User consumption anomaly detection method and device, equipment, storage medium and product

Abstract

According to the user consumption anomaly detection method and device, the equipment, the storage medium and the product provided by the embodiment of the invention, the current consumption behavior data of the user, the cross-category consumption probability of the user and the reference user category are acquired, and the reference user category is determined according to the historical consumption behavior data of the user and the cross-category consumption probability of the user; the cross-category consumption probability of the users is obtained according to statistical analysis of user group consumption behavior data; determining a current user category according to the current consumption behavior data and the cross-category consumption probability of the user; when the current user category is different from the reference user category, the current consumption of the user is abnormal. According to the embodiment of the invention, whether the current consumption behavior of the user is abnormal or not is judged on the basis of the cross-category consumption probability of the user and the historical consumption behavior data of the user, so that the recognition accuracy of the abnormal consumption behavior is improved, the probability of misjudgment of reasonable consumption is reduced, and the user experience is effectively improved.

Description

Technical Field

[0001] This invention relates to the field of consumer safety technology, and in particular to a method, apparatus, equipment, storage medium, and product for detecting abnormal consumer behavior. Background Technology

[0002] Online shopping malls are online service platforms that rely on the internet and support operation on devices such as computers and mobile phones. They provide users with diverse services such as purchasing goods and handling services, and support various online payment methods including electronic account balances, mobile phone balances, and third-party payments. Users can conveniently handle transactions, check order records, and provide feedback without visiting physical stores. Furthermore, the platform features clear product categorization and a variety of special promotions, helping users quickly match their needs.

[0003] However, while these platforms offer convenience, they also pose consumer security risks, potentially leading to financial losses for users. The platforms then need to handle related claims, increasing operational costs. Currently, the industry primarily uses big data models to analyze user consumption behavior data, combining historical consumption patterns to set limiting parameters. If a user's consumption behavior exceeds these parameters, it is considered abnormal, and corresponding measures are taken.

[0004] However, users' consumption behavior is not fixed. In special scenarios such as holiday promotions, normal consumption behavior that exceeds the limit parameters may occur. Such reasonable consumption is easily misjudged as abnormal by the existing system, which reduces the user experience. Summary of the Invention

[0005] This invention provides a method, apparatus, device, storage medium, and product for detecting abnormal user consumption. By using the probability of cross-category consumption and the user's historical consumption behavior data as a basis, it determines whether the user's current consumption behavior is abnormal, thereby improving the accuracy of identifying abnormal consumption behavior, reducing the probability of reasonable consumption being misjudged, and effectively improving the user experience.

[0006] In a first aspect, embodiments of the present invention provide a method for detecting abnormal user consumption, including: The system acquires the user's current consumption behavior data, the probability of the user making cross-category purchases, and a reference user category; wherein the reference user category is determined based on the user's historical consumption behavior data and the probability of the user making cross-category purchases, and the probability of the user making cross-category purchases is obtained through statistical analysis of user group consumption behavior data. The current user category is determined based on the current consumption behavior data and the probability of the user making cross-category purchases; When the current user category is different from the reference user category, the user's current consumption is determined to be abnormal.

[0007] Secondly, embodiments of the present invention also provide a user consumption anomaly detection device, comprising: The data acquisition module is used to acquire the user's current consumption behavior data, the probability of the user's cross-category consumption, and the reference user category; wherein, the reference user category is determined based on the user's historical consumption behavior data and the probability of the user's cross-category consumption, and the probability of the user's cross-category consumption is obtained through statistical analysis of user group consumption behavior data; The category determination module is used to determine the current user category based on the current consumption behavior data and the probability of the user consuming across categories; The anomaly detection module is used to determine that the user's current consumption is abnormal when the current user category is different from the reference user category.

[0008] Thirdly, embodiments of the present invention also provide a user consumption anomaly detection device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the user consumption anomaly detection method as described in any of the above embodiments.

[0009] Fourthly, embodiments of the present invention also provide a computer-readable storage medium, the computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform the user consumption anomaly detection method as described in any of the above embodiments.

