Recommendation model optimization method, management and control product recommendation method and device

Abstract

The embodiment of the invention discloses a recommendation model optimization method and a management and control product recommendation method and device.Firstly, reward labels corresponding to various management and control service states are determined through a preset reward rule, and a preset network model is supervised and trained based on the reward labels and two-tuple features so as to construct a reward model; wherein the two-tuple features are composed of transaction event features and control product features. And further, evaluating the candidate controlled product corresponding to each transaction event by using a reward model, and generating a reward signal of the candidate controlled product. And then, correcting the original management and control tag of each transaction event according to the reward signal to obtain a target management and control tag, training a preset recommendation model by using the target management and control tag to realize optimization of the recommendation model, and finally realizing accurate recommendation of the management and control product by using the optimized recommendation model.

Description

Technical Field

[0001] The present invention relates to data processing technology, and in particular to a recommendation model optimization method, a management and control product recommendation method and device. Background Art

[0002] Internet finance is the product of the combination of the Internet and the financial industry, including various forms such as third-party payment and big data finance. Internet finance can achieve symmetry of new market information, and also has a convenient payment function, so that the supply and demand sides of funds can trade directly.

[0003] Among them, risk control is an important part of payment security management, which can determine what kind of control action to output for risky transaction events, such as directly letting them go, performing multiple verifications such as SMS and face verification on transaction events, etc. In the transaction process, various recommendation models are usually used to recommend products that are more in line with user expectations and have risk control capabilities to users, so as to ensure user payment security. At present, there is an urgent need to provide a solution for optimizing the recommendation model, so as to use the optimized recommendation model to more accurately recommend control products to users. Summary of the invention

[0004] The embodiments of this specification provide a recommendation model optimization method, a management product recommendation method and a device, which can optimize the recommendation model more quickly and simply, enhance the generalization ability of the recommendation model, and thus improve the accuracy of management product recommendations. The recommendation model optimization method includes:

[0005] Use preset reward rules to determine the reward labels corresponding to various management and control business statuses;

[0006] Performing supervised training on a preset network model based on the reward label and the binary features to construct a reward model, wherein the binary features are composed of transaction event features and regulated product features;

[0007] Using the reward model to evaluate the candidate controlled products corresponding to each transaction event, and generating a reward signal for the candidate controlled products;

[0008] The original control label of each transaction event is corrected according to the reward signal to obtain a target control label, and the preset recommendation model is trained using the target control label.

[0009] Further, in some embodiments, the supervised training of a preset network model based on the reward label and the binary feature to construct a reward model includes:

[0010] Using the reward label, the binary feature and the auxiliary label as supervised training data, and using the supervised training data to perform multi-task learning on a preset network model to construct the reward model;

[0011] Among them, the auxiliary labels include risk reporting labels and customer complaint reporting labels.

[0012] Further, in some embodiments, the preset network model includes a plurality of expert networks, a plurality of gating networks, and a plurality of combiners;

[0013] The step of using the reward label, the binary feature and the auxiliary label as supervised training data and performing multi-task learning on a preset network model using the supervised training data to construct the reward model includes:

[0014] Using each of the expert networks to extract features from the supervised training data to obtain a plurality of feature vectors;

[0015] Based on the supervised training data, determining the weight of each of the feature vectors through each of the gating networks corresponding to each task;

[0016] The feature vectors and the weights of the feature vectors are combined by the combiners corresponding to each task to obtain a task prediction value;

[0017] The reward label, the risk report label, the customer complaint report label and the task prediction value are used to perform supervised training on each expert network, each gating network and each combiner corresponding to each task. After the training is completed, the reward model is obtained.

[0018] Further, in some embodiments, each of the expert networks includes a dynamic embedding layer, a first deep neural network layer, and a second deep neural network layer;

[0019] The method of extracting features from the supervised training data using the multiple expert networks to obtain multiple feature vectors includes:

[0020] Performing feature conversion on the supervised training data through each of the dynamic embedding layers to obtain underlying features;

[0021] Performing nonlinear transformation on the underlying features through each of the first deep neural network layers to obtain cross-task shared features in the underlying features;

[0022] Independent features for the target task are extracted from the shared features through each of the second deep neural network layers, and the independent features are integrated to obtain a corresponding feature vector.

[0023] Further, in some embodiments, the dynamic embedding layer includes a multi-embedding layer and a third deep neural network layer;

[0024] The step of converting the supervised training data into features through each of the dynamic embedding layers to obtain underlying features includes:

[0025] Binning the supervised training data to obtain bin features;

[0026] Performing quantile division processing on the supervised training data to obtain quantile features;

[0027] The binning feature is converted into a first intermediate feature through the multi-embedding layer, and the quantile feature is nonlinearly transformed through the third deep neural network layer to obtain a second intermediate feature;

[0028] The first intermediate feature and the second intermediate feature are fused to obtain the underlying feature.

[0029] Furthermore, in some embodiments, the using the reward model to evaluate the candidate regulated product corresponding to each transaction event to generate a reward signal for the candidate regulated product includes:

[0030] Traversing multiple candidate regulated products corresponding to each of the transaction events, and generating multiple sample pair data according to each of the transaction events and each of the candidate regulated products;

[0031] The sample pair data are input into the reward model to generate a reward signal for each candidate regulated product corresponding to each transaction event.

[0032] Furthermore, in some implementations, correcting the original control label of each transaction event according to the reward signal to obtain a target control label includes:

[0033] Selecting a target controlled product whose reward signal meets a preset screening condition from the candidate controlled products;

[0034] The original control label of each transaction event is adjusted to the target control label generated by the target control product.

[0035] Furthermore, in some implementations, the use of preset reward rules to determine reward labels corresponding to various management and control business states includes:

[0036] Determine corresponding reward rules based on control business preferences;

[0037] The reward rules are used to calculate the reward values ​​of each of the managed business states, and each of the reward values ​​is normalized to generate a corresponding reward label.

[0038] The present specification also provides a method for recommending a controlled product, the method comprising:

[0039] Get the characteristic information of the current transaction event;

[0040] Inputting the feature information into a trained recommendation model to obtain a to-be-recommended regulated product corresponding to the current transaction event, and displaying the to-be-recommended regulated product on a user interface;

[0041] The recommendation model is trained by any one of the recommendation model optimization methods.

