Risk warning method, device, equipment and medium based on dynamic rule construction

Abstract

The present invention discloses a risk warning method, device, equipment and medium based on dynamic rule construction, which relates to the technical field of risk warning. The method extracts user static attributes and dynamic behavior sequences, constructs original data sets, performs time series analysis, generates behavior time series response features, performs cross-modal feature fusion, and constructs spatiotemporal perception behavior vectors; constructs a dynamic rule graph structure according to risk rules and corresponding standard spatiotemporal perception behavior vectors, iteratively updates risk rule weights based on graph convolutional networks, constructs a wavelet domain risk quantification model, and generates multi-scale risk values; inputs the newly generated spatiotemporal perception behavior vector into the wavelet domain risk quantification model to generate a multi-scale risk value to be judged, and performs a risk warning when the threshold condition is not met. Through time series analysis and multi-scale risk quantification models, the accuracy of risk identification is improved, and multi-dimensional and multi-scale risk analysis is supported, thereby effectively improving the accuracy and real-time performance of risk management.

Description

Technical Field

[0001] The present invention relates to the field of risk warning technology, and specifically to a risk warning method, device, equipment and medium based on dynamic rule construction. Background Art

[0002] With the rapid development of the digital economy, the point system is increasingly used in finance, e-commerce, membership services and other fields, becoming an important tool for user incentives and behavior management. The point system issues points based on factors such as user consumption behavior and activity, and promotes user interaction and loyalty through point redemption and point transfer. However, as the scale of the point system continues to expand and business scenarios become increasingly complex, traditional risk control methods face many challenges.

[0003] Existing risk control methods usually rely on static rules, which are based on preset user behavior patterns, such as the amount of points obtained, the frequency of redemption, and other indicators, to make risk control judgments. These static rules are generally configured manually, requiring detailed analysis and definition of possible risk scenarios, and setting constraints based on these definitions. Although this method can deal with some traditional risks, it is not flexible and has poor adaptability when dealing with new, complex and dynamically changing risks. The main problems include: difficulty in rule formulation. With the diversity and complexity of the business scenarios of the points system, the preset static rules often cannot fully cover all possible risk scenarios; poor adaptability. Since static rules cannot self-learn and update according to new data or user behavior patterns, they are prone to ignore new potential risks; large manual intervention. Existing risk control rules require a lot of manual intervention and adjustment, which not only increases the workload, but is also easily affected by human errors; false positives and false negatives. Static rules often cannot effectively distinguish between normal user behavior and risky behavior, resulting in false positives and false negatives, thereby affecting user experience. Summary of the invention

[0004] Based on the above-mentioned shortcomings of the prior art, the purpose of the present invention is to provide a risk warning method, device, equipment and medium based on dynamic rule construction to solve the above-mentioned technical problems.

[0005] To achieve the above object, the present invention provides the following technical solution: a risk early warning method based on dynamic rule construction, comprising:

[0006] S1: extracting user static attributes and dynamic behavior sequences from the points system database, constructing an original data set, and preprocessing the original data set;

[0007] S2: Perform temporal analysis on the preprocessed dynamic behavior sequence to generate behavior temporal response features, perform cross-modal feature fusion based on user static attributes, dynamic behavior sequence and behavior temporal response features, and construct a spatiotemporal perception behavior vector;

[0008] S3: construct a dynamic rule graph structure according to the risk rules and the corresponding standard spatiotemporal perception behavior vectors, and iteratively update the risk rule weights of the dynamic rule graph structure based on a graph convolutional network;

[0009] S4: All standard spatiotemporal perception behavior vectors are stacked row by row to form a spatiotemporal perception behavior matrix. According to the spatiotemporal perception behavior matrix and risk rule weights, a wavelet domain risk quantification model is constructed to generate a multi-scale risk value.

[0010] S5: The newly generated spatiotemporal perception behavior vectors are repeatedly stacked in rows to form a spatiotemporal perception behavior matrix, which is input into the wavelet domain risk quantification model to generate a multi-scale risk value to be judged, and a threshold judgment is performed on the multi-scale risk value to be judged. When the threshold judgment result does not meet the threshold condition, a risk warning is issued.

[0011] The present invention is further configured such that the user static attributes include: user ID, age, gender, region, membership level and credit score; the dynamic behavior sequence includes: point change amount, point change type, point change frequency, and behavior time interval; and the original data set is preprocessed, including normalizing numerical parameters and discretizing non-numerical parameters.

