Financial product recommendation method and device, computer equipment and storage medium

Abstract

The invention relates to the technical field of data processing, and particularly discloses a financial product recommendation method and device, computer equipment and a storage medium, and the financial product recommendation method comprises the steps: obtaining a life picture of a user, and obtaining a first data feature according to the life picture; acquiring historical investment data of the user, and performing feature extraction processing on the historical investment data to obtain a second data feature; performing feature fusion on the first data feature and the second data feature to obtain a fused data feature; and obtaining a classification result corresponding to the fused data features based on a KNN algorithm, and determining a financial product corresponding to the classification result as a target financial product. Factors such as life styles and consumption habits of the users are combined with the investment risk bearing capacity of the users to recommend the financial products, so that the accuracy of recommending the financial products is improved.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a financial product recommendation method, device and storage medium. Background Art

[0002] With the continuous development and innovation of the financial market, recommending personalized financial products to users has become an important issue in the field of financial services. At present, financial institutions usually assess users' investment risk tolerance based on questionnaires and financial status forms filled out by users themselves, and then recommend corresponding financial products. However, this traditional assessment method has many limitations.

[0003] On the one hand, the information filled in by users may be inaccurate or incomplete. Users may provide information that is inconsistent with the actual situation due to unclear understanding of their own financial situation or subjective cognitive bias. For example, when filling in the asset status, they may overestimate or underestimate their wealth level, and when describing their risk preferences, they may be affected by their current emotions or the opinions of others and fail to accurately express their true attitude.

[0004] On the other hand, existing evaluation methods often rely only on numerical financial data and lack a comprehensive consideration of the actual living conditions of users. Financial investment is not just a numbers game. The user's lifestyle, consumption habits and other factors also greatly affect their risk tolerance.

[0005] To sum up, the existing financial product recommendation methods have problems such as low accuracy and lack of comprehensiveness. Summary of the invention

[0006] Based on this, the purpose of the present invention is to provide a financial product recommendation method, device, computer equipment and storage medium, which adopts image recognition technology to identify the user's life pictures, infer the user's accurate economic strength, and comprehensively evaluate the user's investment risk tolerance in combination with the user's financial data to recommend suitable financial products to the user.

[0007] In a first aspect, the present invention provides a financial product recommendation method, the method comprising:

[0008] Acquire a user's life picture, and obtain a first data feature based on the life picture; the first data feature represents the user's risk tolerance amount; the life picture includes: at least one of a living environment picture, a travel vehicle picture, a person's clothing picture, and a dining environment picture;

[0009] Acquire the user's historical investment data, perform feature extraction processing on the historical investment data, and obtain a second data feature; the historical investment data includes the name of the user's historical investment target and the historical profit and loss data corresponding to the name of the historical investment target; the second feature data is used to characterize the user's risk tolerance ratio;

[0010] Performing feature fusion on the first data feature and the second data feature to obtain a fused data feature;

[0011] Based on the KNN algorithm, the classification results corresponding to the fused data features are obtained, and the financial products corresponding to the classification results are determined as the target financial products.

[0012] In a second aspect, the present invention provides a financial product recommendation device, the device comprising:

[0013] A picture acquisition module is used to acquire a user's life picture and obtain a first data feature based on the life picture; the first data feature represents the user's risk tolerance amount; the life picture includes: at least one of a living environment picture, a travel vehicle picture, a person's clothing picture, and a dining environment picture;

[0014] A feature extraction module is used to obtain the user's historical investment data, perform feature extraction processing on the historical investment data, and obtain a second data feature; the historical investment data includes the name of the user's historical investment target and the historical profit and loss data corresponding to the name of the historical investment target; the second feature data is used to characterize the user's risk tolerance ratio;

[0015] A feature fusion module, used to fuse the first data feature and the second data feature to obtain a fused data feature;

[0016] The product recommendation module is used to obtain the classification results corresponding to the fusion data features based on the KNN algorithm, and determine the financial products corresponding to the classification results as the target financial products.

[0017] In a third aspect, the present invention provides a computer device, comprising:

[0018] one or more processors;

[0019] A memory for storing one or more programs;

[0020] When one or more programs are executed by one or more processors, the one or more processors implement the financial product recommendation method in the first aspect.

