Investor risk assessment method and related device

Abstract

The invention discloses an investor risk assessment method and a related device, and relates to the technical field of software, and the method comprises the steps: obtaining a first electroencephalogram signal corresponding to a to-be-assessed investor, and carrying out the preprocessing of the first electroencephalogram signal; inputting the preprocessing result of the first electroencephalogram signal into a pre-trained risk classification model to obtain a target risk category corresponding to the to-be-assessed investor output by the risk classification model; wherein the risk classification model is obtained by training a deep learning network based on EEGNet. According to the method, the deep learning network based on the EEGNet is used as a basic model for training to obtain the risk classification model, the electroencephalogram signals of the investor to be assessed are preprocessed, and then the risk category of the investor is predicted through the risk classification model, so that the purpose of accurately assessing the risk bearing capacity of the investor through the electroencephalogram signals is achieved; and an effective basis is provided for subsequent product recommendation.

Description

Technical Field

[0001] The present application relates to the field of software technology, and in particular to an investor risk assessment method and related devices. Background Art

[0002] In banking, providing investors with products that match their risk tolerance can maximize the needs of all parties.

[0003] However, it is difficult to accurately assess investors' risk tolerance at present. In most cases, investors' risk tolerance can only be assessed through subjective judgment and questionnaire surveys. However, these methods are usually not objective and accurate, and are easily affected by irrelevant factors, resulting in deviations in the assessment results. Summary of the invention

[0004] In view of the above problems, this application provides an investor risk assessment method and related devices to achieve the purpose of accurately assessing investor risks. The specific scheme is as follows:

[0005] The first aspect of the present application provides an investor risk assessment method, the investor risk assessment method comprising:

[0006] Acquire a first electroencephalogram signal corresponding to the investor to be evaluated, and pre-process the first electroencephalogram signal;

[0007] The preprocessing result of the first EEG signal is input into a pre-trained risk classification model to obtain the target risk category corresponding to the investor to be evaluated output by the risk classification model; wherein the risk classification model is obtained by training a deep learning network based on EEGNet.

[0008] In a possible implementation, preprocessing the first EEG signal includes:

[0009] The first EEG signal is subjected to filtering processing, baseline correction, artifact removal by independent component analysis, average reference and EEG signal segmentation processing in sequence.

[0010] In a possible implementation, the preprocessing result of the first EEG signal includes a plurality of EEG segments corresponding to the first EEG signal;

[0011] The step of inputting the preprocessing result of the first EEG signal into a pre-trained risk classification model to obtain a target risk category corresponding to the investor to be evaluated output by the risk classification model includes:

[0012] For each EEG segment of the multiple EEG segments, input the EEG segment into the risk classification model to obtain a candidate risk category output by the risk classification model for the EEG segment;

[0013] Classify and count the candidate risk categories corresponding to each EEG segment to obtain the number of EEG segments under different candidate risk categories;

[0014] A candidate risk category with the largest number of EEG segments is selected as the target risk category.

[0015] In a possible implementation, the EEGNet-based deep learning network includes a 2D convolution layer, a depth convolution layer, a depth separable convolution layer and an output layer;

[0016] The process of training the EEGNet-based deep learning network to obtain the risk classification model includes:

[0017] Obtaining sample EEG signals corresponding to the evaluated investors in this iterative training, and preprocessing the sample EEG signals; wherein the sample EEG signals are marked with a first risk category;

[0018] Inputting the preprocessing result of the sample EEG signal into the deep learning network based on EEGNet to achieve: calculating the feature map of the sample EEG signal through the 2D convolution layer; extracting the spatial features in the feature map through the deep convolution layer; decomposing and calculating the spatial features through the deep separable convolution layer; and classifying the decomposition and calculation results of the spatial features through the output layer to obtain a second risk category;

[0019] Calculate the loss function value corresponding to this iterative training according to the first risk category and the second risk category;

[0020] If the loss function value does not meet the preset training end condition, the weight parameters of the 2D convolution layer, the depth convolution layer, the depth separable convolution layer and the output layer are adjusted according to the loss function value, the next iteration training is started, and the step of obtaining the sample EEG signals corresponding to the evaluated investors under this iteration training is returned to be executed;

[0021] If the loss function value satisfies the training end condition, the iterative training is ended, and the EEGNet-based deep learning network after the iterative training is used as the risk classification model.

[0022] In a possible implementation, obtaining sample EEG signals corresponding to the evaluated investors in this iterative training includes:

[0023] Acquiring a second electroencephalogram signal corresponding to the evaluated investor;

[0024] Obtaining a training set and a test set for each round of iterative training from the second EEG signal based on a ten-fold cross validation method;

[0025] During each round of iterative training, the sample EEG signal of this iterative training is obtained from the training set of this round of iterative training;

[0026] Accordingly, the deep learning network based on EEGNet after the iterative training is used as the risk classification model, including:

[0027] The EEGNet-based deep learning network after this iterative training is used as a candidate risk classification model for this round of iterative training;

[0028] Use the test set of this round of iterative training to test the corresponding candidate risk classification model to obtain the model accuracy of this round of iterative training;

[0029] The candidate risk classification model of a round of iterative training with the highest model accuracy is selected as the risk classification model.