[0012] The present invention is further configured such that the calculation logic of the behavior timing response feature is:

[0013]

[0014] Among them, P dynamic is the behavior time series response feature, T is the number of time points in the time window, t is the tth time point in the time window, LSTM (B t ) is the tth dynamic behavior sequence B t The features of the LSTM network output, η is the decay coefficient, K is the number of behavior patterns, λ k is the weight coefficient of the kth behavior mode, R k (B t ) is the tth dynamic behavior sequence B t The risk measure belonging to the kth behavior pattern, R is the number of dimensions of the behavior time series response characteristics, δ r is the weight coefficient, F r (B t ) is a nonlinear mapping function.

[0015] The present invention is further configured such that the t-th dynamic behavior sequence B t The risk measure R belonging to the kth behavior pattern k (B t ) is calculated as follows: Among them, sigmoid(·) is used to ensure that the contribution of the behavior pattern to the risk measurement is in the range of [0,1], and γ k is the coefficient of the kth behavior mode, J k is the number of features in the kth behavior mode, α k，j is the weight coefficient of the jth feature in the kth behavior mode, B j，t is the dynamic behavior sequence B t The jth feature of

[0016] Nonlinear mapping function F r (B t ) is calculated as follows: r (B t )=tanh(W r ·B t +b r ), where W r and b r is the transformation parameter used to control the dynamic behavior sequence B t The linear mapping of the data, tanh is the activation function, which is used to compress the output value to the range of [-1,1].

[0017] The present invention is further configured such that the calculation logic of cross-modal feature fusion is: Among them, F is the spatiotemporal perception behavior vector, D is a two-dimensional tensor representing the interactive information of two-dimensional data, U is the user's static attribute, B is the dynamic behavior sequence, ×1 and ×2 represent tensor multiplication, σ(·) is the activation function, W is the number of features of the dynamic behavior sequence, α w is the attention weight of the w-th feature, φ w is a nonlinear basis function used to perform nonlinear transformation on the w-th feature, WX w is the weight matrix, is a convolution operation, Pool(B) is a pooling operation on the dynamic behavior sequence B, and tanh is a hyperbolic tangent activation function.

[0018] The present invention is further configured such that, in step S3, the risk rule is set as a dynamic rule graph node, and the feature similarity of the corresponding standard spatiotemporal perception behavior vector is set as an edge weight;

[0019] The calculation logic of the feature similarity of the corresponding standard spatiotemporal perception behavior vector is: Among them, E jk(t) is the edge weight between the jth and kth risk control rules at time t, α is the similarity sensitivity coefficient, and f j (t), f k (t) and f m (t) are the standard spatiotemporal perception behavior vectors corresponding to the jth, kth, and mth risk rules, respectively, and d is the total number of risk rules;

[0020] The update logic of risk rule weight is: (l+1) (t) = ReLU(G(t)·W (l) (t)·Θ (l) ), W (l+1) (t) is the risk rule weight of the graph convolution layer l+1, G(t) is the dynamic rule graph structure, W (l) (t) is the risk rule weight of the graph convolution layer 1, Θ (l) is the time-varying weight matrix of the lth layer, and ReLU(·) is the activation function.

[0021] The present invention is further configured that the calculation logic of the wavelet domain risk quantification model is: Among them, R(t) is the risk value at time t, S is the time scale, ψ s (t) is the scale adaptive weight, W s is the scale feature weight matrix, F is the spatiotemporal perception behavior matrix, * is the convolution operation, F*W (s) (t) represents the spatiotemporal perception behavior matrix F and the scale weight matrix W (s) The convolution operation between (t) is used to extract feature information at a specific scale. is the square of the Frobenius norm, ρ s is the scale-dependent threshold, and the scale-adaptive weight ψ s The calculation logic of (t) is: β s is the scale adjustment factor.

[0022] The present invention also provides a risk warning device based on dynamic rule construction, which is used to implement the above-mentioned risk warning method based on dynamic rule construction. The system includes:

[0023] The first construction module: extracting user static attributes and dynamic behavior sequences from the points system database, constructing an original data set, and preprocessing the original data set;

[0024] Time series analysis module: performs time series analysis on the preprocessed dynamic behavior sequence, generates behavior time series response features, performs cross-modal feature fusion based on user static attributes, dynamic behavior sequence and behavior time series response features, and constructs spatiotemporal perception behavior vectors;

[0025] The second construction module: construct a dynamic rule graph structure according to the risk rules and the corresponding standard spatiotemporal perception behavior vectors, and iteratively update the risk rule weights of the dynamic rule graph structure based on the graph convolutional network;

[0026] The third construction module: stack all standard spatiotemporal perception behavior vectors in rows to form a spatiotemporal perception behavior matrix. According to the spatiotemporal perception behavior matrix and risk rule weights, a wavelet domain risk quantification model is constructed to generate multi-scale risk values.