[0021] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the financial product recommendation method in the first aspect.

[0022] The present invention provides a financial product recommendation method, device, computer equipment and storage medium, which obtains a user's life picture and obtains a first data feature based on the life picture; obtains the user's historical investment data, performs feature extraction processing on the historical investment data, and obtains a second data feature; performs feature fusion on the first data feature and the second data feature to obtain a fused data feature; obtains a classification result corresponding to the fused data feature based on a KNN algorithm, and determines that the financial product corresponding to the classification result is a target financial product. Financial products are recommended by combining factors such as the user's lifestyle, consumption habits and the user's investment risk tolerance, thereby improving the accuracy of recommended financial products. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 A flowchart of a financial product recommendation method provided by an embodiment of the present invention;

[0024] Figure 2 A schematic diagram of the steps of obtaining a first data feature in an optional embodiment of the present invention;

[0025] Figure 3 A schematic diagram of the steps of training an image feature extraction model in an optional embodiment of the present invention;

[0026] Figure 4 A schematic diagram of the steps of training a deep reinforcement learning model in an optional embodiment of the present invention;

[0027] Figure 5 A schematic diagram of the steps of obtaining a target financial product based on a KNN algorithm in an optional embodiment of the present invention;

[0028] Figure 6 A schematic diagram of the structure of a financial product recommendation device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0029] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0030] refer to Figure 1, which shows a flow chart of a financial product recommendation method provided by an embodiment of the present invention. This embodiment uses the method applied to a server as an example for illustration. It is understandable that the method can also be applied to a terminal, and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0031] S101. Acquire a user's life picture, and obtain a first data feature based on the life picture.

[0032] Specifically, users can upload life pictures to the server through mobile terminals, and the server receives and stores these life pictures. At the same time, in order to ensure that the format and size of the pictures are within the processable range, non-standard format pictures can be converted and oversized pictures can be appropriately compressed.

[0033] The life pictures uploaded by the server and received by the user may be at least one of the following: living environment pictures, travel vehicle pictures, people's clothing pictures and dining environment pictures. These life pictures are identified by using image recognition technology. For living environment pictures: the type of house (such as apartment, villa, etc.), decoration style and grade can be identified. Luxurious decorated villas may indicate a higher risk tolerance amount, while ordinary apartments may correspond to a lower risk tolerance amount. The surrounding environment can also be analyzed, such as whether it is located in a high-end community, whether there are supporting high-end facilities, etc. For travel vehicle pictures: the brand and model of the vehicle can be identified. High-end car brands and luxury models usually mean that the user has a higher economic strength and a possible high risk tolerance amount. The newness and maintenance of the vehicle can also be considered. For people's clothing pictures: the brand and style of clothing can be identified. Famous brand clothing and high-end custom clothing may indicate that the user has a higher consumption ability and risk tolerance amount. The occasion suitability of clothing can also be identified, such as the difference between high-end dresses for formal occasions and daily casual clothing. For dining environment pictures: the grade and type of the restaurant can be identified. High-end restaurants may indicate that the user has a higher living standard and risk tolerance amount. At the same time, the dishes and tableware on the table can be analyzed to judge the consumption level.

[0034] According to the analysis results, the user's risk tolerance amount characteristic value is determined, and the analysis results of different types of pictures can be combined to obtain a comprehensive risk tolerance amount characteristic value.

[0035] S102: Obtain historical investment data of the user, perform feature extraction processing on the historical investment data, and obtain a second data feature.

[0036] Specifically, the user can upload historical investment data to the server through the mobile terminal, and the server receives and stores the historical investment data. The data may include the name of the user's historical investment target and the historical profit and loss data corresponding to the target. The second characteristic data can be used to characterize the user's risk tolerance ratio.

[0037] For each historical investment target, calculate its profit and loss ratio. For example, if the user's total investment in a target is 10,000 yuan and the profit is 2,000 yuan, the profit ratio is 2,000 / 10,000 = 0.2. If the loss is 1,000 yuan, the loss ratio is 1,000 / 10,000 = 0.1. Consider the profit and loss ratios of all historical investment targets and obtain the user's risk tolerance ratio characteristic value. The average, median or weighted average of the profit ratio and loss ratio can be calculated as the risk tolerance ratio characteristic value; or a model can be trained, and these historical investment data are input into the model, and the risk tolerance ratio characteristic value is obtained through the model algorithm.