[0010] Fifthly, embodiments of the present invention also provide a computer program product, including a computer program / instruction, which, when executed by a processor, implements the user consumption anomaly detection method as described in any of the above embodiments.

[0011] Compared with existing technologies, the user consumption anomaly detection method, apparatus, device, storage medium, and product provided in this invention first acquire the user's current consumption behavior data, the probability of the user's cross-category consumption, and a reference user category. The reference user category is determined based on the user's historical consumption behavior data and the probability of the user's cross-category consumption, and the probability of the user's cross-category consumption is obtained through statistical analysis of user group consumption behavior data. Then, the current user category is determined based on the current consumption behavior data and the probability of the user's cross-category consumption. When the current user category differs from the reference user category, the user's current consumption is deemed abnormal. Therefore, this invention, by using the probability of the user's cross-category consumption and the user's historical consumption behavior data as a basis to determine whether the user's current consumption behavior is abnormal, improves the accuracy of identifying abnormal consumption behavior, reduces the probability of reasonable consumption being misjudged, and effectively improves the user experience. Attached Figure Description

[0012] Figure 1 This is a flowchart illustrating a method for detecting abnormal user consumption according to an embodiment of the present invention; Figure 2 This is a schematic diagram of a graph convolutional neural network processing according to an embodiment of the present invention; Figure 3 This is a user category topology diagram provided in one embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of a user consumption anomaly detection device provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the structure of a user consumption anomaly detection device provided in an embodiment of the present invention. Detailed Implementation

[0013] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0014] See Figure 1 An embodiment of the present invention provides a method for detecting abnormal user consumption, see below. Figure 1 The flowchart of the user consumption anomaly detection method shown includes steps S11 to S13: S11. Obtain the user's current consumption behavior data and reference user category; wherein, the reference user category is determined based on the user's historical consumption behavior data and the probability of the user's cross-category consumption, and the probability of the user's cross-category consumption is obtained through statistical analysis of user group consumption behavior data; S12. Determine the current user category based on the current consumption behavior data; S13. When the current user category is different from the reference user category, the user's current consumption is determined to be abnormal.

[0015] It is worth noting that the method described can be used for consumer risk detection on various existing shopping platforms, such as operator online stores. The method can also be applied to client-side applications, allowing users to analyze their own consumption patterns, etc., without limitation. The operator online store is a comprehensive online service platform built by operators, providing users with diverse services covering core operator businesses such as package selection, broadband application, and smart device purchase, as well as various lifestyle services. Users can initiate payment instructions via the internet using terminals such as computers and mobile phones, supporting various online payment methods such as electronic account balances and phone bill balances. Through the operator online store, users can handle various transactions without leaving home, check transaction lists and records, stay informed about the latest business updates, and submit complaints and suggestions, eliminating the need to visit offline service centers and significantly saving time and effort. Furthermore, the store features clearly categorized device selection zones and a variety of special events, helping users quickly locate and find products and services that meet their needs. Specifically, the core logic of the method is to construct a "reference user category" benchmark based on the user's cross-category consumption probability. By comparing the current consumption category with the reference user category, it determines whether the consumption behavior is abnormal. The key steps of the method's specific process are as follows: 1. Determine three sets of core data.

[0016] (1) Current consumption behavior data: The specific information of the user's current consumption, such as the type of goods purchased, the amount of consumption, and the type of business handled, is the target object for determining "normal / abnormal".

[0017] (2) Probability of cross-category consumption: Based on statistical analysis of the consumption behavior data of user groups, it reflects the general probability that users of a certain category in the user group will exhibit normal consumption behavior in other categories.

[0018] (3) Reference User Category: The consumption preference benchmark category set for users is determined based on the user's own historical consumption behavior data and the probability of users consuming across categories.

[0019] 2. Key Judgment Criteria: Based on the user's current consumption behavior data and the probability of the user making cross-category purchases, the current user category is defined according to the current consumption behavior.