[0027] Risk warning module: The newly generated spatiotemporal perception behavior vectors are repeatedly stacked in rows to form a spatiotemporal perception behavior matrix, the wavelet domain risk quantification model is input, a multi-scale risk value to be judged is generated, a threshold judgment is performed on the multi-scale risk value to be judged, and a risk warning is issued when the threshold judgment result does not meet the threshold condition.

[0028] The present invention further provides an electronic device, comprising:

[0029] one or more processors;

[0030] A storage device is used to store one or more programs. When the one or more programs are executed by the one or more processors, the electronic device implements the risk warning method based on dynamic rules as described in any of the above.

[0031] The present invention also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor of a computer, the computer is enabled to execute the risk warning method based on dynamic rules as described in any one of the above.

[0032] The present invention provides a risk warning method, device, equipment and medium based on dynamic rule construction. The method extracts user static attributes and dynamic behavior sequences from a scoring system database to construct an original data set, and preprocesses the original data set; performs time series analysis on the preprocessed dynamic behavior sequence to generate behavior time series response features, performs cross-modal feature fusion based on user static attributes, dynamic behavior sequences and behavior time series response features, and constructs a spatiotemporal perception behavior vector; constructs a dynamic rule graph structure according to risk rules and corresponding standard spatiotemporal perception behavior vectors, and iteratively updates risk rule weights for the dynamic rule graph structure based on a graph convolutional network; stacks all standard spatiotemporal perception behavior vectors in rows to form a spatiotemporal perception behavior matrix, and constructs a wavelet domain risk quantification model based on the spatiotemporal perception behavior matrix and risk rule weights to generate a multi-scale risk value; repeatedly stacks the newly generated spatiotemporal perception behavior vectors in rows to form a spatiotemporal perception behavior matrix, inputs the wavelet domain risk quantification model, generates a multi-scale risk value to be judged, performs threshold judgment on the multi-scale risk value to be judged, and performs a risk warning when the threshold judgment result does not meet the threshold condition. The beneficial effects produced include:

[0033] 1. Enhanced the adaptability of the risk control system: By introducing dynamic rules and risk warning methods based on machine learning, the present invention can automatically learn and update risk control rules based on real-time data, thereby achieving dynamic adaptation of the risk warning system. The system can adjust risk assessment standards in a timely manner according to changes in user behavior patterns, avoiding the situation where traditional static rules cannot cope with emerging risks;

[0034] 2. Improved accuracy of risk identification: Through time series analysis, behavior pattern recognition, and multi-scale risk quantification, it is possible to accurately identify potential risk behaviors. Compared with traditional static rule methods, this invention can effectively reduce false positives and false negatives, and improve the accuracy and effectiveness of risk control decisions. Especially when user behaviors are complex and changeable, risk identification based on dynamic rules significantly improves the ability to capture complex risks;

[0035] 3. Support multi-dimensional risk analysis: A multi-scale risk quantification model is introduced, and combined with multi-dimensional information such as spatiotemporal perception behavior vectors, behavior time series response characteristics, and dynamic rule graph structures, it can comprehensively assess risks and analyze risks at different time scales. This multi-dimensional and multi-scale risk analysis method enables the system to effectively predict and warn both short-term risks and long-term trends, further improving the accuracy of risk management.

[0036] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0038] Figure 1 A flowchart of a risk early warning method based on dynamic rule construction is shown as an exemplary embodiment of the present invention;

[0039] Figure 2 The figure is a schematic diagram of the structure of a risk warning device based on dynamic rules according to an exemplary embodiment of the present invention. DETAILED DESCRIPTION

[0040] The following will describe the embodiments of the present invention with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are only for illustrating the present invention, not for limiting the scope of protection of the present invention.

[0041] It should be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present invention, and thus the drawings only show components related to the present invention rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed arbitrarily, and the component layout may also be more complicated.

[0042] In the following description, numerous details are discussed to provide a more thorough explanation of the embodiments of the present invention. However, it is obvious to those skilled in the art that the embodiments of the present invention can be implemented without these specific details. In other embodiments, well-known structures and devices are shown in the form of block diagrams rather than in detail to avoid making the embodiments of the present invention difficult to understand.

[0043] Embodiment 1

[0044] Risk early warning methods based on dynamic rules, such as Figure 1 As shown, including:

[0045] S1: extracting user static attributes and dynamic behavior sequences from the points system database, constructing an original data set, and preprocessing the original data set;

[0046] S2: Perform temporal analysis on the preprocessed dynamic behavior sequence to generate behavior temporal response features, perform cross-modal feature fusion based on user static attributes, dynamic behavior sequence and behavior temporal response features, and construct a spatiotemporal perception behavior vector;

[0047] S3: construct a dynamic rule graph structure according to the risk rules and the corresponding standard spatiotemporal perception behavior vectors, and iteratively update the risk rule weights of the dynamic rule graph structure based on a graph convolutional network;

[0048] S4: All standard spatiotemporal perception behavior vectors are stacked row by row to form a spatiotemporal perception behavior matrix. According to the spatiotemporal perception behavior matrix and risk rule weights, a wavelet domain risk quantification model is constructed to generate a multi-scale risk value.