[0030] A second aspect of the present application provides an investor risk assessment device, the investor risk assessment device comprising:

[0031] Model training module, used to train the EEGNet-based deep learning network to obtain a risk classification model;

[0032] The risk assessment module is used to obtain the first EEG signal corresponding to the investor to be assessed and preprocess the first EEG signal; input the preprocessing result of the first EEG signal into the risk classification model to obtain the target risk category corresponding to the investor to be assessed output by the risk classification model.

[0033] In a possible implementation, the risk assessment module for preprocessing the first EEG signal is specifically used to:

[0034] The first EEG signal is subjected to filtering processing, baseline correction, artifact removal by independent component analysis, average reference and EEG signal segmentation processing in sequence.

[0035] A third aspect of the present application provides a computer program product, including computer-readable instructions. When the computer-readable instructions are executed on an electronic device, the electronic device implements the investor risk assessment method of the first aspect or any implementation of the first aspect.

[0036] A fourth aspect of the present application provides an electronic device, comprising at least one processor and a memory connected to the processor, wherein:

[0037] The memory is used to store computer programs;

[0038] The processor is used to execute the computer program so that the electronic device can implement the investor risk assessment method of the first aspect or any implementation method of the first aspect.

[0039] The fifth aspect of the present application provides a computer storage medium, which carries one or more computer programs. When the one or more computer programs are executed by an electronic device, the electronic device can implement the investor risk assessment method of the first aspect or any implementation method of the first aspect.

[0040] By means of the above technical scheme, the present application provides an investor risk assessment method and related devices, which obtain the first EEG signal corresponding to the investor to be assessed, and pre-process the first EEG signal; the pre-processing result of the first EEG signal is input into the pre-trained risk classification model to obtain the target risk category corresponding to the investor to be assessed output by the risk classification model; wherein the risk classification model is obtained by training a deep learning network based on EEGNet. The present application uses a deep learning network based on EEGNet as the basic model to train a risk classification model, and predicts the risk category of the investor to be assessed through the risk classification model after pre-processing the EEG signal of the investor to be assessed, thereby achieving the purpose of accurately assessing the risk tolerance of investors with EEG signals, and providing an effective basis for subsequent product recommendations. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The above and other features, advantages and aspects of the embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the accompanying drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and the originals and elements are not necessarily drawn to scale.

[0042] Figure 1 A flowchart of an investor risk assessment method provided in an embodiment of the present application;

[0043] Figure 2 A partial flow chart of an investor risk assessment method provided in an embodiment of the present application;

[0044] Figure 3 A schematic diagram of the structure of a deep learning network based on EEGNet provided in an embodiment of the present application;

[0045] Figure 4 A schematic diagram of another part of the flow chart of an investor risk assessment method provided in an embodiment of the present application;

[0046] Figure 5 A schematic diagram of a processing flow of a deep convolutional layer provided in an embodiment of the present application;

[0047] Figure 6 A schematic diagram of a processing flow of a depth-separable convolutional layer provided in an embodiment of the present application;

[0048] Figure 7 A schematic diagram of the structure of an investor risk assessment device provided in an embodiment of the present application;

[0049] Figure 8 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0050] The following describes the embodiments of the present application in conjunction with the drawings in the embodiments of the present application. The terms used in the implementation method section of the present application are only used to explain the specific embodiments of the present application, and are not intended to limit the present application.

[0051] The embodiments of the present application are described below in conjunction with the accompanying drawings. Those skilled in the art will appreciate that, with the development of technology and the emergence of new scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0052] The terms "first", "second" etc. in the specification of the application and the above-mentioned drawings are used to distinguish similar objects, and need not be used to describe a specific order or sequential order. It should be understood that the terms used in this way can be interchangeable in appropriate circumstances, and this is only to describe the distinction mode adopted by the objects of the same attributes when describing in the embodiments of the application. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, so that the process, method, system, product or equipment comprising a series of units need not be limited to those units, but may include other units that are not clearly listed or inherent to these processes, methods, products or equipment.

[0053] Accurately recommending financial products to investors has always been a pain point for major banks. The risk level classification of financial products has been relatively complete, but there is no more accurate assessment method for investors' risk tolerance at this stage. The risk tolerance of different investors varies greatly. If the recommended products do not match the actual acceptance of investors, it will bring business losses and risks, which will in turn affect investors' confidence and the reputation of the bank. Therefore, it is very important to accurately assess investors' risk tolerance and accurately recommend relevant financial products.

[0054] Human biological characteristics can reflect a lot of important information. There are usually large differences in biological characteristics between different individuals, which provides the possibility of classification based on biological characteristics. Among many biological characteristics, EEG signals are active in various emotion recognition and classification fields due to their sensitivity to emotional changes. Investors' risk tolerance can also be reflected as part of emotions and can be closely related to EEG signals. Therefore, the differences in EEG signals, a biological characteristic, can be used to distinguish investors.