[0038] S103: Fusing the first data feature and the second data feature to obtain a fused data feature.

[0039] Specifically, the first data feature (the risk tolerance amount feature value) and the second data feature (the risk tolerance ratio feature value) are fused. Exemplarily, a simple concatenation method can be used to combine the two feature values ​​into a vector. For example, if the risk tolerance amount feature value is 61 and the risk tolerance ratio feature value is 0.3, the fused feature vector is (61, 0.3).

[0040] S104. Obtain a classification result corresponding to the fused data feature based on the KNN algorithm, and determine that the financial product corresponding to the classification result is the target financial product.

[0041] Specifically, a large amount of user data with labeled classification results can be collected. Each user's data includes fused data features obtained based on life pictures and historical investment data, as well as the corresponding financial product classification results. The data is preprocessed, such as standardization and normalization, to ensure that different features are comparable in value.

[0042] For the user to whom the financial product is to be recommended, the fusion data features are used as input. The feature distance between the user and other users in the training data is calculated. The Euclidean distance, Manhattan distance, etc. can be used to select the K neighbor users with the closest distance. The classification result of the user to be recommended is determined based on the financial product classification results of these K neighbor users. The majority voting method can be used, that is, the financial product classification with the most occurrences among the K neighbors is selected as the classification result of the user to be recommended.

[0043] According to the classification results, the corresponding financial products are found from the financial product library as target financial products and recommended to users. At the same time, detailed information about the target financial products can be provided to users, such as product features, risk level, expected returns, etc., to help users make decisions.

[0044] By using the above method, financial products are recommended by combining factors such as the user's lifestyle, consumption habits, and the user's investment risk tolerance, thereby improving the accuracy of recommended financial products.

[0045] In an optional embodiment, obtaining a first data feature based on life pictures includes the following steps: based on a trained image feature extraction model, obtaining comprehensive data features of living environment pictures, travel vehicle pictures, people's clothing pictures and dining environment pictures based on life pictures as the first data feature.

[0046] Specifically, the server starts the image feature extraction model and inputs the received life pictures into the model. The model processes the living environment pictures, travel vehicle pictures, people's clothing pictures and dining environment pictures respectively, and extracts their respective feature information. The model integrates and synthesizes the features of these different types of pictures to obtain a comprehensive data feature, which represents the overall feature information of this group of life pictures. The server receives the comprehensive data feature output by the model as the first data feature, and prepares for subsequent processing or storage.

[0047] refer to Figure 2 In an optional embodiment, based on the trained image feature extraction model, comprehensive data features of living environment pictures, travel vehicle pictures, people's clothing pictures and dining environment pictures are obtained according to life pictures as the first data features, including the following steps:

[0048] S201 . Perform a first preprocessing on each life picture to obtain an image to be extracted corresponding to each life picture.

[0049] Specifically, the server performs a first preprocessing operation on each received life picture. This may include adjusting the image size, normalizing the color, removing noise, etc., converting it into a format suitable for input into the image feature extraction model, and obtaining the image to be extracted corresponding to each life picture.

[0050] S202: Input the image to be extracted into a trained image feature extraction model to obtain feature vectors corresponding to each life picture.

[0051] Specifically, the server inputs the images to be extracted into the trained image feature extraction model one by one. The model analyzes each image, extracts unique feature information, and outputs it in the form of a feature vector. In this way, the server obtains the feature vector corresponding to each life picture.

[0052] S203: Fusing the feature vectors by weighted summation to obtain comprehensive feature data as the first data feature.