[0049] S5: The newly generated spatiotemporal perception behavior vectors are repeatedly stacked in rows to form a spatiotemporal perception behavior matrix, which is input into the wavelet domain risk quantification model to generate a multi-scale risk value to be judged, and a threshold judgment is performed on the multi-scale risk value to be judged. When the threshold judgment result does not meet the threshold condition, a risk warning is issued.

[0050] The present invention is further configured such that the user static attributes include: user ID, age, gender, region, membership level and credit score; the dynamic behavior sequence includes: point change amount, point change type, point change frequency, and behavior time interval; the original data set is preprocessed, including normalizing the numerical parameters and discretizing the non-numerical parameters. Specifically, the point change amount refers to the increase or decrease in the number of points of the user in each transaction, reflecting the user's point acquisition and consumption behavior; the point change type refers to the category of point increase or decrease; the point change frequency refers to the frequency of the user's point change, that is, the number of times the points change per unit time; the behavior time interval refers to the time difference between user behaviors, indicating the interval time between two behaviors; discretizing the non-numerical parameters means converting them into a numerical form that can be processed by a computer, including: label encoding: mapping each category to a unique integer; unique hot encoding: representing each category as a binary vector. This is a prior art and is not limited here.

[0051] The present invention is further configured such that the calculation logic of the behavior timing response feature is: Among them, P dynamicis the behavior time series response feature, T is the number of time points in the time window, t is the tth time point in the time window, LSTM (B t ) is the tth dynamic behavior sequence B t The features of the LSTM network output, η is the decay coefficient, K is the number of behavior patterns, λ k is the weight coefficient of the kth behavior mode, R k (B t ) is the tth dynamic behavior sequence B t The risk measure belonging to the kth behavior pattern, R is the number of dimensions of the behavior time series response characteristics, δ r is the weight coefficient, F r (B t ) is a nonlinear mapping function. Specifically, the behavior time series response feature is a feature obtained by integrating all time points and behavior sequences in the time window, which represents the behavior response of the user within a period of time. This feature reflects the relationship between the behavior sequence and the user's final response. LSTM (Long Short-Term Memory Network) is a special type of recurrent neural network (RNN) that is particularly suitable for processing and predicting sequence data. The LSTM network can effectively capture and store long-term dependency information, which makes it very suitable for processing dynamic behavior sequences. In the above calculation logic, the LSTM network is used to extract time-dependent features from the dynamic behavior sequence at each time point and generate feature outputs suitable for subsequent calculations; the behavior pattern represents the classification of user behavior, including different types of behavior or state. The risk metric measures the risk or importance of a user when he belongs to a specific behavior pattern at a certain moment. In a feasible embodiment of the present invention, the behavior pattern is a standard spatiotemporal perception behavior vector corresponding to the risk rule; the nonlinear mapping function is used to map the input behavior sequence to a specific response feature dimension, and process the sequence data in a nonlinear manner. By comprehensively considering time dependency, behavior pattern and risk metric, the user's behavior response can be predicted more accurately, especially in a complex dynamic environment.

[0052] The present invention is further configured such that the t-th dynamic behavior sequence B t The risk measure R belonging to the lth behavior pattern k (B t ) is calculated as follows: Among them, sigmoid(·) is used to ensure that the contribution of the behavior pattern to the risk measurement is in the range of [0,1], and γ k is the coefficient of the lth behavior mode, J k is the number of features in the lth behavior mode, α k，j is the weight coefficient of the jth feature in the kth behavior mode, B j，t is the dynamic behavior sequence B tSpecifically, it aims to evaluate the tth dynamic behavior sequence B t The risk level of the k-th behavior pattern. By using the sigmoid activation function, the output risk metric is ensured to be in the range [0,1];

[0053] Nonlinear mapping function F r (B t ) is calculated as follows: r (B t )=tanh(W r ·B t +b r ), where W r and b r is the transformation parameter used to control the dynamic behavior sequence B t The linear mapping of data, tanh is the activation function, which is used to compress the output value to the range of [-1,1]. Specifically, the nonlinear mapping function F r (B t ) is used to analyze the dynamic behavior sequence B t Nonlinear mapping is performed to further extract and process behavioral features. The input data is mapped to the range of [-1,1] through the tanh activation function; by extracting features and performing nonlinear mapping on dynamic behavior sequences, the user's behavior pattern can be effectively identified. Combining risk measurement and nonlinear mapping functions, the model can better understand the potential risks and changing trends of user behavior and improve the accuracy of behavior prediction.