[0055] Based on the above explanation, using EEG signals, a biological feature that is sensitive to emotional changes, to assess investors' risk tolerance will produce more accurate and reliable results. How to use the differences in EEG signals between different individuals as an evaluation criterion is a key issue that needs to be addressed urgently.

[0056] In order to solve the above problems, the embodiment of the present application provides a flowchart of an investor risk assessment method. The investor risk assessment method of the embodiment of the present application is described in detail below in conjunction with the accompanying drawings.

[0057] See also Figure 1 , Figure 1 A flowchart of an investor risk assessment method provided in an embodiment of the present application. Figure 1 As shown, an investor risk assessment method provided in an embodiment of the present application may include steps S101 to S102, and these steps are described in detail below.

[0058] S101, obtaining a first electroencephalogram signal corresponding to the investor to be evaluated, and preprocessing the first electroencephalogram signal.

[0059] In the embodiment of the present application, the EEG signal of the investor to be evaluated when answering the evaluation questions, that is, the first EEG signal, can be acquired by the EEG acquisition device.

[0060] In the process of collecting EEG signals, it is often interfered by various noises or irrelevant component signals, such as the errors of the EEG acquisition equipment itself, the interference caused by the electrical environment in which the equipment is located, and the interference caused by the investor's own posture changes. These errors are mixed in the original EEG signal, which will greatly interfere with the acquisition of effective information. Therefore, the embodiment of the present application uses specific measures to eliminate noise and extract useful information. After preprocessing, it will be cleaner and more real. The quality of the EEG signal will directly affect the classification results of the subsequent EEG signal.

[0061] It should be noted that the electric charge level in the brain is maintained by billions of neurons. Each neuron maintains its resting potential or propagates action potential through ion exchange. When these transmitted ion waves reach the electrodes set on the brain, they can push the electrons on the metal in the electrodes. At this time, the voltage difference caused by this push can be measured. This voltage difference is the EEG signal. Over time, the continuously recorded voltage values ​​form an electroencephalogram.

[0062] S102, inputting the preprocessing result of the first EEG signal into a pre-trained risk classification model to obtain a target risk category corresponding to the investor to be evaluated output by the risk classification model; wherein the risk classification model is obtained by training a deep learning network based on EEGNet.

[0063] In the embodiment of the present application, a deep learning network based on EEGNet (Electroencephalography Net) is used as a basic model to train a risk classification model. After obtaining the preprocessing result of the first EEG signal, it is input into the risk classification model to obtain the risk category of the investor to be evaluated by the risk classification model, that is, the target risk category. It should be noted that the risk category can characterize the risk tolerance of the investor.

[0064] In addition, after obtaining the target risk category corresponding to the investor to be evaluated, the risk tolerance of the investor to be evaluated can be comprehensively analyzed based on the questionnaire score of the investor to be evaluated and the investor's own evaluation, and finally the financial product that best matches his or her risk tolerance can be recommended based on the final evaluation result.

[0065] It should be noted that EEGNet is a compact convolutional neural network for EEG task analysis. It incorporates some classic EEG feature extraction concepts into the neural network model and adds a deep separable convolution module, which greatly reduces the number of parameters. It can extract effective features for different EEG paradigms. When there is only limited training data, EEGNet can still have a high classification performance.

[0066] In a possible implementation, the first EEG signal may be preprocessed using the preprocessing process involved. In this regard, in the embodiment of the present application, the "preprocessing the first EEG signal" in step S10 may be performed using the following steps:

[0067] The first EEG signal is subjected to filtering processing, baseline correction, independent component analysis to remove artifacts, average reference and EEG signal segmentation processing in sequence.

[0068] In the embodiment of the present application, the first EEG signal can be processed in sequence by five steps: filtering, baseline correction, independent component analysis to remove artifacts, average reference, and EEG signal segmentation. Each step is introduced below:

[0069] 1) Filtering. Since the useful information of EEG signals is generally distributed in the range of 0.1 Hz to 100 Hz, the embodiment of the present application uses a high-pass filter to filter out signals less than 0.1 Hz and a low-pass filter to filter out signals greater than 100 Hz. In addition, due to the influence of electrical equipment in the actual acquisition environment, 50 Hz power frequency interference will be generated, and a 50 Hz notch filter can be used to filter out the 50 Hz signal.

[0070] 2) Baseline correction. EEG signals are time-dependent signals, so the signals may drift in time during the acquisition process. These time drifts will change over time, thus affecting the original signals. In this regard, the embodiment of the present application selects EEG signals over a period of time as baseline data, and subtracts the average value of the baseline data from the EEG signals of each data point after the baseline data.

[0071] 3) Independent component analysis to remove artifacts. EEG signals are usually mixed with other biological signals (such as eye movement signals). These mixed biological signals are called artifact signals. As artifact signals, such signals are close to EEG signals and cannot be removed by filtering. In this regard, the embodiment of the present application uses independent component analysis to separate the artifact signals.

[0072] Assume that the collected EEG signals include x1(t), x2(t),…, x n (t), this group of EEG signals can be decomposed into n signal sources s1(t),s2(t),…,s n (t) and an unknown mixing matrix A, each signal source is an independent component, so the reconstruction matrix x = As can be satisfied.