[0053] Specifically, the server assigns different weights to each feature vector according to the importance of different life pictures. The importance can be preset or determined by a certain evaluation method. Then, the server uses the weighted summation method to fuse each feature vector to obtain a comprehensive feature data, which is used as the first data feature for subsequent analysis and processing. For example, the weight of the living environment is 0.3, the travel vehicle is 0.3, the character clothing is 0.2, and the dining environment is 0.2. If the living environment analysis result corresponds to a medium risk tolerance amount, the score is 50; the travel vehicle is a high risk tolerance amount, the score is 80; the character clothing is a medium to high risk tolerance amount, the score is 60; the dining environment is a medium risk tolerance amount, the score is 50. Then the comprehensive risk tolerance amount feature value is 0.3×50+0.3×80+0.2×60+0.2×50=61.

[0054] refer to Figure 3 In an optional embodiment, the training method of the image feature extraction model includes the following steps:

[0055] S301. Obtain a life picture data set and labels corresponding to the life pictures to obtain a labeled data set.

[0056] Specifically, first, the server obtains a large number of life picture data sets, which can come from various channels, such as user uploads, database storage, etc. For each life picture, professionals will label it with the corresponding risk tolerance amount according to certain standards. In this way, a labeled data set is obtained.

[0057] Then, the server divides the labeled data set into training set, validation set and test set. Usually, it can be divided according to a certain ratio, such as 70% as training set, 20% as validation set, and 10% as test set. The training set is used for model training, the validation set is used to evaluate the performance of the model and adjust the hyperparameters during the training process, and the test set is used to evaluate the final performance of the model after the model training is completed.

[0058] S302: Initialize the parameters of the model using a random initialization method.

[0059] Specifically, the server uses random initialization to assign initial values ​​to the parameters of the image feature extraction model. This is because random initialization can break the symmetry, making the model perform better during training.

[0060] S303, perform cyclic training on the model; wherein the steps of cyclic training include: step 1, randomly extract a batch of samples from the training set, input the samples into the model, and obtain the prediction results; step 2, calculate the loss value between the prediction result and the true label according to the loss function; step 3, use the optimization algorithm to update the parameters of the model according to the loss value.

[0061] Specifically, the server randomly extracts a batch of samples from the training set, assuming the batch size is n. These samples will be input into the model for training.

[0062] The model processes the input samples and obtains the prediction results through a series of operations such as convolution, pooling, and full connection. Assume that for a sample image, the risk tolerance amount predicted by the model is y pred .

[0063] The server calculates the loss value between the prediction result and the true label according to the pre-set loss function. Here, the mean square error (MSE) loss function can be used, and its calculation formula is:

[0064]

[0065] Where n is the batch size, y true,i is the true risk tolerance label of the i-th sample, y pred,i is the risk tolerance amount predicted by the model for the i-th sample.

[0066] The server uses an optimization algorithm, such as Stochastic Gradient Descent (SGD), to update the model parameters based on the calculated loss value. The update formula of SGD is:

[0067]

[0068] Among them, θ is the parameter of the model, α is the learning rate, is the gradient of the loss function with respect to the parameters.

[0069] S304: After each training round, the performance of the model is evaluated based on the validation set. If the performance of the model on the validation set does not improve any more, the cyclic training is stopped to obtain a trained feature extraction model.

[0070] Specifically, after each training round, the server evaluates the performance of the model based on the validation set. The specific method is to calculate the accuracy of the model on the validation set. The accuracy calculation formula is:

[0071]

[0072] Among them, Accuracy is the accuracy, N is the number of samples predicted correctly by the model on the validation set, and M is the total number of samples in the validation set.

[0073] If the accuracy of the model on the validation set does not improve any more, the server stops the cyclic training, and the model obtained at this time is the trained feature extraction model. This can avoid overfitting of the model and improve the generalization ability of the model.

[0074] In an optional embodiment, performing feature extraction processing on historical investment data to obtain a second data feature includes the following steps:

[0075] Performing a second preprocessing on the historical profit and loss data corresponding to each historical investment target name to obtain a plurality of data to be processed;

[0076] The multiple data to be processed are respectively input into the trained deep reinforcement learning model to obtain multiple second data features corresponding to the name of each historical investment target.

[0077] Specifically, the server first obtains historical investment data, including the names of each historical investment target and the corresponding historical profit and loss data. The server performs a second preprocessing on the historical profit and loss data corresponding to each historical investment target name. This may include data cleaning, removing outliers, missing values, etc.; data normalization, unifying the data into a specific range or distribution for better processing by the model. After preprocessing, multiple data to be processed are obtained.