[0042] The present specification also proposes a recommendation model optimization device, the device comprising:

[0043] A reward label determination module is used to determine the reward labels corresponding to various management and control business states using preset reward rules;

[0044] A reward model building module, used for supervising and training a preset network model based on the reward label and the binary features to build a reward model, wherein the binary features are composed of transaction event features and regulated product features;

[0045] A reward signal generating module, used to evaluate the candidate regulated product corresponding to each transaction event using the reward model, and generate a reward signal for the candidate regulated product;

[0046] The recommendation model training module is used to correct the original control label of each transaction event according to the reward signal to obtain a target control label, and use the target control label to train a preset recommendation model.

[0047] The embodiment of this specification also provides a device for recommending managed products, the device comprising:

[0048] A feature information acquisition module is used to acquire feature information of the current transaction event;

[0049] A controlled product recommendation module, used for inputting the feature information into a trained recommendation model, obtaining a to-be-recommended controlled product corresponding to the current transaction event, and displaying the to-be-recommended controlled product on a user interface;

[0050] The recommendation model is trained by any one of the recommendation model optimization methods.

[0051] The embodiments of the present specification also provide a storage medium, wherein the storage medium stores a computer program, and the computer program is suitable for being loaded by a processor and executing the steps of the above method.

[0052] An embodiment of the present specification also provides an electronic device, comprising: a processor and a memory; wherein the memory stores a computer program, and the computer program is suitable for being loaded by the processor and executing the steps of the above method.

[0053] The embodiments of the present specification also provide a computer program product, wherein the computer program product stores at least one instruction, and the at least one instruction is suitable for being loaded by a processor and executing the above method steps.

[0054] In the embodiments of the present specification, first, the reward labels corresponding to various management and control business states are determined using preset reward rules, and the preset network model is supervised and trained based on the reward labels and binary features to construct a reward model, wherein the binary features are composed of transaction event features and management and control product features. Furthermore, the reward model is used to evaluate the candidate management and control products corresponding to each transaction event to generate a reward signal for the candidate management and control products. Then, the original management and control label of each transaction event is corrected according to the reward signal to obtain the target management and control label, and the preset recommendation model is trained using the target management and control label to optimize the recommendation model, and finally the optimized recommendation model is used to achieve accurate recommendations for management and control products. By adopting the method in the embodiments of this specification, on the one hand, by combining transaction event features and control product features to form binary features and constructing a reward model, the reward model can deeply understand the correlation between different transaction events and control products, thereby objectively quantifying the performance of different control products under specific transaction events, reducing dependence on manual experience, and improving the intelligence and automation level of the decision-making process; on the other hand, based on the automatic correction of the original control label based on the reward signal, the generated target control label is closer to the actual business needs and risk conditions, so that the recommendation model can learn more accurate control strategies during the training process, thereby providing control product recommendations that are more in line with scenario requirements in actual applications, that is, the optimization of the recommendation model is realized more quickly and simply, and the generalization ability of the recommendation model is enhanced, so that the recommendation model can more accurately identify and respond to users' specific needs and preferences, thereby improving the accuracy of control product recommendations. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 A schematic diagram of a system architecture for applying the embodiments of this specification;

[0056] Figure 2 A flowchart of a recommendation model optimization method is provided for the embodiments of this specification;

[0057] Figure 3 A flowchart of another recommendation model optimization method provided in an embodiment of this specification;

[0058] Figure 4 A flowchart of another recommendation model optimization method provided in an embodiment of this specification;

[0059] Figure 5A schematic diagram of a process of extracting feature vectors using an expert network provided in an embodiment of this specification;

[0060] Figure 6 Another flow chart of extracting feature vectors using an expert network is provided for the embodiments of this specification;

[0061] Figure 7 A schematic diagram of the structure of an expert network provided in an embodiment of this specification;

[0062] Figure 8 A flowchart of a method for managing product recommendations is provided for the embodiments of this specification;

[0063] Fig. 9 A schematic diagram of the structure of a recommendation model optimization device provided in an embodiment of this specification;

[0064] Fig.10 A schematic diagram of the structure of a device for recommending products for management and control provided in an embodiment of this specification;

[0065] Fig.11 A schematic diagram of the structure of an electronic device provided in an embodiment of this specification. DETAILED DESCRIPTION

[0066] In order to make the purpose, technical solutions and advantages of this specification more clear, the technical solutions of this specification will be clearly and completely described below in combination with the specific embodiments of this specification and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of this specification, not all of them. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this specification.

[0067] Figure 1 A schematic diagram of an architecture to which the embodiments of this specification can be applied is shown.

[0068] like Figure 1 As shown, the system architecture 100 may include one or more of terminal devices such as a smart phone 101, a portable computer 102, a desktop computer 103, a network 104, and a server 105. The network 104 is used to provide a medium for a communication link between the terminal device and the server 105. The network 104 may include various connection types, such as wired or wireless communication links or optical fiber cables, etc. The terminal device may be various electronic devices with data processing functions, and the electronic device may have a display screen, which is used to display reward tags corresponding to various management and control business states, original management tags and target management tags of each transaction event, and recommended management products corresponding to the current transaction event, etc.

[0069] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. According to the implementation requirements, there may be any number of terminal devices, networks and servers. For example, the server 105 may be a server cluster composed of multiple servers.

[0070] See also Figure 2 , which is a flowchart of a recommendation model optimization method provided in an embodiment of this specification. In an embodiment of this specification, the recommendation model optimization method is applied to a recommendation model optimization device or an electronic device equipped with a recommendation model optimization device. Figure 2 The process shown is described in detail, and the recommendation model optimization method may specifically include the following steps:

[0071] S202, using preset reward rules to determine reward labels corresponding to various management and control business states;

[0072] In one or more embodiments of the present specification, the management and control business status can be determined according to transaction events, management and control products, risk reporting, and customer complaint reporting, etc. For example, one management and control business status is: successful transaction, management and control product A, no risk reporting, no customer complaint reporting, another management and control business status is: failed transaction, management and control product B, risk reporting, no customer complaint reporting, and another management and control business status is: successful transaction, management and control product C, risk reporting, and customer complaint reporting.

[0073] It can be understood that regulated products refer to services provided to users based on their operations in transaction events, which are used to optimize the risks, compliance and efficiency involved in transaction events.

[0074] The reward label is used to quantify the current management and control business status to reflect the degree of achievement of the management and control business objectives. Optionally, the corresponding reward rules can be determined based on the management and control business preferences, and the reward values ​​of various management and control business states can be calculated using the reward rules, and each reward value is normalized to generate a corresponding reward label.