[0073] The process of independent component analysis can include two steps: signal decomposition and artifact removal. The following introduces signal decomposition and artifact removal respectively:

[0074] 1) Signal decomposition: You can manually select different numbers of signal sources n for multiple decompositions. The calculated mixing matrix A is multiplied by the separated signal to obtain the reconstructed matrix x. The reconstructed matrix x is the same as the original matrix x1(t), x2(t),…, x n The mean square error between the reconstructed matrix and the original matrix under different signal source numbers n is calculated, and the decomposition result corresponding to the calculation result with the smallest mean square error is selected as the final result of signal decomposition.

[0075] 2) Artifact Removal Common artifacts in EEG signals include eye movement artifacts, myoelectric noise, and electrocardiographic noise. Each signal decomposed in the previous step is analyzed. For eye movement artifacts, the time domain usually appears as a slow wave with a large amplitude, and the spikes caused by blinking can be clearly identified, and are distributed in the channel signal corresponding to the forehead area. For electromyographic noise, the time domain appears as irregular high-frequency fluctuations, and the frequency domain is usually distributed in the higher frequency band above 20Hz. For electrocardiographic noise, it usually appears as a heartbeat waveform with the same period, and is usually distributed in the channels corresponding to the acquisition electrodes at the back of the head. After removing various artifacts from the separated signal, the filtered signal is multiplied by the mixing matrix A, and the reconstructed matrix x obtained is the final output of this step.

[0076] The purpose of independent component analysis is to estimate the mixing matrix A and the signal source s when only the observed data x is known. After separating the EEG signal and the artifact signal, a threshold is set to filter the artifact signal and retain the EEG signal.

[0077] 4) Average reference. Through the above steps 1) to 3), high-amplitude signals such as eye movement signals have been removed from the EEG signal, which will have a certain impact on the real EEG signal. Therefore, after removing the artifact signal, the EEG signal obtained through the above steps 1) to 3) can be re-referenced according to its average value, that is, the average value of the EEG signal is used as the zero point to readjust the amplitude of the EEG signal.

[0078] 5) EEG signal segmentation. Considering the relevance of EEG signals to external stimuli, EEG signals containing the most information are usually distributed between 1000ms before and after the stimulation point, so the EEG signals within this range are intercepted as EEG segments.

[0079] Thus, through the above steps 1) to 5), the preprocessing result of the first EEG signal includes a plurality of EEG segments corresponding to the first EEG signal. Figure 2 , Figure 2 A partial flow chart of an investor risk assessment method provided in an embodiment of the present application. Figure 2 As shown, an investor risk assessment method provided in an embodiment of the present application, wherein in step S102 "the preprocessing result of the first EEG signal is input into a pre-trained risk classification model to obtain a target risk category corresponding to the investor to be assessed output by the risk classification model", may include steps S201 to S203, and these steps are described in detail below.

[0080] S201, for each EEG segment among the multiple EEG segments, input the EEG segment into a risk classification model to obtain a candidate risk category output by the risk classification model for the EEG segment.

[0081] In an embodiment of the present application, for each EEG segment corresponding to the first EEG signal, the EEG segment can be input into a risk classification model to obtain a risk category output by the risk classification model for the EEG segment, that is, a candidate risk category.

[0082] S202, classifying and counting the candidate risk categories corresponding to each EEG segment to obtain the number of EEG segments under different candidate risk categories.

[0083] S203, selecting a candidate risk category with the largest number of EEG segments as the target risk category.

[0084] In an embodiment of the present application, for multiple EEG segments corresponding to the first EEG signal, after obtaining the corresponding candidate risk categories, a voting method can be used to perform classification and statistics to obtain the number of EEG segments under different candidate risk categories, and then, select the candidate risk category with the largest number of EEG segments as the target risk category.

[0085] See also Figure 3 , Figure 3 A schematic diagram of the structure of a deep learning network based on EEGNet provided in an embodiment of the present application. Figure 3 As shown in Figure 1, the deep learning network based on EEGNet includes 2D convolution layer A, depth convolution layer B, depth separable convolution layer C and output layer D. Figure 4 , Figure 4 Another partial flow chart of an investor risk assessment method provided in an embodiment of the present application. Figure 4 As shown, an investor risk assessment method provided by an embodiment of the present application, wherein a process of obtaining a risk classification model by training a deep learning network based on EEGNet includes steps S401 to S405, and these steps are described in detail below.

[0086] S401, obtaining sample EEG signals corresponding to the evaluated investors in this iterative training, and preprocessing the sample EEG signals; wherein the sample EEG signals are marked with a first risk category.

[0087] In the embodiment of the present application, a risk classification model is obtained by performing multiple iterations of training on a deep learning network based on EEGNet. During any iteration of training, the EEG signals of the evaluated investors under this iteration of training, i.e., sample EEG signals, can be obtained. The sample EEG signals are pre-labeled with risk categories, i.e., the first risk category. Then, the sample EEG signals are pre-processed, wherein the preprocessing process of the sample EEG signals can refer to the preprocessing process of the first EEG signals mentioned above, and the embodiment of the present application will not be repeated here.

