to obtain a classification prediction value, including: extracting features of the sample feature data based on the input layer to obtain a sample feature vector; separating features of the sample feature vector based on the hidden layer to separate features that can be used for classification prediction from the sample feature vector to obtain a classification sample vector; and classifying and predicting the classification sample vector based on the activation function of the output layer to obtain a classification prediction value; If the performance data is the first data, the positive policy sample and the negative policy sample are screened and processed according to the preset quantile and the performance data to obtain the second sample data, including: If the performance data is the first data, the first sample data is sorted according to the classification prediction result to obtain a sample data sequence; the sample data sequence is divided into a plurality of sample intervals according to a preset quantile; the positive policy samples and the negative policy samples in the sample interval are screened according to the performance data to obtain the second sample data, and the performance data includes precision, cumulative recall and cumulative lift; The extracting features of the first sample data to obtain target policy features includes: The label probability of the first sample data is calculated by using a preset function and a preset feature label, and a probability distribution of the first sample data on each preset feature label is created to obtain the label probability of the first sample data; feature extraction is performed on the first sample data whose label probability is greater than or equal to a preset probability threshold to obtain the target policy feature.

2. The training method according to claim 1, characterized in that: The step of constructing sample characteristic data according to the target policy characteristics comprises: Get the preset calculation dimension and sample range; The target policy characteristics are formatted within the calculation dimension and sample range to obtain the sample characteristic data.

3. A method for identifying an insurance policy, characterized in that: The method comprises: Obtaining the target insurance policy to be identified; Input the target insurance policy into the insurance policy recognition model for recognition processing to obtain a classification probability value of the target insurance policy, wherein the classification probability value is used to characterize the possibility that the target insurance policy contains complaint content; the insurance policy recognition model is trained according to the training method of the insurance policy recognition model according to any one of claims 1 to 2; The target insurance policy is categorized according to the classification probability value to obtain a category label.

4. A training device for an insurance policy recognition model, characterized in that: The training device comprises: A sample data acquisition module, used to acquire first sample data, wherein the first sample data includes positive policy samples and negative policy samples marked with complaint labels, and the complaint labels are used to mark complaint categories of the first sample data; A feature extraction module, used to extract features from the first sample data to obtain target policy features; A construction module, used to construct sample characteristic data according to the target policy characteristics; A training module, used to train a preset neural network model according to the sample feature data to obtain a classification prediction result of the neural network model; An evaluation module, used to perform performance evaluation on the neural network model according to the classification prediction result to obtain performance data of the neural network model; A screening module, for screening the positive policy sample and the negative policy sample according to a preset quantile and the performance data to obtain second sample data if the performance data is the first data; a loop module, used for inputting the second sample data and the classification prediction result into the neural network model for training, until the performance data is transformed from the first data into the second data, so as to obtain a policy recognition model, wherein the policy recognition model is used for identifying the complaint category of the target policy, wherein the first data and the second data are used for characterizing whether the neural network model meets the training condition; The step of training a preset neural network model according to the sample feature data to obtain a classification prediction result of the neural network model includes: The sample feature data is classified and predicted by the preset algorithm of the neural network model to obtain a classification prediction value; the classification prediction result is obtained according to the classification prediction value and a preset classification threshold; wherein the preset algorithm is a deep learning algorithm, the neural network model includes an input layer, a hidden layer and an output layer, and the sample feature data is classified and predicted by the preset algorithm of the neural network model to obtain a classification prediction value, including: extracting features of the sample feature data based on the input layer to obtain a sample feature vector; separating features of the sample feature vector based on the hidden layer to separate features that can be used for classification prediction from the sample feature vector to obtain a classification sample vector; and classifying and predicting the classification sample vector based on the activation function of the output layer to obtain a classification prediction value; If the performance data is the first data, the positive policy samples and the negative policy samples are screened and processed according to the preset quantile and the performance data to obtain the second sample data, including: if the performance data is the first data, the first sample data is sorted and processed according to the classification prediction result to obtain a sample data sequence; the sample data sequence is divided into a plurality of sample intervals according to the preset quantile; the positive policy samples and the negative policy samples in the sample interval are screened and processed according to the performance data to obtain the second sample data, and the performance data includes precision, cumulative recall and cumulative lift; The extracting features of the first sample data to obtain target policy features includes: The label probability of the first sample data is calculated by using a preset function and a preset feature label, and a probability distribution of the first sample data on each preset feature label is created to obtain the label probability of the first sample data; feature extraction is performed on the first sample data whose label probability is greater than or equal to a preset probability threshold to obtain the target policy feature.

5. A policy identification device, characterized in that: The insurance policy identification device comprises: The insurance policy acquisition module is used to acquire the target insurance policy to be identified; An insurance policy identification module, used for inputting the target insurance policy into an insurance policy identification model for identification processing to obtain a classification probability value of the target insurance policy, wherein the classification probability value is used to characterize the possibility that the target insurance policy contains complaint content; the insurance policy identification model is trained according to the training method of the insurance policy identification model according to any one of claims 1 to 2; The labeling module is used to label the target insurance policy according to the classification probability value to obtain a category label.

6. An electronic device, characterized in that: The electronic device includes a memory, a processor, a program stored in the memory and executable on the processor, and a data bus for realizing connection and communication between the processor and the memory. When the program is executed by the processor, the steps of the training method as described in any one of claims 1 to 2 or the steps of the insurance policy identification method as described in claim 3 are realized.

7. A storage medium, the storage medium being a computer-readable storage medium, used for computer-readable storage, characterized in that: The storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps of the training method as described in any one of claims 1 to 2 or the steps of the insurance policy identification method as described in claim 3.

Images