[0075] Among them, the management and control business preferences include maximizing the transaction success rate or transaction amount, minimizing transaction risks, and improving user satisfaction. Specific reward rules can be set based on the management and control business preferences. For example, for successful transaction events, positive rewards are given, corresponding to higher reward values, and for failed transaction events, rewards are reduced, corresponding to lower reward values; for risk-free reported transaction events, positive rewards are given; for risk-free reported transaction events, rewards are reduced or penalties are imposed; for transaction events without customer complaints, positive rewards are given, and vice versa, if there are customer complaints, penalties are imposed; in the case of risk-free reported transactions, the lighter the control product provided, the higher the reward value given; in the case of risky reported transactions, the stricter the control product provided, the higher the reward value given; in the case of failed transactions, there are customer complaints, the lighter the control product provided, the higher the reward value given; the corresponding reward rules can be flexibly set according to specific application scenarios, and this manual does not limit this.

[0076] Next, analyze the characteristics of various management and control business states, including whether the transaction is successful, whether there is a risk report, whether there is a customer complaint report, and the management and control products provided, and use the reward rules to calculate the reward values ​​of various management and control business states. For example, a basic reward value is calculated based on the success or failure of the transaction event, and the basic reward value is adjusted based on the risk report information and customer complaint report information to increase or decrease the reward value to obtain the final reward value. It should be noted that when generating reward labels, other factors of business preferences, such as transaction size, can also be considered to further fine-tune the reward value.

[0077] In order to enable the reward model to learn effectively, the calculated reward value needs to be normalized to ensure that the processed reward label is within a reasonable range to avoid values ​​that are too large or too small affecting the convergence and stability of the model.

[0078] By converting the management and control business status into specific numerical values ​​through reward labels, the reward model can quantitatively understand the consequences of its decisions. For example, a high reward label may indicate that the risk control measures taken in a certain transaction event successfully reduced losses, while a low reward label may indicate that the risk was not effectively identified or mitigated.

[0079] S204, performing supervised training on a preset network model based on the reward label and the binary feature to construct a reward model, wherein the binary feature consists of a transaction event feature and a managed product feature;

[0080] In one or more embodiments of this specification, a basic neural network architecture can be selected as a preset network model, such as a convolutional neural network, a recurrent neural network, a long short-term memory network or a Transformer network, etc., and this specification does not limit this. After obtaining the reward labels of various management and control business states, the reward labels and the binary features consisting of transaction event features and management and control product features can be integrated into a complete training data set for the preset network model to use for supervised learning.

[0081] Among them, transaction event features include features extracted from information such as transaction amount, transaction time and transaction location, and regulated product features include features extracted from information such as product type, regulation intensity, and regulation effect.

[0082] The training data set can be divided into a training set, a validation set, and a test set. The parameters of the preset network model are adjusted on the training set, and the loss function is minimized through the back propagation algorithm. The hyperparameters of the preset network model are adjusted using the validation set to prevent overfitting. The loss function can be the mean square error, cross entropy loss function, etc. The model performance is evaluated on the test set, and the model structure or training parameters are continuously adjusted based on the performance of the model on the test set to further improve the model performance, thereby constructing a reward model.

[0083] S206, using the reward model to evaluate the candidate controlled product corresponding to each transaction event, and generating a reward signal for the candidate controlled product;

[0084] For each transaction event, all candidate control products are traversed, and multiple sets of sample data are generated according to each transaction event and each candidate control product, denoted as {T i ,P j}, where T i represents the i-th transaction event, P j Represents the jth candidate regulated product. Based on each set of sample data, a binary feature vector can be constructed, which contains a combination of transaction event features and regulated product features to ensure consistency with the features used in building the reward model.

[0085] The constructed reward model is used to evaluate the binary feature vector corresponding to each group of sample data to generate a reward signal for each group of samples, that is, a reward signal for each candidate regulated product under a specific transaction event. The reward signal can be used to quantify the control effect of each candidate regulated product under a specific transaction event and the degree of adaptation to the transaction event.

[0086] S208, correcting the original control label of each transaction event according to the reward signal to obtain a target control label, and using the target control label to train a preset recommendation model.

[0087] Among them, the original control label of each transaction event is a label generated according to the current control product assigned to each transaction event. The process of determining the original control label of each transaction event may be the same as step S202 of determining the reward label, or may be different. This specification does not limit this.

[0088] After obtaining the reward signal of each candidate control product under each transaction event, a target control product whose reward signal meets the preset screening conditions can be selected from multiple candidate control products, and the original control label of each transaction event can be adjusted to the target control label generated by the target control product.

[0089] Optionally, the target control product that achieves the best control effect under each transaction event can be determined from all candidate control products based on the reward signal. For example, the preset screening condition is to set a reward signal threshold, and the candidate control products whose reward signals are greater than the reward signal threshold are used as target control products, and the current control product of the transaction event is replaced with the target control product, and a new control label is regenerated based on the transaction event and the corresponding target control product, that is, a target control label is generated.

[0090] It is understandable that the number of target control products selected will be different depending on the set screening conditions. For example, if the screening condition is to select the control product with the highest reward signal, then there is only one target control product; if the screening condition is to select the control product with a reward signal greater than the reward signal threshold, then the target control product can be one, multiple, or none; if the screening condition is to select the top ten control products with higher reward signals, then there will be multiple target control products.

[0091] By correcting the original control label of each transaction event and obtaining the target control label of each transaction event, the preset recommendation model can be trained using a data set containing the target control label. Of course, the data set also includes sample data consisting of transaction events and control products to ensure that the recommendation model can receive the characteristics of transaction events as input and predict the most suitable control product as output.

[0092] The preset recommendation models may include random forests, gradient boosting trees, or deep learning models. The appropriate machine learning algorithm may be selected based on the complexity of the data and specific needs, and this description does not limit this.

[0093] By combining transaction event features and regulated product features to form binary features and constructing a reward model, the reward model can deeply understand the correlation between different transaction events and regulated products, thereby objectively quantifying the performance of different regulated products under specific transaction events, reducing reliance on manual experience, and improving the intelligence and automation of the decision-making process. Moreover, based on the automatic correction of the original regulated labels by reward signals, the generated target regulated labels are closer to actual business needs and risk conditions, allowing the recommendation model to learn more accurate regulated strategies during training, thereby providing regulated product recommendations that are more in line with scenario needs in actual applications, and optimizing the recommendation model more quickly and simply.