[0088] S402, input the preprocessing results of the sample EEG signal into the deep learning network based on EEGNet to achieve: calculating the feature map of the sample EEG signal through the 2D convolution layer; extracting the spatial features in the feature map through the deep convolution layer; decomposing and calculating the spatial features through the deep separable convolution layer; and classifying the decomposition and calculation results of the spatial features through the output layer to obtain the second risk category.

[0089] Continue to see Figure 3 , the preprocessing result of the sample EEG signal is used as the input of the 2D convolution layer A, which consists of a multi-way convolution kernel (i.e. Figure 3 Each convolution kernel a performs 2D convolution calculation on the sample EEG signal to output the feature map of the sample EEG signal.

[0090] Continue to see Figure 3 , the output of 2D convolution layer A is used as the input of deep convolution layer B, which consists of multi-way convolution kernels (i.e. Figure 3 The convolution kernel b in the image is composed of one convolution kernel b corresponding to one convolution kernel a. Each convolution kernel b performs convolution calculation on the feature map output by the corresponding convolution kernel a. Figure 5 , Figure 5 A schematic diagram of a processing flow of a deep convolutional layer provided in an embodiment of the present application. Figure 5 As shown in the figure, assuming that one convolution kernel a outputs a feature map of size 12x 12x 3, and one convolution kernel b contains three convolution kernels of size 5x5x1, then each convolution kernel of size 5x5x1 only performs convolution calculation on one channel of the feature map of size 12x12x3 to obtain a feature map of size 8x8x1, and then the calculation results of the three convolution kernels are stacked up, and finally a feature map of size 8x8x3 is obtained as the spatial feature output. In this way, deep convolution can extract the spatial features of each channel of the EEG signal and reduce the number of parameters that need to be fitted.

[0091] Continue to see Figure 3 , the output of the deep convolution layer B is used as the input of the deep separable convolution layer C. The deep separable convolution layer C is composed of multiple convolution kernel groups. Each convolution kernel group includes a deep convolution kernel (i.e. Figure 3 The convolution kernel c1 in ) and the point-by-point convolution kernel (i.e. Figure 3 The convolution kernel c2 in the figure is a convolution kernel group (including a convolution kernel c1 and a convolution kernel c2) corresponding to a convolution kernel b. Each convolution kernel c1 and convolution kernel c2 decompose and calculate the spatial features of the output of the corresponding convolution kernel b. Figure 6 , Figure 6 A schematic diagram of a processing flow of a depth-separable convolutional layer provided in an embodiment of the present application. Figure 6As shown, for one convolution kernel group, assuming that the convolution kernel c1 of this path is used to process a feature map of size 8x8x3, and the convolution kernel c2 of this path contains a convolution kernel of size 1x1x3, then a convolution kernel of size 1x1x3 is used to perform convolution calculation on the feature map of size 8x8x3, and the spatial features output by the deep convolution layer B can be combined to obtain as much useful information as possible.

[0092] It should be noted that the depthwise separable convolution consists of two parts: depthwise convolution and pointwise convolution. It is essentially a decomposition calculation of the ordinary convolution process. Each convolution kernel of the depthwise convolution only performs convolution calculation on one channel of the input feature map, and finally stacks the calculation results of each channel to obtain the output. The convolution kernel size of the pointwise convolution is 1x1. The convolution calculation is performed on the results of all channels after the depthwise convolution is stacked, and the final output is obtained using the effective information on each channel in the input feature map.

[0093] Continue to see Figure 3 , the output of the depthwise separable convolutional layer C is used as the input of the output layer, and the output layer can classify the decomposition calculation results of the depthwise separable convolutional layer C to obtain the corresponding risk category, that is, the second risk category.

[0094] S403: Calculate the loss function value corresponding to this iterative training according to the first risk category and the second risk category.

[0095] In the embodiment of the present application, the first risk category is taken as the target, and the loss function value of this iterative training is calculated according to the first risk category and the second risk category.

[0096] S404: If the loss function value does not meet the preset training end condition, adjust the weight parameters of the 2D convolution layer, the depth convolution layer, the depth separable convolution layer and the output layer according to the loss function value, enter the next iterative training, and return to execute step S401.

[0097] In an embodiment of the present application, if the loss function value does not meet the preset training end conditions, such as the loss function value does not tend to converge, the weight parameters of the 2D convolution layer, the depth convolution layer, the depth separable convolution layer and the output layer are adjusted in reverse according to the loss function value, and the next iterative training is entered, and the execution of step S401 is returned.

[0098] S405: If the loss function value meets the training end condition, end this iterative training, and use the EEGNet-based deep learning network after this iterative training as a risk classification model.

[0099] In an embodiment of the present application, if the loss function value meets the training end condition, such as the loss function value tends to converge, the iterative training is terminated, and the EEGNet-based deep learning network after the iterative training is used as the risk classification model.