[0020] 3. Anomaly Detection Rules: Category Consistency Comparison. This is the core decision-making step of the solution. If the current user category is the same as the reference user category, the current consumption behavior is determined to be normal consumption. If the current user category is different from the reference user category, the current consumption behavior is determined to be abnormal consumption.

[0021] Compared with existing technologies, the embodiments of the present invention improve the accuracy of identifying abnormal consumption behavior and reduce the probability of reasonable consumption being misjudged by using the probability of users making cross-category consumption and the user's historical consumption behavior data as a basis for determining whether the user's current consumption behavior is abnormal. This effectively improves the user experience.

[0022] In a preferred embodiment, the reference user category is generated in the following manner: Obtain the user's historical consumption behavior data and the probability of the user making cross-category purchases; Extract the common features of the historical consumption behavior data; The reference user category of the user is determined based on the probability of the user consuming across categories and the common characteristics.

[0023] Specifically, the core basis for generating user categories is users' historical consumption behavior data and the probability of users making cross-category purchases. More specifically, its generation is based on common features extracted from users' historical consumption behavior data, as well as the probability of users making cross-category purchases. These common features represent the aggregation of a user's core consumption preferences and are deep feature extractions from historical consumption behavior data. Common features specifically refer to stable and shared key information in a user's historical consumption behavior, rather than scattered single-purchase records. The core purpose of extracting these features is to filter out interference from accidental consumption behavior and accurately capture the core of a user's regular consumption preferences.

[0024] In the process of generating reference user categories, the common characteristics of individual users are calibrated using the group pattern of the probability of users consuming across categories, ultimately determining the reference user category. This fusion of individual characteristics and group patterns ensures that the reference user category closely matches individual user consumption habits, while also using group probability to verify the rationality of category association, making the final generated reference user category more accurate and stable.

[0025] In a preferred embodiment, the reference user category is generated in the following manner: Obtain the user's historical consumption behavior data and the probability of the user making cross-category purchases; The historical consumption behavior data and the probability of the user making cross-category purchases are input into a pre-trained classification model; The backbone network of the classification model is used to extract features from the historical consumption behavior data to obtain the common features of the historical consumption behavior data. The graph convolutional neural network of the classification model is used to process the probability of cross-category consumption by the user and the common features to obtain reference graph convolutional features; The reference user category is obtained by classifying the convolutional features of the reference image using the classifier of the classification model.

[0026] Specifically, this implementation method uses a machine learning model (backbone network + graph convolutional neural network + classifier) ​​to accurately characterize the core consumption attributes of users. First, the backbone network captures the core consumption characteristics of individual users to avoid interference from accidental consumption; then, the graph convolutional neural network incorporates the probability of users' cross-category consumption to verify the rationality of individual characteristics; finally, the output reference user category not only fits the user's personal habits but also conforms to the group's consumption patterns, which is more accurate and stable than simply looking at historical data or simply looking at group patterns.

[0027] For example, assuming historical consumption behavior data is generated in the operator's online store, the steps for generating user categories are as follows: 1. Input the historical consumption behavior data of a single user from the operator's online store into the backbone network of the classification model for feature extraction, extracting common features from the multiple historical consumption behavior data. The backbone network can be a conventional convolutional neural network (CNN) or ResNet, and is not limited here. The backbone network processes the multiple historical consumption behavior data to extract the common features from the multiple historical consumption behavior data.

[0028] 2. The obtained probability of users consuming across categories and the common features extracted in step 1 are input into the graph convolutional neural network of the classification model for processing to obtain graph convolutional features.

[0029] 3. Input the graph convolutional features obtained in step 2 into the classifier of the classification model for classification to obtain the user's user category A, which serves as the reference user category. Specifically, the graph convolutional features are input into the classifier, which maps the input graph convolutional features to probability values ​​belonging to each user category and outputs the category corresponding to the highest probability value as the user category A to which the user belongs.