[0078] The server inputs multiple data to be processed into the trained deep reinforcement learning model. After being trained with a large amount of data, the model can effectively extract features from the input data. The model processes each data to be processed, analyzes the pattern and trend of historical profit and loss data, and finally obtains multiple second data features corresponding to the name of each historical investment target. These second data features can reflect the risk, return and other characteristics of each historical investment target, providing an important basis for subsequent investment decision analysis.

[0079] refer to Figure 4 In an optional embodiment, the training process of the deep reinforcement learning model includes the following steps:

[0080] S401. Obtain a historical profit and loss data set of users of the trading platform through the trading platform, perform a second preprocessing on the historical profit and loss curve data in the historical profit and loss data set, and obtain test data.

[0081] S402. Build a deep reinforcement learning model.

[0082] S403: Input the training data into the deep reinforcement learning model to train the deep reinforcement learning model.

[0083] S404: Input the verification data into the trained deep reinforcement learning model to optimize the parameters of the deep reinforcement learning model.

[0084] Specifically, the server obtains the user's historical profit and loss data set from the trading platform. For each sample, it contains stock data, fund data, and bond data and the corresponding profit and loss values, as well as information such as the user's investment ratio in different investment products.

[0085] Calculate the volatility of different investment products in a specific time period. Taking stocks as an example, suppose the returns of stocks in T time periods are R 1,t , then the average rate of return

[0086] The formula for calculating the volatility of a stock is: Similarly, the volatility σ2 of the fund and the volatility σ3 of the bond can be calculated.

[0087] Construct a comprehensive risk index, set the user's investment ratio in stocks, funds, and bonds as p1, p2, and p3 respectively, then the comprehensive risk index can be expressed as: RI = p1×σ1+p2×σ2+p3×σ3. After preprocessing, the data is divided into training data and verification data to prepare for subsequent model training and optimization.

[0088] When building a deep reinforcement learning model, the state representation s should include not only conventional market information, but also features such as the user's historical risk tolerance index. The action selection strategy π(s) can include operations such as adjusting the proportion of different investment products. The reward calculation function r(s,a) can be set according to the following formula:

[0089] r(s,a)=α×A+β×B

[0090] Among them, A represents the investment return, for example, it can be the difference between the total value of the current investment portfolio and the initial investment value. B represents the matching degree of the risk tolerance ratio, which can be defined as the degree of proximity between the risk level of the actual investment portfolio and the risk tolerance ratio expected by the user. For example, it can be expressed as 1-|RI1-RI2|, where RI1 is the comprehensive risk indicator of the actual investment portfolio, and RI2 is the comprehensive risk indicator corresponding to the risk tolerance ratio expected by the user. α and β are weight coefficients used to balance the importance of return and risk matching in rewards.

[0091] The training data is input into the deep reinforcement learning model. At each time step t, the model calculates the state s according to the current state s. t Select an action t , that is, adjust the proportion of different varieties in the investment portfolio. Then get the new state s based on environmental feedback t+1 and reward r t The goal of the model is to maximize the cumulative reward where γ is a discount factor that controls the importance of future rewards.

[0092] The model can be trained using the policy gradient algorithm, and its update formula is:

[0093]

[0094] Among them, θ is the parameter of the model, α is the learning rate, and J(θ) is the objective function, which can be defined as the expectation of the cumulative reward E[R]. By continuously adjusting the parameter θ, the model gradually optimizes its strategy to obtain higher cumulative rewards on the training data.

[0095] Input the verification data into the trained deep reinforcement learning model. Evaluate the generalization ability of the model by calculating the performance indicators of the model on the verification data. For example, you can calculate the average cumulative reward Where N is the number of samples of the verification data, R i is the cumulative reward of the i-th sample. At the same time, the accuracy of the risk tolerance ratio can also be calculated, for example, using Indicates that Accuracy i It is the degree of match between the actual portfolio risk level in the ith sample and the user's expected risk tolerance ratio, which can be measured by the matching degree B of the risk tolerance ratio mentioned above.