[0100] In a possible implementation, a ten-fold cross validation can be used to select the final classification model during the risk classification model training process. In this regard, in the embodiment of the present application, step S401 "obtaining sample EEG signals corresponding to the evaluated investors under this iterative training" can be performed using the following steps:

[0101] Obtain the second EEG signal corresponding to the evaluated investor; obtain the training set and test set of each round of iterative training from the second EEG signal based on the ten-fold cross-validation method; in the process of each round of iterative training, obtain the sample EEG signal of this iterative training from the training set of this round of iterative training;

[0102] Correspondingly, in the above step S305, "using the deep learning network based on EEGNet after this iteration training as a risk classification model" can be performed by the following steps:

[0103] The EEGNet-based deep learning network after this iterative training is used as the candidate risk classification model for this round of iterative training; the corresponding candidate risk classification model is tested using the test set of this round of iterative training to obtain the model accuracy of this round of iterative training; the candidate risk classification model of the round of iterative training with the highest model accuracy is selected as the risk classification model.

[0104] It should be noted that the basic principle of 10-fold cross validation is to divide the samples to be trained into two parts: training set and validation set, use the training set to train the classification model, and then use the validation set to test the accuracy of the classification model. 10-fold cross validation means dividing the original sample into ten parts, taking turns to use 9 of them as training sets and 1 as validation set, calculating the classification accuracy through the classification model, and finally selecting the model with the highest accuracy as the final classification model.

[0105] In the embodiment of the present application, the EEG signal of the evaluated investor when answering the evaluation question, i.e., the second EEG signal, can be collected by the EEG acquisition device. The second EEG signal is evenly divided into ten parts based on the ten-fold cross-validation method, and nine of them are used as training sets and one as a validation set in turn, thereby obtaining the training set and test set of each round of iterative training in ten rounds of iterative training.

[0106] In each round of iterative training, the training set of this round of iterative training can be used to perform multiple iterative training on the EEGNet-based deep learning network. Therefore, the sample EEG signals under this iterative training can be obtained from the training set of this round of iterative training, and input into the EEGNet-based deep learning network after preprocessing to complete this iterative training. After multiple iterative trainings, the EEGNet-based deep learning network after this iterative training is used as the candidate risk classification model for this round of iterative training. The corresponding candidate risk classification model is tested using the test set of this round of iterative training to obtain the model accuracy of this round of iterative training.

[0107] After ten rounds of iterative training, ten candidate risk classification models and their respective model accuracy rates are obtained, from which the candidate risk classification model of a round of iterative training with the highest model accuracy rate is selected as the final risk classification model.

[0108] Through the above description, in view of the problems that the existing investor risk assessment methods have low assessment accuracy and the assessment results are easily interfered with, the embodiment of the present application provides an investor risk assessment method, which can accurately assess the risk of investors based on their EEG signals when they answer assessment questions. The advantages of this application are:

[0109] 1) EEG signals are used as the basis for risk tolerance assessment. Compared with other biometrics, EEG signals have a stronger correlation with emotional changes and are more credible in assessing investors’ risk tolerance.

[0110] 2) A complete process for risk assessment based on EEG signal processing has been established. This application makes full use of the correlation between EEG signals and investor's emotional changes, and obtains corresponding risk assessment results by analyzing investors' EEG signals. Compared with existing risk assessment methods, it is completely objective and can obtain more accurate assessment results.

[0111] Therefore, the risk tolerance assessment results obtained through this application have stronger anti-interference ability and higher accuracy, and can better solve the problems that need to be solved in the current bank risk assessment method.

[0112] An investor risk assessment method provided by an embodiment of the present application is introduced above, and a device for executing the above investor risk assessment method will be introduced below.

[0113] See also Figure 7 , Figure 7 This is a schematic diagram of the structure of an investor risk assessment device provided in an embodiment of the present application. Figure 6 As shown, an investor risk assessment device provided in an embodiment of the present application includes:

[0114] A model training module 10 is used to train a deep learning network based on EEGNet to obtain a risk classification model;

[0115] The risk assessment module 20 is used to obtain the first EEG signal corresponding to the investor to be assessed and preprocess the first EEG signal; input the preprocessing result of the first EEG signal into the risk classification model to obtain the target risk category corresponding to the investor to be assessed output by the risk classification model.

[0116] In a possible implementation, the risk assessment module 20 for preprocessing the first EEG signal is specifically configured to:

[0117] The first EEG signal is subjected to filtering processing, baseline correction, independent component analysis to remove artifacts, average reference and EEG signal segmentation processing in sequence.

[0118] In a possible implementation, the preprocessing result of the first EEG signal includes a plurality of EEG segments corresponding to the first EEG signal;

[0119] The risk assessment module 20 is used to input the preprocessing result of the first EEG signal into a pre-trained risk classification model to obtain a target risk category of the investor to be assessed output by the risk classification model, and is specifically used to:

[0120] For each of the multiple EEG segments, the EEG segment is input into the risk classification model to obtain the candidate risk category output by the risk classification model for the EEG segment; the candidate risk categories corresponding to each EEG segment are classified and counted to obtain the number of EEG segments under different candidate risk categories; and the candidate risk category with the largest number of EEG segments is selected as the target risk category.