[0094] See also Figure 3 , is a flow chart of another recommendation model optimization method provided in the embodiments of this specification. Figure 3 As shown, the method comprises the following steps:

[0095] S302, using preset reward rules to determine reward labels corresponding to various management and control business states;

[0096] For step S302, please refer to the detailed description of step S202 in another embodiment of this specification, which will not be repeated here.

[0097] S304, using the reward label, the binary feature and the auxiliary label as supervised training data, and using the supervised training data to perform multi-task learning on a preset network model to construct the reward model;

[0098] Among them, the binary features are composed of transaction event features and control product features. The transaction event features include features extracted from information such as transaction amount, transaction time and transaction location, and the control product features include features extracted from information such as product type, control intensity, and control effect. The preset network models include convolutional neural networks, recurrent neural networks, long short-term memory networks, or Transformer networks. Auxiliary labels include risk reporting labels and customer complaint reporting labels. The risk reporting label is used to reflect whether there is a risk in the transaction event and the specific type of risk. The customer complaint reporting label is used to record customer complaints, including complaint frequency, severity, and resolution efficiency.

[0099] In one or more embodiments of the present specification, reward labels and binary features can be used as training data sets for supervised learning by a preset network model to directly optimize the model's predictive ability for the main tasks, so that the model can learn to predict the expected effects of different regulated products when processing different transaction events.

[0100] In order to improve the interpretability and predictive performance of the model, risk reporting and customer complaint reporting can be used as auxiliary labels for multi-task learning, that is, risk reporting labels and customer complaint reporting labels are introduced as additional learning targets. The model should not only learn to predict the control effect, but also learn to identify which transaction events may lead to risks or customer complaints. After the preset network model completes multi-task learning, the reward model is obtained.

[0101] S306, using the reward model to evaluate the candidate controlled product corresponding to each transaction event, and generating a reward signal for the candidate controlled product;

[0102] For step S306, please refer to the detailed description of step S206 in another embodiment of this specification, which will not be repeated here.

[0103] S308, correcting the original control label of each transaction event according to the reward signal to obtain a target control label, and using the target control label to train a preset recommendation model;

[0104] For step S308, please refer to the detailed description of step S208 in another embodiment of this specification, which will not be repeated here.

[0105] By introducing reward mechanisms and multi-task learning, the model can learn richer data structures and patterns, improve the generalization and robustness of the model, and thus improve the comprehensive prediction ability of the model. Label correction and model optimization also promote the iteration and improvement of the model, so that it can be continuously optimized as the business environment changes.

[0106] See also Figure 4 , which is a flow chart of another method for optimizing a recommendation model provided in an embodiment of this specification. The preset network model may include multiple expert networks, multiple gated networks, and multiple combiners. Figure 4 As shown, the method comprises the following steps:

[0107] S402, using preset reward rules to determine reward labels corresponding to various management and control business states;

[0108] For step S402, please refer to the detailed description of step S202 in another embodiment of this specification, which will not be repeated here.

[0109] S404, using each of the expert networks to perform feature extraction on the supervised training data to obtain a plurality of feature vectors;

[0110] Among them, the supervised training data consists of reward labels, binary features and auxiliary labels. The binary features consist of transaction event features and control product features. Transaction event features include features extracted from information such as transaction amount, transaction time and transaction location. Control product features include features extracted from information such as product type, control intensity and control effect. Auxiliary labels include risk reporting labels and customer complaint reporting labels. Risk reporting labels are used to reflect whether there is a risk in a transaction event and the specific type of risk. Customer complaint reporting labels are used to record customer complaints, including complaint frequency, severity and resolution efficiency.

[0111] In one or more embodiments of the present specification, when constructing a reward model, supervised learning is required for three tasks: reward prediction, risk return prediction, and customer complaint return prediction. Each task corresponds to an expert network, and each expert network generates an output related to a specific task.

[0112] It should be noted that multiple expert networks are parallel networks, and each expert network learns for the entire data set. However, as the training progresses, each expert network becomes an "expert" in processing specific types of data or specific tasks.

[0113] The supervised training data is input into each expert network for feature extraction. Each expert network outputs a feature vector. Each feature vector captures the key attributes of the input data for a specific task for subsequent model use.

[0114] S406, based on the supervised training data, determining the weight of each of the feature vectors through each of the gating networks corresponding to each task;

[0115] The gating network is used to determine which expert network outputs are most relevant to the input supervised training data, and then dynamically assign weights to different expert networks at each data point. The gating network dynamically adjusts the weights by learning the features of different data points to ensure that the most suitable expert network can lead the prediction process at different data points. It is understood that in the embodiments of this specification, a data point refers to a set of sample data consisting of a reward label, a binary feature, and an auxiliary label.

[0116] Optionally, each gated network corresponds to a task, namely reward prediction, risk report prediction and customer complaint report prediction. For example, suppose there are three expert networks E1, E2 and E3 and a gated network G. For task T1, when a data point is input into the model, the gated network G calculates a set of weights w1, w2, w3 based on the characteristics of the data point based on the attention mechanism, corresponding to E1, E2 and E3 respectively. This set of weights determines the contribution of the feature vectors output by each expert network at this data point to the final prediction.

[0117] Therefore, the gating network can intelligently select the most appropriate expert network for prediction according to the characteristics and requirements of different data points, thereby improving the overall prediction accuracy and effect of the model.

[0118] S408, combining the feature vectors and the weights of the feature vectors through the combiners corresponding to each task to obtain a task prediction value;

[0119] Each task has a combiner that generates task-related output, i.e., the task prediction value. Specifically, the output is obtained by combining the feature vectors and the weights of the feature vectors by the combiner corresponding to each task, for example, multiplying the feature vectors by the corresponding weights to generate the final task prediction value.

[0120] Optionally, the combiner can be a multi-layer perceptron, a loop layer, an attention layer, etc., which is not limited in this specification. For the reward model, the task prediction value obtained includes the output of three tasks: reward prediction, risk return prediction, and customer complaint return prediction.