[0096] If the performance is not ideal, you can adjust the model parameters, such as the learning rate α, discount factor γ, etc., to optimize the model's ability to characterize the user's risk tolerance ratio. At the same time, you can also adjust the model's structure or hyperparameters based on the verification results to improve the model's accuracy and stability.

[0097] refer to Figure 5In an optional embodiment, obtaining a classification result corresponding to the fusion data feature based on the KNN algorithm and determining the financial product corresponding to the classification result as the target financial product includes the following steps:

[0098] S501. Obtain a fused feature data set with labeled classifications, where the fused feature data set includes fused data feature samples and corresponding financial product classification labels.

[0099] S502: Take the fused feature data as a new sample and use the Euclidean distance calculation formula to calculate the distance d between the new sample and each sample in the fused feature data set:

[0100]

[0101] Among them, x is the feature vector of the new sample, y is the feature vector of the sample in the fused feature dataset, and n is the dimension of the feature.

[0102] S503: Determine the optimal K value.

[0103] S504: According to the calculated distance d, select the K samples with the closest distance as neighbor samples.

[0104] S505: Count the number of each category in the neighboring samples, classify the new sample into the category with the largest number, and obtain the classification result of the new sample.

[0105] S506: Based on the preset mapping relationship between the classification results and the financial products, determine the target financial product according to the classification results.

[0106] Specifically, we need a labeled classification dataset that contains a large number of fused data feature samples and corresponding financial product classification labels. These samples can be collected from historical data and their classification can be determined by manual labeling or other reliable methods.

[0107] For the new fused data feature (i.e. the user data feature to be classified), calculate its distance from each sample in the data set. There are many ways to calculate the distance, such as Euclidean distance and Manhattan distance. For example, the Euclidean distance calculation formula is:

[0108]

[0109] Among them, x is the feature vector of the new sample, y is the feature vector of the sample in the fused feature dataset, and n is the dimension of the feature.

[0110] The K value is a key parameter that determines how many neighbor samples are considered during classification. The choice of the K value needs to be adjusted according to the specific problem and data set. The optimal K value can be determined by methods such as cross-validation. If the K value is too small, it may be affected by noise data, resulting in unstable classification results; if the K value is too large, the classification results may be too smooth and some local feature information may be lost.

[0111] According to the calculated distance, the K samples with the closest distance are selected as neighbors. The classification labels of these neighbor samples will be used to determine the classification of the new sample.

[0112] Count the number of each category in these K nearest neighbor samples, and classify the new sample into the category with the largest number. For example, if K = 5, three of the nearest neighbor samples belong to category A and two of the nearest neighbor samples belong to category B, then the new sample is classified into category A.

[0113] Each classification result corresponds to one or more financial products. This is a pre-set mapping relationship when establishing the recommendation system. For example, classification A may correspond to financial products such as stock funds and high-risk bonds; classification B may correspond to financial products such as mixed funds and medium-risk bonds.

[0114] Once the classification results of the new samples are determined, the corresponding financial products can be determined as target financial products based on the mapping relationship. If a classification corresponds to multiple financial products, they can be further screened and recommended based on other factors, such as product yield, risk rating, liquidity, etc.

[0115] refer to Figure 6 , is a financial product recommendation device 600 provided by an embodiment of the present invention, the device comprising:

[0116] The picture acquisition module 601 is used to acquire the user's life pictures and obtain the first data feature based on the life pictures; the first data feature is used to characterize the user's risk tolerance amount; the life pictures include: at least one of: living environment pictures, travel vehicle pictures, people's clothing pictures and dining environment pictures.

[0117] Feature extraction module 602 is used to obtain the user's historical investment data, perform feature extraction processing on the historical investment data, and obtain a second data feature; the historical investment data includes the name of the user's historical investment target and the historical profit and loss data corresponding to the name of the historical investment target; the second feature data is used to characterize the user's risk tolerance ratio.

[0118] The feature fusion module 603 is used to fuse the first data feature and the second data feature to obtain a fused data feature.

[0119] The product recommendation module 604 is used to obtain the classification result corresponding to the fusion data feature based on the KNN algorithm, and determine the financial product corresponding to the classification result as the target financial product.