[0121] In one possible implementation, the EEGNet-based deep learning network includes a 2D convolution layer, a depth convolution layer, a depth-wise separable convolution layer, and an output layer;

[0122] The model training module 10 is specifically used for:

[0123] Obtain sample EEG signals corresponding to the evaluated investors under this iterative training, and preprocess the sample EEG signals; wherein the sample EEG signals are marked with the first risk category; input the preprocessing results of the sample EEG signals into the deep learning network based on EEGNet to achieve: calculating the feature map of the sample EEG signals through the 2D convolution layer; extracting the spatial features in the feature map through the deep convolution layer; decomposing and calculating the spatial features through the deep separable convolution layer; classifying the decomposition and calculation results of the spatial features through the output layer to obtain the second risk category; calculating the loss function value corresponding to this iterative training according to the first risk category and the second risk category; if the loss function value does not meet the preset training end condition, adjust the weight parameters of the 2D convolution layer, the deep convolution layer, the deep separable convolution layer and the output layer according to the loss function value, enter the next iterative training, and return to execute the step of obtaining the sample EEG signals corresponding to the evaluated investors under this iterative training; if the loss function value meets the training end condition, end this iterative training, and use the deep learning network based on EEGNet after this iterative training as the risk classification model.

[0124] In a possible implementation, the model training module 10 for obtaining sample EEG signals corresponding to the evaluated investors in this iterative training is specifically used for:

[0125] Obtain the second EEG signal corresponding to the evaluated investor; obtain the training set and test set of each round of iterative training from the second EEG signal based on the ten-fold cross-validation method; in the process of each round of iterative training, obtain the sample EEG signal of this iterative training from the training set of this round of iterative training;

[0126] Accordingly, the model training module 10 for using the EEGNet-based deep learning network after this iterative training as a risk classification model is specifically used for:

[0127] The EEGNet-based deep learning network after this iterative training is used as the candidate risk classification model for this round of iterative training; the corresponding candidate risk classification model is tested using the test set of this round of iterative training to obtain the model accuracy of this round of iterative training; the candidate risk classification model of the round of iterative training with the highest model accuracy is selected as the risk classification model.

[0128] It should be noted that the detailed functions of each module in the embodiment of the present application can be found in the corresponding public part of the above-mentioned investor risk assessment method embodiment, and will not be repeated here.

[0129] The present application also provides an electronic device in an embodiment. Figure 8 , Figure 8A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. The electronic device in the embodiment of the present application may include but is not limited to fixed terminals such as mobile phones, notebook computers, PDAs (personal digital assistants), PADs (tablet computers), desktop computers, etc. Figure 8 The electronic device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0130] like Figure 8 As shown, the electronic device may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 801, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 802 or a program loaded from a storage device 808 to a random access memory (RAM) 803. When the electronic device is powered on, various programs and data required for the operation of the electronic device are also stored in the RAM 803. The processing device 801, the ROM 802, and the RAM 803 are connected to each other via a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0131] Typically, the following devices may be connected to the I / O interface 805: input devices 806 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; output devices 807 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 808 including, for example, a memory card, a hard disk, etc.; and communication devices 809. The communication device 809 may allow the electronic device to communicate with other devices wirelessly or by wire to exchange data. Although Figure 8 An electronic device having various devices is shown, but it should be understood that it is not required to implement or possess all the devices shown. More or fewer devices may be implemented or possessed instead.

[0132] Also provided in an embodiment of the present application is a computer program product, including computer-readable instructions. When the computer-readable instructions are executed on an electronic device, the electronic device implements any one of the investor risk assessment methods provided in the embodiments of the present application.

[0133] A computer-readable storage medium is also provided in an embodiment of the present application. The storage medium carries one or more computer programs. When the one or more computer programs are executed by an electronic device, the electronic device can implement any investor risk assessment method provided in the embodiment of the present application.

[0134] It should also be noted that the device embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed over multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. In addition, in the drawings of the device embodiments provided by the present application, the connection relationship between the modules indicates that there is a communication connection between them, which may be specifically implemented as one or more communication buses or signal lines.

[0135] Through the description of the above implementation mode, the technicians in the field can clearly understand that the present application can be implemented by means of software plus necessary general hardware, and of course, it can also be implemented by special hardware including special integrated circuits, special CPUs, special memories, special components, etc. In general, all functions completed by computer programs can be easily implemented by corresponding hardware, and the specific hardware structure used to implement the same function can also be various, such as analog circuits, digital circuits or special circuits. However, for the present application, software program implementation is a better implementation mode in more cases. Based on such an understanding, the technical solution of the present application is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a readable storage medium, such as a computer floppy disk, a U disk, a mobile hard disk, a ROM, a RAM, a disk or an optical disk, etc., including a number of instructions to enable a computer device (which can be a personal computer, a training device, or a network device, etc.) to execute the methods described in each embodiment of the present application.

[0136] In the above embodiments, all or part of the embodiments may be implemented by software, hardware, firmware or any combination thereof. When implemented by software, all or part of the embodiments may be implemented in the form of a computer program product.