[0054] The present invention is further configured such that the calculation logic of cross-modal feature fusion is: Among them, F is the spatiotemporal perception behavior vector, D is a two-dimensional tensor representing the interactive information of two-dimensional data, U is the user's static attribute, B is the dynamic behavior sequence, ×1 and ×2 represent tensor multiplication, σ(·) is the activation function, W is the number of features of the dynamic behavior sequence, α w is the attention weight of the w-th feature, φ w is a nonlinear basis function used to perform nonlinear transformation on the w-th feature, WX w is the weight matrix, is a convolution operation, Pool(B) is a pooling operation on the dynamic behavior sequence B, and tanh is a hyperbolic tangent activation function. Specifically, the spatiotemporal perception behavior vector is the final representation obtained through cross-modal feature fusion, which integrates the user's static attributes and dynamic behavior sequence, and transforms the user's behavior, preferences, and historical activities into a feature that can be used by machine learning algorithms; the two-dimensional tensor D is a matrix that represents the interaction information between the user's static attributes and dynamic behavior sequence. The tensor multiplication D×1U×2LSTM(B) is used to model the relationship between the two. Through the calculation of the two-dimensional tensor, the model can capture the interaction between static attributes and dynamic behaviors, thereby generating a more accurate representation of the user's behavior characteristics. The convolution operation is an operation commonly used to process structured grid data, especially in computer vision and time series analysis. Here, the convolution operation is used to combine the weight matrix and the pooled dynamic behavior sequence. The convolution operation helps the model establish local associations between features, thereby enhancing the model's expressive power, especially when processing features with temporal or spatial correlations; the pooling operation is a common data dimensionality reduction technique, especially widely used in deep learning. It is used to downsample dynamic behavior sequences and retain the most important information. Pooling operations help reduce the amount of data, improve computational efficiency, and retain the most representative information in the feature space to prevent overfitting. Nonlinear basis functions are used to perform nonlinear transformations on input features to enhance the nonlinear expression capabilities of the model. By introducing nonlinear transformations, the model can better fit complex behavior patterns and data relationships. Nonlinear mapping improves the model's ability to learn complex feature relationships and can capture potential patterns and changing trends in user behavior. Through cross-modal feature fusion, the user's static attributes and dynamic behavior sequences can be effectively combined to generate more accurate spatiotemporal perception behavior vectors, thereby improving the accuracy of behavior prediction. By introducing nonlinear mapping and attention mechanisms, the model can capture complex feature relationships and improve the ability to understand and predict user behavior.

[0055] The present invention is further configured such that, in step S3, the risk rule is set as a dynamic rule graph node, and the feature similarity of the corresponding standard spatiotemporal perception behavior vector is set as an edge weight;

[0056] The calculation logic of the feature similarity of the corresponding standard spatiotemporal perception behavior vector is: Among them, E jk (t) is the edge weight between the jth and kth risk control rules at time t, α is the similarity sensitivity coefficient, and f j (t), f k (t) and f m(t) are the standard spatiotemporal perception behavior vectors corresponding to the jth, kth, and mth risk rules, respectively, and d is the total number of risk rules. Specifically, the feature similarity between risk rules is calculated, and the difference measure (Euclidean distance) of spatiotemporal perception behavior vectors is used to determine the edge weight according to the distance. By standardizing all rules, the similarity of each pair of rules is obtained.

[0057] The update logic of risk rule weight is: (l+1) (t) = ReLU(G(t)·W (l) (t)·Θ (l) ), W (l+1) (t) is the risk rule weight of the graph convolution layer l+1, G(t) is the dynamic rule graph structure, W (l) (t) is the risk rule weight of the graph convolution layer l, Θ (l) is the time-varying weight matrix of the lth layer, and ReLU(·) is the activation function. Specifically, the graph convolution operation is performed through the graph structure G(t) and the weight matrix W (l) (t) Calculate the risk rule weight W of the next layer (l+1) (t). By applying the ReLU activation function, the updated weights have nonlinear characteristics, thereby enhancing the model's expressiveness. Through cross-modal feature fusion and graph convolution operations, the model can better capture and represent the relationship between complex user behavior patterns and rules. The risk rule graph combined with the spatiotemporal perception behavior vector can more accurately assess the risk of each behavior pattern, thereby helping the decision-making system make more accurate predictions.