[0030] In a preferred embodiment, the step of processing the probability of cross-category consumption by the user and the common features using the graph convolutional neural network of the classification model to obtain reference graph convolutional features includes: The probability of a user making cross-category purchases is regularized to obtain a regularization matrix; In the graph convolutional neural network of the classification model, the common features are used as the initial feature vector, and the regularization matrix guides the features to propagate and aggregate layer by layer to obtain the reference graph convolutional features.

[0031] Specifically, graph convolutional neural networks include multiple layers of graph convolutional layers. For example, see [link to example]. Figure 2The diagram illustrates the processing of a graph convolutional neural network (Graph Convolutional Neural Network), which employs three graph convolutional layers. The probability of a user's cross-category consumption exists as a coexistence probability matrix. This matrix is ​​first regularized to obtain a regularization matrix. The shared features are used as initial feature vectors, and together with the regularization matrix, are input into the first graph convolutional layer to obtain the first feature. The first feature and the regularization matrix are then input into the second graph convolutional layer to obtain the second feature. Finally, the second feature and the regularization matrix are input into the third graph convolutional layer to obtain image features, which serve as the user's reference graph convolutional features. Graph convolutional neural networks can effectively mine the potential relationships between multiple historical consumption behavior data, thereby improving the accuracy of subsequent classification and prediction.

[0032] The mathematical expression for the processing of each graph convolutional layer is shown in the following formula: ; in, This represents the features obtained after processing by the l-th graph convolutional layer. Here are the graph convolution parameters for the l-th graph convolutional layer. For regularization matrix, This refers to the activation function. For example, any existing activation function can be chosen, such as the ReLU function.

[0033] It is worth noting that the number of layers in the graph convolutional neural network is not limited to the specific number mentioned above, and can be set according to the actual situation, which is not limited here.

[0034] In a preferred embodiment, determining the current user category based on the current consumption behavior data and the probability of the user making cross-category purchases includes: Input the current consumption behavior data and the probability of the user making cross-category purchases into a pre-trained classification model; The current consumption characteristics are obtained by extracting features from the current consumption behavior data using the backbone network of the classification model. The graph convolutional neural network of the classification model is used to process the probability of cross-category consumption by the user and the current consumption features to obtain the current graph convolutional features; The current user category is obtained by classifying the convolutional features of the current graph using the classifier of the classification model.

[0035] Further, the process of using the graph convolutional neural network of the classification model to process the probability of the user's cross-category consumption and the current consumption features to obtain the current graph convolutional features includes: The probability of a user making cross-category purchases is regularized to obtain a regularization matrix; In the graph convolutional neural network of the classification model, the current consumption feature is used as the initial feature vector, and the regularization matrix guides the feature to propagate and aggregate layer by layer to obtain the current graph convolutional feature.

[0036] Specifically, taking the example of consumer behavior data generated from the operator's online mall, the user's current consumer behavior data in the operator's online mall is input into a classification model for processing to predict the user category B to which the current consumer behavior belongs. If it is the same as the user's reference category, it is considered normal; if it is different from the user's reference category, it is considered abnormal. Specifically, the backbone network processes the user's current consumer behavior data in the operator's online mall to obtain the current consumer characteristics. The current consumer characteristics and the probability of the user's cross-category consumption are input into a graph convolutional neural network for processing to obtain a prediction of the user category B to which the current consumer behavior belongs. In this process, the workflow of the backbone network, graph convolutional neural network, and classifier is the same as the reference user category confirmation process described in the above implementation method, and will not be repeated here.

[0037] In a preferred embodiment, determining the current user category based on the current consumption behavior data and the probability of the user making cross-category purchases includes: Acquire several consumer behavior data of the user, including the current consumer behavior data; Feature extraction is performed on each of the consumer behavior data to be analyzed to obtain several consumer features to be analyzed. Construct a similarity matrix among the consumption features to be analyzed; The consumption characteristics to be analyzed corresponding to the current consumption behavior data are fused with the similarity matrix to obtain fused features; The current user category is determined based on the fusion features and the probability of the user consuming across categories.