[0121] S410, using the reward label, the risk report label, the customer complaint report label and the task prediction value to perform supervised training on each of the expert networks, each of the gating networks and each of the combiners corresponding to each task, and after the training is completed, the reward model is obtained;

[0122] The reward label, risk report label, customer complaint report label and task prediction value are integrated to obtain the training data set, and the training samples in the training data set are input into each expert network. Each expert network will output a feature vector accordingly. At the same time, the gating network evaluates the output of all expert networks and generates a weight distribution to guide the output of all expert networks. The combiner performs weighted average on the output of the expert network according to the weight distribution generated by the gating network to generate the final task prediction value.

[0123] The final task prediction value is compared with the reward label, risk return label, and customer complaint return label in the training data set to calculate the loss function. It is understandable that the design of the loss function should take all labels into consideration to ensure that the model can achieve good performance in different aspects. Through the back-propagation algorithm, the gradient is calculated according to the loss function, and the parameters of the expert network, gating network, and combiner are updated using an optimization algorithm (such as gradient descent) to minimize the loss. The entire training process will iterate multiple times, and each iteration will adjust the parameters of the model expert network, gating network, and combiner until the preset iteration termination conditions are met to obtain the reward model. Among them, the preset iteration termination conditions include reaching the maximum number of iterations, loss function convergence, etc.

[0124] S412, using the reward model to evaluate the candidate controlled product corresponding to each transaction event, and generating a reward signal for the candidate controlled product;

[0125] For step S412, please refer to the detailed description of step S206 in another embodiment of this specification, which will not be repeated here.

[0126] S414, correcting the original control label of each transaction event according to the reward signal to obtain a target control label, and using the target control label to train a preset recommendation model;

[0127] For step S414, please refer to the detailed description of step S208 in another embodiment of this specification, which will not be repeated here.

[0128] In the embodiments of this specification, through the design of the gated network and the expert network, the model can simultaneously learn three related but different tasks of reward prediction, risk report prediction and customer complaint report prediction during the training process, so that the model can not only predict rewards, but also identify and understand potential risk factors and customer satisfaction issues, thereby providing more comprehensive and in-depth business insights, which will help to recommend management products more accurately in the future.

[0129] See also Figure 5 , is a flow chart of extracting feature vectors using an expert network provided in an embodiment of this specification. Each expert network includes a dynamic embedding layer, a first deep neural network layer, and a second deep neural network layer. Figure 5 As shown, the process includes the following steps:

[0130] S502, performing feature conversion on the supervised training data through each of the dynamic embedding layers to obtain underlying features;

[0131] The dynamic embedding layer is used to perform preliminary feature conversion on the input supervised training data, and the parameters of the dynamic embedding layer can be dynamically adjusted according to the different input data, so as to generate more accurate feature representation for different input data.

[0132] For example, the underlying features can be embedded vectors obtained by mapping discrete or continuous features in supervised training data to a high-dimensional continuous space, and the embedded vectors can capture the complex relationships between features.

[0133] S504, performing nonlinear transformation on the underlying features through each of the first deep neural network layers to obtain cross-task shared features in the underlying features;

[0134] After obtaining the underlying features, it is necessary to further explore the common information in the underlying features. The first deep neural network layer is used to extract cross-task shared features from the underlying features. Shared features refer to features that are important for all or most tasks, which can help the model better understand the common patterns and rules between different tasks.

[0135] S506, extracting independent features for the target task from the shared features through each of the second deep neural network layers, and integrating the independent features to obtain a corresponding feature vector;

[0136] The second deep neural network layer focuses on extracting the independent features required for each specific task from the shared features, allowing the model to optimize the performance of each task in a targeted manner without being interfered with by other tasks. Among them, independent features refer to features that are important for a specific task but unimportant relative to other tasks.

[0137] In the embodiments of this specification, by extracting shared features, the model can learn more general representations from multiple related tasks, thereby improving the generalization ability on new tasks. Moreover, the existence of shared features allows the model to learn together on multiple tasks, avoiding the problem of overfitting on a single task. The extraction of independent features helps the model to effectively transfer knowledge between similar tasks, and can quickly adapt to new tasks even when the amount of data is limited. By cleverly combining the extraction of shared features and independent features, both global shared information and task-specific details are taken into account, achieving high efficiency and high performance in multi-task learning scenarios.

[0138] See also Figure 6 , is another flow chart of extracting feature vectors using an expert network provided in the embodiments of this specification. Figure 7 As shown, a structural diagram of an expert network is given, wherein the expert network includes a dynamic embedding layer 701, a first deep neural network layer 702, a second deep neural network layer 703 and a fourth deep neural network layer 704, the dynamic embedding layer 701 includes a multi-embedding layer 711 and a third deep neural network layer 712, and the dynamic embedding layer 701 is used to convert discrete features into continuous vector representations.

[0139] Furthermore, combined with Figure 7 The network structure given in Figure 6 The individual steps shown in are explained below:

[0140] S602, binning the supervised training data to obtain bin features;

[0141] The supervised training data includes various types of data, such as numerical and categorical data. Binning refers to dividing the continuous numerical features in the supervised training data into several intervals, each interval corresponding to a discrete value, so as to convert the numerical features into categorical features, which can better handle nonlinearity and outliers, and facilitate the subsequent embedding layer processing.

[0142] S604, performing quantile division processing on the supervised training data to obtain quantile features;

[0143] The quantile partitioning process divides the continuous numerical features in the supervised training data into certain proportions, such as quartiles, deciles, etc. It can also convert continuous features into categorical form, while also taking into account the characteristics of data distribution.

[0144] S606, converting the binning feature into a first intermediate feature through the multi-embedding layer, and performing a nonlinear transformation on the quantile feature through the third deep neural network layer to obtain a second intermediate feature;

[0145] The preprocessed features are input into the dynamic embedding layer 701. Specifically, the binned features include multiple category features, and the multi-embedding layer 711 can be regarded as multiple separate embedding layers, each of which is responsible for processing a specific feature. Therefore, each category feature is input into the corresponding embedding layer, and the corresponding first intermediate feature is output. The setting of the multi-embedding layer 711 enables the model to capture the information of each category feature more finely.

[0146] The quantile features can be input into the third deep neural network layer 712, which is composed of multiple fully connected layers. The second intermediate features are obtained by nonlinearly transforming the quantile features and extracting the deep expression of the features.

[0147] S608, fusing the first intermediate feature and the second intermediate feature to obtain the underlying feature;

[0148] The first intermediate feature output from the multi-embedding layer 711 and the second intermediate feature output from the third deep neural network layer 712 are fused, such as performing element-wise multiplication of the first intermediate feature and the second intermediate feature to obtain the underlying feature. This fusion method helps to retain the important information of each feature while enhancing the interdependence between the features.