[0137] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from a website site, a computer, a training device, or a data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, training device, or data center. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a training device, a data center, etc. that includes one or more available media integrations. The available medium may be a magnetic medium, (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)), etc.

Claims

1. An investor risk assessment method, characterized in that: The investor risk assessment method includes: Acquire a first electroencephalogram signal corresponding to the investor to be evaluated, and pre-process the first electroencephalogram signal; The preprocessing result of the first EEG signal is input into a pre-trained risk classification model to obtain the target risk category corresponding to the investor to be evaluated output by the risk classification model; wherein the risk classification model is obtained by training a deep learning network based on EEGNet.

2. The investor risk assessment method according to claim 1, characterized in that: The preprocessing of the first EEG signal comprises: The first EEG signal is subjected to filtering processing, baseline correction, artifact removal by independent component analysis, average reference and EEG signal segmentation processing in sequence.

3. The investor risk assessment method according to claim 2, characterized in that: The preprocessing result of the first EEG signal includes a plurality of EEG segments corresponding to the first EEG signal; The step of inputting the preprocessing result of the first EEG signal into a pre-trained risk classification model to obtain a target risk category corresponding to the investor to be evaluated output by the risk classification model includes: For each EEG segment of the multiple EEG segments, input the EEG segment into the risk classification model to obtain a candidate risk category output by the risk classification model for the EEG segment; Classify and count the candidate risk categories corresponding to each EEG segment to obtain the number of EEG segments under different candidate risk categories; A candidate risk category with the largest number of EEG segments is selected as the target risk category.

4. The investor risk assessment method according to claim 1, characterized in that: The EEGNet-based deep learning network includes a 2D convolution layer, a depth convolution layer, a depth separable convolution layer and an output layer; The process of training the EEGNet-based deep learning network to obtain the risk classification model includes: Obtaining sample EEG signals corresponding to the evaluated investors in this iterative training, and preprocessing the sample EEG signals; wherein the sample EEG signals are marked with a first risk category; Inputting the preprocessing result of the sample EEG signal into the deep learning network based on EEGNet to achieve: calculating the feature map of the sample EEG signal through the 2D convolution layer; extracting the spatial features in the feature map through the deep convolution layer; decomposing and calculating the spatial features through the deep separable convolution layer; and classifying the decomposition and calculation results of the spatial features through the output layer to obtain a second risk category; Calculate the loss function value corresponding to this iterative training according to the first risk category and the second risk category; If the loss function value does not meet the preset training end condition, the weight parameters of the 2D convolution layer, the depth convolution layer, the depth separable convolution layer and the output layer are adjusted according to the loss function value, the next iteration training is started, and the step of obtaining the sample EEG signals corresponding to the evaluated investors under this iteration training is returned to be executed; If the loss function value satisfies the training end condition, the iterative training is ended, and the EEGNet-based deep learning network after the iterative training is used as the risk classification model.

5. The investor risk assessment method according to claim 4, characterized in that: The obtaining of sample EEG signals corresponding to the evaluated investors in this iterative training includes: Acquiring a second electroencephalogram signal corresponding to the evaluated investor; Obtaining a training set and a test set for each round of iterative training from the second EEG signal based on a ten-fold cross validation method; During each round of iterative training, the sample EEG signal of this iterative training is obtained from the training set of this round of iterative training; Accordingly, the deep learning network based on EEGNet after the iterative training is used as the risk classification model, including: The EEGNet-based deep learning network after this iterative training is used as a candidate risk classification model for this round of iterative training; Use the test set of this round of iterative training to test the corresponding candidate risk classification model to obtain the model accuracy of this round of iterative training; The candidate risk classification model of a round of iterative training with the highest model accuracy is selected as the risk classification model.

6. An investor risk assessment device, characterized in that: The investor risk assessment device comprises: Model training module, used to train the EEGNet-based deep learning network to obtain a risk classification model; The risk assessment module is used to obtain the first EEG signal corresponding to the investor to be assessed and preprocess the first EEG signal; input the preprocessing result of the first EEG signal into the risk classification model to obtain the target risk category corresponding to the investor to be assessed output by the risk classification model.

7. The investor risk assessment device according to claim 6, characterized in that: The risk assessment module for preprocessing the first EEG signal is specifically used for: The first EEG signal is subjected to filtering processing, baseline correction, artifact removal by independent component analysis, average reference and EEG signal segmentation processing in sequence.

8. A computer program product, characterized in that It includes computer-readable instructions, and when the computer-readable instructions are executed on an electronic device, the electronic device implements the investor risk assessment method as described in any one of claims 1 to 5.

9. An electronic device, characterized in that: The method comprises at least one processor and a memory connected to the processor, wherein: The memory is used to store computer programs; The processor is used to execute the computer program so that the electronic device can implement the investor risk assessment method as described in any one of claims 1 to 5.

10. A computer storage medium, characterized in that: The storage medium carries one or more computer programs, and when the one or more computer programs are executed by an electronic device, the electronic device can implement the investor risk assessment method as described in any one of claims 1 to 5.

Images