[0038] Specifically, to further improve the accuracy of anomaly detection, multiple (first K) consumer behavior data sets, including the user's current consumption behavior data, are taken. A backbone network is used to process each of these data sets to obtain the consumption features of each data set. A similarity matrix is ​​calculated between these consumption features. Then, the similarity matrix and the consumption features of the current consumption behavior data are fused. Finally, the fused features and the probability of the user engaging in cross-category consumption are input into a graph convolutional neural network for processing to obtain a prediction of the user category to which the current consumption behavior belongs. The application principle of the classification model in this process is consistent with the aforementioned content and will not be repeated here. Specifically, the similarity matrix is ​​calculated as follows: For any two consumption features from the multiple consumption behavior data sets, the cosine similarity between the two consumption features is calculated to obtain the similarity matrix.

[0039] In this implementation, a similarity matrix of multiple consumption behavior data, including the user's current consumption behavior data, is integrated. The subsequent graph convolutional neural network is used to fully explore the similarity relationships between the multiple consumption behavior data, including the user's current consumption behavior data, so as to avoid the influence of the user making normal consumption behaviors that do not belong to the category and improve the detection accuracy.

[0040] In a preferred embodiment, the probability of a user making cross-category purchases exists in the form of a coexistence probability matrix, which is generated in the following manner: Obtain the consumption behavior data of the aforementioned user group; Statistical analysis is performed on the consumption behavior data of the user group to divide the user group into several user categories and construct a user category topology graph; wherein, the nodes of the user category topology graph represent the user category, and the lines between the nodes represent the probability that the user categories of the connected nodes coexist. Extract the coexistence probability matrix from the user category topology graph.

[0041] For example, the probability of a user making cross-category purchases exists in the form of a coexistence probability matrix, which is generated in the following way: 1. Based on prior knowledge, construct a topology graph for each user category.

[0042] like Figure 3 The user category topology diagram shown initially divides users into n categories (e.g., based on existing user consumption behavior data in the operator's online store) Figure 3 We have n user categories (A, B, C, N) and construct a topology graph of these n categories. The topology graph consists of nodes and connections between them. Nodes represent user categories, and connections between nodes represent the probability of two categories coexisting.

[0043] for example, Figure 3 In the graph, 0.02 represents the probability that a user belongs to category A but exhibits consumption behavior similar to that of a user from category B (since consumption behavior of users in category A is not always strictly category A, it may occasionally include normal consumption behavior from other categories), and 0.17 represents the probability that a user belongs to category B but exhibits consumption behavior similar to that of a user from category A. The coexistence probability is directional; that is, the connection between the two nodes is directional, with A pointing to B, representing the probability that a user belongs to category A but exhibits consumption behavior similar to that of a user from category B.

[0044] It's worth noting that in practical applications, by summarizing and statistically analyzing a large number of existing consumer behaviors, more accurate categorization can be achieved, and the coexistence probability of each category can be obtained. The larger the total number of consumer behaviors analyzed, the closer the coexistence probability value will be to the true value of its distribution in the real world.

[0045] 2. Extract the coexistence probability matrix of each category from the topological graph.

[0046] Representing the relationships between nodes in the topology graph using matrices yields coexistence probability matrices for each category. For example, if users are divided into 10 categories, the coexistence probability matrix would be a 10×10 matrix, where the value at position (i, j) represents the probability that the consumption behavior of the j-th category occurs in the i-th category.

[0047] In one implementation, when the user's current consumption is abnormal, the consumption system is triggered to perform risk control operations.

[0048] Optionally, the risk control operation includes risk warning operation and / or secondary verification operation.

[0049] Optionally, the consumption system is an operator system.

[0050] For example, abnormal user spending behavior can be reported to the operator, who can then manage the risk by issuing risk warnings, requiring secondary verification, etc., to prevent users from suffering financial losses.