[0120] Optionally, the picture acquisition module 601 includes an image processing unit, which is used to obtain comprehensive data features of living environment pictures, travel vehicle pictures, people's clothing pictures and dining environment pictures according to life pictures based on a trained image feature extraction model, as the first data feature.

[0121] Optionally, the image processing unit is further configured to:

[0122] Performing a first preprocessing on each life picture to obtain an image to be extracted corresponding to each life picture;

[0123] The image to be extracted is input into the trained image feature extraction model to obtain the feature vectors corresponding to each life picture.

[0124] The eigenvectors are fused by a weighted summation method to obtain comprehensive feature data as the first data feature; wherein different weights are assigned to different eigenvectors according to the importance of the life pictures corresponding to different eigenvectors.

[0125] Optionally, the feature extraction module 602 includes:

[0126] A preprocessing unit, used for performing a second preprocessing on the historical profit and loss data corresponding to each historical investment target name to obtain a plurality of data to be processed;

[0127] The feature extraction unit is used to input multiple data to be processed into the trained deep reinforcement learning model respectively to obtain multiple second data features corresponding to the name of each historical investment target.

[0128] Optionally, the product recommendation module 604 includes:

[0129] A sample setting unit is used to obtain a fused feature data set with labeled classifications, where the fused feature data set includes fused data feature samples and corresponding financial product classification labels;

[0130] The calculation unit is used to use the fused feature data as a new sample and use the Euclidean distance calculation formula to calculate the distance d between the new sample and each sample in the fused feature data set:

[0131]

[0132] Among them, x is the feature vector of the new sample, y is the feature vector of the sample in the fused feature dataset, and n is the dimension of the feature;

[0133] A K value determination unit, used to determine an optimal K value;

[0134] A neighbor sample determination unit is used to select K samples with the closest distance as neighbor samples according to the calculated distance d;

[0135] The classification unit is used to count the number of each category in the neighboring samples, classify the new sample into the category with the largest number, and obtain the classification result of the new sample;

[0136] The product recommendation unit is used to determine the target financial product according to the classification result based on the preset mapping relationship between the classification result and the financial product.

[0137] An embodiment of the present invention further provides a computer device, the device comprising:

[0138] one or more processors;

[0139] A memory for storing one or more programs;

[0140] When one or more programs are executed by one or more processors, the one or more processors implement any of the financial product recommendation methods described above.

[0141] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the financial product recommendation methods described above.

[0142] The above-mentioned embodiments only express several implementation methods of the present invention, and the description is relatively specific and detailed, but it cannot be understood as limiting the scope of the invention patent. It should be pointed out that for ordinary technicians in this field, several modifications and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention.

Claims

1. A financial product recommendation method, characterized in that: The method comprises: Acquire a user's life picture, and obtain a first data feature based on the life picture; the first data feature is used to characterize the risk tolerance amount of the user; the life picture includes: at least one of a living environment picture, a travel vehicle picture, a person's clothing picture, and a dining environment picture; Acquire historical investment data of the user, perform feature extraction processing on the historical investment data, and obtain a second data feature; the historical investment data includes the name of the historical investment target of the user and the historical profit and loss data corresponding to the name of the historical investment target; the second feature data is used to characterize the risk tolerance ratio of the user; Performing feature fusion on the first data feature and the second data feature to obtain a fused data feature; A classification result corresponding to the fused data feature is obtained based on the KNN algorithm, and the financial product corresponding to the classification result is determined as the target financial product.

2. The method according to claim 1, characterized in that The obtaining of the first data feature according to the life picture includes: Based on the trained image feature extraction model, the comprehensive data features of the living environment picture, the travel vehicle picture, the person's clothing picture and the dining environment picture are obtained according to the life picture as the first data feature.

3. The method according to claim 2, characterized in that The comprehensive data features of the living environment picture, the travel vehicle picture, the person's clothing picture and the dining environment picture are obtained according to the life picture based on the trained image feature extraction model as the first data features, including: Performing a first preprocessing on each of the life pictures to obtain an image to be extracted corresponding to each of the life pictures; Inputting the image to be extracted into a trained image feature extraction model to obtain feature vectors corresponding to each of the life pictures; The feature vectors are fused by a weighted summation method to obtain comprehensive feature data as the first data feature; wherein different weights are assigned to different feature vectors according to the importance of the life pictures corresponding to different feature vectors.