[0149] Taking reward labels, bigram features, and auxiliary labels as input for low-level feature mining ensures that the model can understand the data from multiple perspectives.

[0150] S610, performing nonlinear transformation on the underlying features through each of the first deep neural network layers to obtain cross-task shared features in the underlying features;

[0151] Among them, the first deep neural network layer 702 can automatically learn and extract high-level abstract representations between features through multi-layer neuron connections, that is, learn the common features required for all tasks, that is, shared features.

[0152] The fused underlying features are input into the first deep neural network layer 702, and the multi-layer nonlinear transformation of the first deep neural network layer 702 is used to mine the shared features across tasks in the underlying features, which helps the model understand the basic patterns and structures of the data.

[0153] S612, extracting independent features for the target task from the shared features through each of the second deep neural network layers, and integrating the independent features to obtain a corresponding feature vector;

[0154] In one or more embodiments of the present specification, the second deep neural network layer 703 may include multiple deep neural network layers. Optionally, deeper feature mining can be performed through two independent deep neural network layers, each deep neural network layer mining different feature information for a specific task, such as the specificity of transaction event features or the uniqueness of regulated product features. By extracting information specific to each task, the model can make accurate predictions for different tasks.

[0155] Each deep neural network layer in the second deep neural network layer 703 can output an independent feature. Finally, all independent features can be concatenated, and the concatenated feature vectors are nonlinearly transformed through the fourth deep neural network layer 704 to obtain feature vectors corresponding to specific tasks, ensuring that the model can utilize shared features while taking into account task specificity.

[0156] See also Figure 8 , is a flow chart of a method for recommending a product management system provided in an embodiment of this specification. Figure 8 As shown, the method comprises the following steps:

[0157] S802, obtaining characteristic information of the current transaction event;

[0158] In one or more embodiments of the present specification, the feature information of the current transaction event includes a feature vector of the transaction event. Optionally, the transaction features of the current transaction event may be obtained first, such as the transaction amount, transaction time, and transaction location. Further, the transaction features of the current transaction event are preprocessed and feature engineered, such as missing value processing, data cleaning, feature coding, and feature selection, to obtain a feature vector of the current transaction event, thereby ensuring the quality and validity of the input data of the recommendation model.

[0159] S804: Input the feature information into a trained recommendation model to obtain a to-be-recommended regulated product corresponding to the current transaction event, and display the to-be-recommended regulated product on a user interface.

[0160] It should be noted that the recommendation model is optimized by the method described in detail in steps S202 to S208 of another embodiment of this specification, or optimized by the method described in detail in steps S302 to S308 of yet another embodiment of this specification, or optimized by the method described in detail in steps S402 to S414 of yet another embodiment of this specification, which will not be repeated here.

[0161] The feature vector of the current transaction event is input into the trained recommendation model. Based on the learned prediction model, the recommendation model obtains the recommended control product that best matches the current transaction event. Finally, the recommendation results output by the model are presented to the user through the user interface, including the name of the recommended control product, product description, and recommendation reason.

[0162] In the embodiments of this specification, the optimization of the recommendation model can be realized more quickly and simply. Since the optimized recommendation model has a strong generalization ability and can more accurately identify and respond to the specific needs and preferences of users, the recommendation model can be used to accurately recommend regulated products to users who perform transaction operations, thereby improving the user experience. Moreover, it is possible to provide highly personalized regulated product recommendations based on the specific characteristics of transaction events, further improving the accuracy and pertinence of the recommendations. The automated recommendation system greatly reduces the time for manual screening and decision-making, speeds up the transaction process, and improves overall operational efficiency.

[0163] See also Fig. 9 , is a schematic diagram of the structure of a recommendation model optimization device provided in the embodiments of this specification. Fig. 9 As shown, the recommendation model optimization device 1 can be implemented as all or part of an electronic device through software, hardware or a combination of both. According to some embodiments, the recommendation model optimization device 1 includes a reward label determination module 11, a reward model construction module 12, a reward signal generation module 13 and a recommendation model training module 14, specifically including:

[0164] The reward label determination module 11 is used to determine the reward labels corresponding to various management and control business states by using preset reward rules;

[0165] A reward model building module 12, used for supervising and training a preset network model based on the reward label and the binary feature to build a reward model, wherein the binary feature consists of a transaction event feature and a regulated product feature;

[0166] A reward signal generating module 13, configured to evaluate the candidate regulated product corresponding to each transaction event using the reward model, and generate a reward signal for the candidate regulated product;

[0167] The recommendation model training module 14 is used to correct the original control label of each transaction event according to the reward signal to obtain a target control label, and use the target control label to train a preset recommendation model.

[0168] Optionally, when the reward model construction module 12 performs supervised training on a preset network model based on the reward label and the binary feature to construct the reward model, it is specifically used to:

[0169] Using the reward label, the binary feature and the auxiliary label as supervised training data, and using the supervised training data to perform multi-task learning on a preset network model to construct the reward model;

[0170] Among them, the auxiliary labels include risk reporting labels and customer complaint reporting labels.

[0171] Optionally, the preset network model includes multiple expert networks, multiple gating networks and multiple combiners; when the reward model construction module 12 uses the reward label, the binary feature and the auxiliary label as supervised training data and performs multi-task learning on the preset network model using the supervised training data to construct the reward model, it is specifically used to:

[0172] Using each of the expert networks to extract features from the supervised training data to obtain a plurality of feature vectors;

[0173] Based on the supervised training data, determining the weight of each of the feature vectors through each of the gating networks corresponding to each task;

[0174] The feature vectors and the weights of the feature vectors are combined by the combiners corresponding to each task to obtain a task prediction value;

[0175] The reward label, the risk report label, the customer complaint report label and the task prediction value are used to perform supervised training on each expert network, each gating network and each combiner corresponding to each task. After the training is completed, the reward model is obtained.

[0176] Optionally, each of the expert networks includes a dynamic embedding layer, a first deep neural network layer, and a second deep neural network layer;

[0177] When the reward model building module 12 uses each of the expert networks to extract features from the supervised training data to obtain multiple feature vectors, it is specifically used to:

[0178] Performing feature conversion on the supervised training data through each of the dynamic embedding layers to obtain underlying features;

[0179] Performing nonlinear transformation on the underlying features through each of the first deep neural network layers to obtain cross-task shared features in the underlying features;

[0180] Independent features for the target task are extracted from the shared features through each of the second deep neural network layers, and the independent features are integrated to obtain a corresponding feature vector.