[0058] The present invention is further configured that the calculation logic of the wavelet domain risk quantification model is: Among them, R(t) is the risk value at time t, S is the time scale, ψ s (t) is the scale adaptive weight, W s is the scale feature weight matrix, F is the spatiotemporal perception behavior matrix, * is the convolution operation, F*W (s) (t) represents the spatiotemporal perception behavior matrix F and the scale weight matrix W (s) The convolution operation between (t) is used to extract feature information at a specific scale. is the square of the Frobenius norm, ρ s is the scale-dependent threshold, and the scale-adaptive weight ψ s The calculation logic of (t) is: β sSpecifically, the wavelet domain risk quantification model combines the spatiotemporal perception behavior matrix and wavelet transform, realizes the dynamic modeling of risk quantification through multi-scale feature extraction and scale adaptive adjustment, and can extract the user's behavior characteristics at different time scales through wavelet transform and convolution operations, which helps to identify short-term and long-term behavior patterns, thereby improving the accuracy of risk quantification. The scale adaptive weight ψ s (t) It can automatically adjust the importance of features of different scales in the model as time changes, thereby enhancing the adaptability and flexibility of the model.

[0059] In step S5, the generated spatiotemporal perception behavior vector represents the behavior characteristics of the user at a certain moment, and contains information about the user's dynamic behavior and static attributes. The spatiotemporal perception behavior vector is repeatedly stacked row by row to construct a spatiotemporal perception behavior matrix. The repeated stacking operation stacks the user behavior characteristics into a spatiotemporal perception behavior matrix that complies with the input; the spatiotemporal perception behavior matrix is ​​used as input and sent to the wavelet domain risk quantification model for risk calculation. The model will perform in-depth analysis of the user's behavior through multi-scale feature extraction and convolution operations, and calculate the risk value at each time scale. The output result of the wavelet domain risk quantification model is the multi-scale risk value to be judged. Each scale represents a different time dimension. The model will generate a corresponding risk value based on the long-term and short-term dependencies in the behavior data. The multi-scale risk value assigns a risk score to each behavior pattern, reflecting the risk level of the behavior at different time scales; the generated multi-scale risk value is threshold judged, that is, by comparing with the preset risk threshold, it is determined whether the current risk exceeds the predetermined safety range. If the threshold judgment result does not meet the preset conditions, that is, the risk value exceeds the set threshold, the system will trigger a risk warning. Through threshold judgment, the system can automatically detect abnormal high-risk behaviors and issue warnings, helping managers or decision-making systems take timely measures.

[0060] Embodiment 2

[0061] See also Figure 2 The exemplary risk warning device based on dynamic rule construction is used to implement the above-mentioned risk warning method based on dynamic rule construction, and the system includes:

[0062] The first construction module: extracting user static attributes and dynamic behavior sequences from the points system database, constructing an original data set, and preprocessing the original data set;

[0063] Time series analysis module: performs time series analysis on the preprocessed dynamic behavior sequence, generates behavior time series response features, performs cross-modal feature fusion based on user static attributes, dynamic behavior sequence and behavior time series response features, and constructs spatiotemporal perception behavior vectors;

[0064] The second construction module: construct a dynamic rule graph structure according to the risk rules and the corresponding standard spatiotemporal perception behavior vectors, and iteratively update the risk rule weights of the dynamic rule graph structure based on the graph convolutional network;

[0065] The third construction module: stack all standard spatiotemporal perception behavior vectors in rows to form a spatiotemporal perception behavior matrix. According to the spatiotemporal perception behavior matrix and risk rule weights, a wavelet domain risk quantification model is constructed to generate multi-scale risk values.

[0066] Risk warning module: The newly generated spatiotemporal perception behavior vectors are repeatedly stacked in rows to form a spatiotemporal perception behavior matrix, the wavelet domain risk quantification model is input, a multi-scale risk value to be judged is generated, a threshold judgment is performed on the multi-scale risk value to be judged, and a risk warning is issued when the threshold judgment result does not meet the threshold condition.

[0067] It should be noted that the risk warning device based on dynamic rules provided in the above embodiment and the risk warning method based on dynamic rules provided in the above embodiment belong to the same concept, wherein the specific manner in which each module and unit performs operations has been described in detail in the method embodiment and will not be repeated here. In actual applications, the risk warning device based on dynamic rules provided in the above embodiment can distribute the above functions to different functional modules as needed, that is, divide the internal structure of the system into different functional modules to complete all or part of the functions described above, and this is not limited here.

[0068] An embodiment of the present application also provides an electronic device, comprising: one or more processors; a storage device for storing one or more programs, and when the one or more programs are executed by the one or more processors, the electronic device implements the risk warning method based on dynamic rule construction provided in the above-mentioned embodiments.