4. The method according to claim 2, characterized in that: The training method of the image feature extraction model includes: Acquire a life picture data set and labels corresponding to the life pictures to obtain a labeled data set; wherein the labels are the risk tolerance amounts corresponding to the life pictures; the labeled data set is divided into a training set, a validation set and a test set; Use random initialization method to initialize the parameters of the model; The model is subjected to cyclic training; wherein the cyclic training steps include: step 1, randomly extracting a batch of samples from the training set, inputting the samples into the model, and obtaining a prediction result; step 2, calculating the loss value between the prediction result and the true label according to the loss function; step 3, using an optimization algorithm to update the parameters of the model according to the loss value; After each training round, the performance of the model is evaluated based on the validation set. If the performance of the model on the validation set does not improve any more, the cyclic training is stopped to obtain a trained feature extraction model. The method of evaluating the performance of the model based on the validation set is to calculate the accuracy of the model on the validation set, and the accuracy is used to characterize the performance of the model.

5. The method according to claim 1, characterized in that The performing feature extraction processing on the historical investment data to obtain the second data feature includes: Performing a second preprocessing on the historical profit and loss data corresponding to each of the historical investment target names to obtain a plurality of data to be processed; The multiple data to be processed are respectively input into the trained deep reinforcement learning model to obtain multiple second data features corresponding to the name of each of the historical investment targets.

6. The method according to claim 5, characterized in that The training process of the deep reinforcement learning model includes: Obtaining a historical profit and loss data set of the users of the trading platform through the trading platform, performing a second preprocessing on the historical profit and loss curve data in the historical profit and loss data set, and obtaining test data; wherein the test data is divided into training data and verification data, and the profit and loss curve data is divided into stock data, fund data, and bond data; Constructing the deep reinforcement learning model; Inputting the training data into the deep reinforcement learning model to train the deep reinforcement learning model; The verification data is input into the trained deep reinforcement learning model to optimize the parameters of the deep reinforcement learning model.

7. The method according to claim 1, characterized in that The obtaining of the classification result corresponding to the fused data feature based on the KNN algorithm, and determining the financial product corresponding to the classification result as the target financial product, includes: Obtaining a fused feature data set with labeled classifications, wherein the fused feature data set includes fused data feature samples and corresponding financial product classification labels; The fused feature data is used as a new sample, and the Euclidean distance calculation formula is used to calculate the distance d between the new sample and each sample in the fused feature data set: Wherein, x is the feature vector of the new sample, y is the feature vector of the sample in the fused feature dataset, and n is the dimension of the feature; Determine the optimal K value; According to the calculated distance d, select the K samples with the closest distance as the nearest neighbor samples; Counting the number of each category in the neighboring samples, classifying the new sample into the category with the largest number, and obtaining the classification result of the new sample; Based on a preset mapping relationship between the classification result and the financial product, a target financial product is determined according to the classification result.

8. A financial product recommendation device, characterized in that: include: A picture acquisition module, used to acquire a user's life picture and obtain a first data feature according to the life picture; The first data feature is used to characterize the risk tolerance amount of the user; The life pictures include at least one of: living environment pictures, travel vehicle pictures, people's clothing pictures and dining environment pictures; a feature extraction module, used to obtain historical investment data of a user, perform feature extraction processing on the historical investment data, and obtain a second data feature; the historical investment data includes the name of the historical investment target of the user and the historical profit and loss data corresponding to the name of the historical investment target; the second feature data is used to characterize the risk tolerance ratio of the user; A feature fusion module, used for fusing the first data feature and the second data feature to obtain a fused data feature; The product recommendation module is used to obtain the classification result corresponding to the fused data feature based on the KNN algorithm, and determine the financial product corresponding to the classification result as the target financial product.

9. A computer device, characterized in that: include: one or more processors; A memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the financial product recommendation method as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the financial product recommendation method as described in any one of claims 1 to 7 is implemented.

Images