[0181] Optionally, the dynamic embedding layer includes a multi-embedding layer and a third deep neural network layer;

[0182] When the reward model building module 12 performs feature conversion on the supervised training data through each of the dynamic embedding layers to obtain the underlying features, it is specifically used to:

[0183] Binning the supervised training data to obtain bin features;

[0184] Performing quantile division processing on the supervised training data to obtain quantile features;

[0185] The binning feature is converted into a first intermediate feature through the multi-embedding layer, and the quantile feature is nonlinearly transformed through the third deep neural network layer to obtain a second intermediate feature;

[0186] The first intermediate feature and the second intermediate feature are fused to obtain the underlying feature.

[0187] Optionally, when the reward signal generating module 13 uses the reward model to evaluate the candidate regulated product corresponding to each transaction event and generates the reward signal of the candidate regulated product, it is specifically used to:

[0188] Traversing multiple candidate regulated products corresponding to each of the transaction events, and generating multiple sample pair data according to each of the transaction events and each of the candidate regulated products;

[0189] The sample pair data are input into the reward model to generate a reward signal for each candidate regulated product corresponding to each transaction event.

[0190] Optionally, when the recommendation model training module 14 corrects the original control label of each transaction event according to the reward signal to obtain the target control label, it is specifically used to:

[0191] Selecting a target controlled product whose reward signal meets a preset screening condition from the candidate controlled products;

[0192] The original control label of each transaction event is adjusted to the target control label generated by the target control product.

[0193] Optionally, when the reward label determination module 11 determines the reward labels corresponding to various management and control business states by using preset reward rules, it is specifically used to:

[0194] Determine corresponding reward rules based on control business preferences;

[0195] The reward rules are used to calculate the reward values ​​of each of the managed business states, and each of the reward values ​​is normalized to generate a corresponding reward label.

[0196] The above device embodiments correspond to the method embodiments. For specific descriptions, please refer to the description of the method embodiments, which will not be repeated here. The device embodiments are obtained based on the corresponding method embodiments and have the same technical effects as the corresponding method embodiments. For specific descriptions, please refer to the corresponding method embodiments.

[0197] See also Fig.10 , is a schematic diagram of the structure of a control product recommendation device provided in an embodiment of this specification. Fig.10 As shown, the control product recommendation device 2 can be implemented as all or part of the electronic device through software, hardware or a combination of both. According to some embodiments, the control product recommendation device 2 includes a feature information acquisition module 21 and a control product recommendation module 22, specifically including:

[0198] The characteristic information acquisition module 21 is used to acquire characteristic information of the current transaction event;

[0199] A controlled product recommendation module 22, configured to input the feature information into a trained recommendation model, obtain a to-be-recommended controlled product corresponding to the current transaction event, and display the to-be-recommended controlled product on a user interface;

[0200] Wherein, the recommendation model is trained by the method described in the recommendation model optimization device.

[0201] The above device embodiments correspond to the method embodiments. For specific descriptions, please refer to the description of the method embodiments, which will not be repeated here. The device embodiments are obtained based on the corresponding method embodiments and have the same technical effects as the corresponding method embodiments. For specific descriptions, please refer to the corresponding method embodiments.

[0202] The present specification also provides a computer storage medium, which can store multiple instructions, and the instructions are suitable for being loaded and executed by a processor as described above. Figures 2 to 6 , Figure 8 The method of the embodiment shown in the figure can be specifically executed by referring to Figures 2 to 6 , Figure 8 The specific description of the illustrated embodiment will not be repeated here.

[0203] The present specification also provides a computer program product, which stores at least one instruction, and the at least one instruction is loaded and executed by the processor as described above. Figures 2 to 6 , Figure 8 The method of the embodiment shown in the figure can be specifically executed by referring to Figures 2 to 6 , Figure 8 The specific description of the illustrated embodiment will not be repeated here.

[0204] The embodiments of this specification also provide Fig.11 The structural diagram of the electronic device shown in FIG. Fig.11 At the hardware level, the electronic device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory, and may also include hardware required for other services. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to implement the above-mentioned voice activity detection method.

[0205] Of course, in addition to software implementation, this specification does not exclude other implementation methods, such as logic devices or a combination of software and hardware, etc., that is to say, the executor of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0206] In the 1990s, improvements to a technology could be clearly distinguished as hardware improvements (for example, improvements to the circuit structure of diodes, transistors, switches, etc.) or software improvements (improvements to the method flow). However, with the development of technology, many improvements to the method flow today can be regarded as direct improvements to the hardware circuit structure. Designers almost always obtain the corresponding hardware circuit structure by programming the improved method flow into the hardware circuit. Therefore, it cannot be said that an improvement in a method flow cannot be implemented using a hardware entity module. For example, a programmable logic device (PLD) (such as a field programmable gate array (FPGA)) is such an integrated circuit whose logical function is determined by the user's programming of the device. Designers can "integrate" a digital system on a PLD by programming it themselves, without having to ask a chip manufacturer to design and produce a dedicated integrated circuit chip. Moreover, nowadays, instead of manually making integrated circuit chips, this kind of programming is mostly implemented by "logic compiler" software, which is similar to the software compiler used when developing and writing programs, and the original code before compilation must also be written in a specific programming language, which is called hardware description language (HDL). There is not only one HDL, but many kinds, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. The most commonly used ones are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also know that it is only necessary to program the method flow slightly in the above-mentioned hardware description languages ​​and program it into the integrated circuit, and then it is easy to obtain the hardware circuit that implements the logic method flow.

[0207] The controller can be implemented in any appropriate manner, for example, the controller can take the form of a microprocessor or processor and a computer-readable medium storing a computer-readable program code (such as software or firmware) that can be executed by the (micro)processor, a logic gate, a switch, an application-specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320. The memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art also know that in addition to implementing the controller in a purely computer-readable program code manner, the controller can be implemented in the form of a logic gate, a switch, an application-specific integrated circuit, a programmable logic controller, and an embedded microcontroller by logically programming the method steps. Therefore, this controller can be considered as a hardware component, and the devices included therein for implementing various functions can also be regarded as structures within the hardware component. Or even, the devices for implementing various functions can be regarded as both software modules for implementing the method and structures within the hardware component.