[0069] The embodiment of the present application also provides a computer system of an electronic device. It should be noted that the computer system of an electronic device is only an example and should not bring any limitation to the function and scope of use of the embodiment of the present application.

[0070] Specifically, the computer system includes a central processing unit (CPU), which can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM) or the program loaded from the storage part to the random access memory (RAM), such as executing the method described in the above embodiment. In the RAM, various programs and data required for system operation are also stored. The CPU, ROM and RAM are connected to each other through a bus. The input / output (I / O) interface is also connected to the bus.

[0071] The following components are connected to the I / O interface: an input part including a keyboard, a mouse, etc.; an output part including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker; a storage part including a hard disk, etc.; and a communication part including a network interface card such as a LAN (Local Area Network) card, a modem, etc. The communication part performs communication processing via a network such as the Internet. A drive is also connected to the I / O interface as needed. Removable media, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., are installed on the drive as needed so that the computer program read therefrom is installed into the storage part as needed.

[0072] In particular, according to an embodiment of the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present application includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a computer program for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication part, and / or installed from a removable medium. When the computer program is executed by a central processing unit (CPU), various functions defined in the system of the present application are executed.

[0073] It should be noted that the computer-readable medium shown in the embodiment of the present application can be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium can be, for example, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media can include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a flash memory, an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, wherein a computer-readable computer program is carried. This propagated data signal can take a variety of forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination of the above. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. A computer program contained on a computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.

[0074] The units involved in the embodiments described in this application may be implemented by software or hardware, and the units described may also be set in a processor. The names of these units do not, in some cases, constitute limitations on the units themselves.

[0075] Another aspect of the present application also provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the risk warning method based on dynamic rule construction as described above is implemented. The computer-readable storage medium can be included in the electronic device described in the above embodiment, or it can exist independently without being assembled into the electronic device.

[0076] Another aspect of the present application also provides a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the risk warning method based on dynamic rule construction provided in each of the above embodiments.

[0077] The above embodiments are merely illustrative of the principles and effects of the present invention, and are not intended to limit the present invention. Anyone familiar with the technology may modify or change the above embodiments without violating the spirit and scope of the present invention. Therefore, all equivalent modifications or changes made by a person of ordinary skill in the art without departing from the spirit and technical ideas disclosed by the present invention shall still be covered by the claims of the present invention.

Claims

1. The risk early warning method based on dynamic rules is characterized by: include: S1: extracting user static attributes and dynamic behavior sequences from the points system database, constructing an original data set, and preprocessing the original data set; The user static attributes include: user ID, age, gender, region, membership level and credit score; the dynamic behavior sequence includes: point change amount, point change type, point change frequency, and behavior time interval; the original data set is preprocessed, including normalizing numerical parameters and discretizing non-numerical parameters; S2: Perform temporal analysis on the preprocessed dynamic behavior sequence to generate behavior temporal response features, perform cross-modal feature fusion based on user static attributes, dynamic behavior sequence and behavior temporal response features, and construct a spatiotemporal perception behavior vector; S3: construct a dynamic rule graph structure according to the risk rules and the corresponding standard spatiotemporal perception behavior vectors, and iteratively update the risk rule weights of the dynamic rule graph structure based on the graph convolution network; set the risk rules as dynamic rule graph nodes, and set the feature similarity of the corresponding standard spatiotemporal perception behavior vectors as edge weights; the calculation logic of the feature similarity of the corresponding standard spatiotemporal perception behavior vectors is: Among them, E jk (t) is the edge weight between the jth and kth risk control rules at time t, α is the similarity sensitivity coefficient, and f j (t), f k (t) and f m (t) are the standard spatiotemporal perception behavior vectors corresponding to the jth, kth, and mth risk rules, respectively, and d is the total number of risk rules; the update logic of the risk rule weight is: W (l+1) (t) = ReLU(G(t)·W (l) (t)·Θ (l) ), W (l+1) (t) is the risk rule weight of the graph convolution layer l+1, G(t) is the dynamic rule graph structure, W (l) (t) is the risk rule weight of the graph convolution layer l, Θ (l) is the time-varying weight matrix of the lth layer, and ReLU(·) is the activation function; S4: All standard spatiotemporal perception behavior vectors are stacked in rows to form a spatiotemporal perception behavior matrix. According to the spatiotemporal perception behavior matrix and risk rule weights, a wavelet domain risk quantification model is constructed to generate multi-scale risk values. The calculation logic of the wavelet domain risk quantification model is: Among them, R(t) is the risk value at time t, S is the time scale, ψ s (t) is the scale adaptive weight, W s is the scale feature weight matrix, F is the spatiotemporal perception behavior matrix, * is the convolution operation, F*W (s) (t) represents the spatiotemporal perception behavior matrix F and the scale weight matrix W (s) The convolution operation between (t) is used to extract feature information at a specific scale. is the square of the Frobenius norm, ρ s is the scale-dependent threshold, and the scale-adaptive weight ψ s The calculation logic of (t) is: β s is the scale adjustment coefficient; S5: The newly generated spatiotemporal perception behavior vectors are repeatedly stacked in rows to form a spatiotemporal perception behavior matrix, which is input into the wavelet domain risk quantification model to generate a multi-scale risk value to be judged, and a threshold judgment is performed on the multi-scale risk value to be judged. When the threshold judgment result does not meet the threshold condition, a risk warning is issued.

2. The risk early warning method based on dynamic rule construction according to claim 1 is characterized in that: The calculation logic of the behavior timing response feature is: Among them, P dynamic is the behavior time series response feature, T is the number of time points in the time window, t is the tth time point in the time window, LSTM (B t ) is the tth dynamic behavior sequence B t The features of the LSTM network output, η is the decay coefficient, K is the number of behavior patterns, λ k is the weight coefficient of the kth behavior mode, R k (B t ) is the tth dynamic behavior sequence B t The risk measure belonging to the kth behavior pattern, R is the number of dimensions of the behavior time series response characteristics, δ r is the weight coefficient, F r (B t ) is a nonlinear mapping function.

3. The risk early warning method based on dynamic rule construction according to claim 2 is characterized in that: The tth dynamic behavior sequence B t The risk measure R belonging to the kth behavior pattern k (B t ) is calculated as follows: Among them, sigmoid(·) is used to ensure that the contribution of the behavior pattern to the risk measurement is in the range of [0,1], and γ k is the coefficient of the kth behavior mode, J k is the number of features in the kth behavior mode, α k，j is the weight coefficient of the jth feature in the kth behavior mode, B j，t is the dynamic behavior sequence B t The jth feature of Nonlinear mapping function F r (B t ) is calculated as follows: r (B t )=tanh(W r ·B t +b r ), where W r and b r is the transformation parameter used to control the dynamic behavior sequence B t The linear mapping of the data, tanh is the activation function, which is used to compress the output value to the range of [-1,1].

4. The risk early warning method based on dynamic rule construction according to claim 3 is characterized in that: The calculation logic of cross-modal feature fusion is: Among them, F is the spatiotemporal perception behavior vector, D is a two-dimensional tensor representing the interactive information of two-dimensional data, U is the user's static attribute, B is the dynamic behavior sequence, x1 and x2 represent tensor multiplication, σ(·) is the activation function, W is the number of features of the dynamic behavior sequence, α w is the attention weight of the w-th feature, φ w is a nonlinear basis function used to perform nonlinear transformation on the w-th feature, WX w is the weight matrix, is a convolution operation, Pool(B) is a pooling operation on the dynamic behavior sequence B, and tanh is a hyperbolic tangent activation function.

5. A risk warning device based on dynamic rule construction, used to implement the risk warning method based on dynamic rule construction as described in any one of claims 1 to 4, characterized in that: include: The first construction module: extracting user static attributes and dynamic behavior sequences from the points system database, constructing an original data set, and preprocessing the original data set; Time series analysis module: performs time series analysis on the preprocessed dynamic behavior sequence, generates behavior time series response features, performs cross-modal feature fusion based on user static attributes, dynamic behavior sequence and behavior time series response features, and constructs spatiotemporal perception behavior vectors; The second construction module: construct a dynamic rule graph structure according to the risk rules and the corresponding standard spatiotemporal perception behavior vectors, and iteratively update the risk rule weights of the dynamic rule graph structure based on the graph convolutional network; The third construction module: stack all standard spatiotemporal perception behavior vectors in rows to form a spatiotemporal perception behavior matrix. According to the spatiotemporal perception behavior matrix and risk rule weights, a wavelet domain risk quantification model is constructed to generate multi-scale risk values. Risk warning module: The newly generated spatiotemporal perception behavior vectors are repeatedly stacked in rows to form a spatiotemporal perception behavior matrix, the wavelet domain risk quantification model is input, a multi-scale risk value to be judged is generated, a threshold judgment is performed on the multi-scale risk value to be judged, and a risk warning is issued when the threshold judgment result does not meet the threshold condition.

6. An electronic device, characterized in that: The electronic device comprises: one or more processors; A storage device for storing one or more programs, which, when executed by the one or more processors, enables the electronic device to implement the risk warning method based on dynamic rule construction as described in any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed by a processor of a computer, the computer is caused to execute the risk warning method based on dynamic rule construction as described in any one of claims 1 to 4.

Images