[0208] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0209] For the convenience of description, the above device is described in various units according to their functions. Of course, when implementing this specification, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0210] Those skilled in the art will appreciate that the embodiments of this specification may be provided as methods, systems, or computer program products. Therefore, this specification may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0211] This specification is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of this specification. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0212] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0213] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0214] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0215] The memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0216] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0217] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0218] Those skilled in the art will appreciate that the embodiments of this specification may be provided as methods, systems or computer program products. Therefore, this specification may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0219] This specification may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. This specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0220] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0221] The above description is only an embodiment of the present specification and is not intended to limit the present specification. For those skilled in the art, the present specification may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present specification shall be included in the scope of the claims of the present specification.

Claims

1. A recommendation model optimization method, the method comprising: Use preset reward rules to determine the reward labels corresponding to various management and control business statuses; Performing supervised training on a preset network model based on the reward label and the binary features to construct a reward model, wherein the binary features are composed of transaction event features and regulated product features; Using the reward model to evaluate the candidate controlled products corresponding to each transaction event, and generating a reward signal for the candidate controlled products; The original control label of each transaction event is corrected according to the reward signal to obtain a target control label, and the preset recommendation model is trained using the target control label.

2. The recommendation model optimization method according to claim 1, wherein the supervised training of the preset network model based on the reward label and the binary feature to construct the reward model comprises: Using the reward label, the binary feature and the auxiliary label as supervised training data, and using the supervised training data to perform multi-task learning on a preset network model to construct the reward model; Among them, the auxiliary labels include risk reporting labels and customer complaint reporting labels.

3. According to the recommendation model optimization method of claim 2, the preset network model includes multiple expert networks, multiple gating networks and multiple combiners; The step of using the reward label, the binary feature and the auxiliary label as supervised training data and performing multi-task learning on a preset network model using the supervised training data to construct the reward model includes: Using each of the expert networks to extract features from the supervised training data to obtain a plurality of feature vectors; Based on the supervised training data, determining the weight of each of the feature vectors through each of the gating networks corresponding to each task; The feature vectors and the weights of the feature vectors are combined by the combiners corresponding to each task to obtain a task prediction value; The reward label, the risk report label, the customer complaint report label and the task prediction value are used to perform supervised training on each expert network, each gating network and each combiner corresponding to each task. After the training is completed, the reward model is obtained.

4. The recommendation model optimization method according to claim 3, wherein each of the expert networks comprises a dynamic embedding layer, a first deep neural network layer, and a second deep neural network layer; The method of extracting features from the supervised training data using the multiple expert networks to obtain multiple feature vectors includes: Performing feature conversion on the supervised training data through each of the dynamic embedding layers to obtain underlying features; Performing nonlinear transformation on the underlying features through each of the first deep neural network layers to obtain cross-task shared features in the underlying features; Independent features for the target task are extracted from the shared features through each of the second deep neural network layers, and the independent features are integrated to obtain a corresponding feature vector.

5. The recommendation model optimization method according to claim 4, wherein the dynamic embedding layer comprises a multi-embedding layer and a third deep neural network layer; The step of converting the supervised training data into features through each of the dynamic embedding layers to obtain underlying features includes: Binning the supervised training data to obtain bin features; Performing quantile division processing on the supervised training data to obtain quantile features; The binning feature is converted into a first intermediate feature through the multi-embedding layer, and the quantile feature is nonlinearly transformed through the third deep neural network layer to obtain a second intermediate feature; The first intermediate feature and the second intermediate feature are fused to obtain the underlying feature.

6. The recommendation model optimization method according to claim 1, wherein the step of evaluating the candidate regulated products corresponding to each transaction event using the reward model to generate a reward signal for the candidate regulated products comprises: Traversing multiple candidate regulated products corresponding to each of the transaction events, and generating multiple sample pair data according to each of the transaction events and each of the candidate regulated products; The sample pair data are input into the reward model to generate a reward signal for each candidate controlled product corresponding to each transaction event.

7. The recommendation model optimization method according to claim 1, wherein the original control label of each transaction event is corrected according to the reward signal to obtain the target control label, including: Selecting a target controlled product whose reward signal meets a preset screening condition from the candidate controlled products; The original control label of each transaction event is adjusted to the target control label generated by the target control product.

8. The recommendation model optimization method according to claim 1, wherein the step of determining the reward labels corresponding to various management and control business states by using preset reward rules comprises: Determine corresponding reward rules based on control business preferences; The reward rules are used to calculate the reward values ​​of each of the managed business states, and each of the reward values ​​is normalized to generate a corresponding reward label.

9. A method for recommending a controlled product, the method comprising: Get the characteristic information of the current transaction event; Inputting the feature information into a trained recommendation model to obtain a to-be-recommended regulated product corresponding to the current transaction event, and displaying the to-be-recommended regulated product on a user interface; Wherein, the recommendation model is trained by the method described in any one of claims 1-8.

10. A recommendation model optimization device, comprising: A reward label determination module is used to determine the reward labels corresponding to various management and control business states using preset reward rules; A reward model building module, used for supervising and training a preset network model based on the reward label and the binary features to build a reward model, wherein the binary features are composed of transaction event features and regulated product features; A reward signal generating module, used to evaluate the candidate regulated product corresponding to each transaction event using the reward model, and generate a reward signal for the candidate regulated product; The recommendation model training module is used to correct the original control label of each transaction event according to the reward signal to obtain a target control label, and use the target control label to train a preset recommendation model.

11. A device for recommending controlled products, the device comprising: A feature information acquisition module is used to acquire feature information of the current transaction event; A controlled product recommendation module, used for inputting the feature information into a trained recommendation model, obtaining a to-be-recommended controlled product corresponding to the current transaction event, and displaying the to-be-recommended controlled product on a user interface; Wherein, the recommendation model is trained by the method described in any one of claims 1-8.

12. A storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method according to any one of claims 1 to 9.

13. An electronic device, comprising: A processor and a memory; wherein the memory stores a computer program, and the computer program is suitable for being loaded by the processor and executing the steps of the method as claimed in any one of claims 1 to 9.

14. A computer program product having at least one instruction stored thereon, wherein when the at least one instruction is executed by a processor, the steps of the method according to any one of claims 1 to 9 are implemented.

Images