[0051] Compared with existing technologies, the embodiments of the present invention determine whether a user's current consumption behavior is abnormal by relying on the probability of cross-category consumption and the user's historical consumption behavior data. Theoretically, this has high market application potential, especially in e-commerce, financial risk control, telecommunications services, and other fields, and has the following beneficial effects: 1. Fraud Prevention: By detecting unusual consumer behavior, potential fraudulent activities can be identified in a timely manner, thereby reducing fraudulent losses. This is especially true in the financial and e-commerce sectors, where such detection can significantly reduce economic losses caused by fraudulent activities.

[0052] 2. Improve customer satisfaction: By detecting abnormal spending behavior, irregularities in customer accounts can be identified and resolved promptly, thereby increasing customer satisfaction and trust in the service. This helps maintain customer loyalty and reduce customer churn.

[0053] 3. Optimize operational efficiency: Automated anomaly detection systems reduce the workload of manual review, thereby lowering operating costs. Furthermore, such systems can operate 24 / 7, improving detection efficiency and response speed.

[0054] 3. Data-driven decision-making: By analyzing consumer behavior data, potential market trends and opportunities can be identified, providing data support for strategic decision-making. For example, marketing strategies can be adjusted and product mix optimized based on unusual consumer behavior.

[0055] 4. Enhance risk management: By detecting anomalies in consumer behavior, businesses can better assess and manage the risks they face. This helps them become more flexible and resilient in the face of market changes, thereby maintaining a competitive advantage.

[0056] 5. Regulatory Compliance: In certain industries, particularly the financial and insurance sectors, regulatory agencies require companies to possess specific risk management and anomaly detection capabilities. By implementing anomaly detection systems, companies can better meet regulatory requirements and avoid fines and other legal risks arising from non-compliance.

[0057] See Figure 4 An embodiment of the present invention also provides a user consumption anomaly detection device, comprising: The data acquisition module 21 is used to acquire the user's current consumption behavior data, the probability of the user's cross-category consumption, and the reference user category; wherein, the reference user category is determined based on the user's historical consumption behavior data and the probability of the user's cross-category consumption, and the probability of the user's cross-category consumption is obtained through statistical analysis of user group consumption behavior data; Category determination module 22 is used to determine the current user category based on the current consumption behavior data and the probability of the user consuming across categories; The anomaly detection module 23 is used to determine that the user's current consumption is abnormal when the current user category is different from the reference user category.

[0058] In one implementation, the category determination module 22 is further configured to: Obtain the user's historical consumption behavior data and the probability of the user making cross-category purchases; Extract the common features of the historical consumption behavior data; The reference user category of the user is determined based on the probability of the user consuming across categories and the common characteristics.

[0059] In one implementation, the category determination module 22 is further configured to: Obtain the user's historical consumption behavior data and the probability of the user making cross-category purchases; The historical consumption behavior data and the probability of the user making cross-category purchases are input into a pre-trained classification model; The backbone network of the classification model is used to extract features from the historical consumption behavior data to obtain the common features of the historical consumption behavior data. The graph convolutional neural network of the classification model is used to process the probability of cross-category consumption by the user and the common features to obtain reference graph convolutional features; The reference user category is obtained by classifying the convolutional features of the reference image using the classifier of the classification model.

[0060] In one implementation, the graph convolutional neural network of the classification model is used to process the probability of cross-category consumption by the user and the common features to obtain reference graph convolutional features, including: The probability of a user making cross-category purchases is regularized to obtain a regularization matrix; In the graph convolutional neural network of the classification model, the common features are used as the initial feature vector, and the regularization matrix guides the features to propagate and aggregate layer by layer to obtain the reference graph convolutional features.

[0061] In one implementation, the category determination module 22 is specifically used for: Input the current consumption behavior data and the probability of the user making cross-category purchases into a pre-trained classification model; The current consumption characteristics are obtained by extracting features from the current consumption behavior data using the backbone network of the classification model. The graph convolutional neural network of the classification model is used to process the probability of cross-category consumption by the user and the current consumption features to obtain the current graph convolutional features; The current user category is obtained by classifying the convolutional features of the current graph using the classifier of the classification model.

[0062] In one implementation, the process of using the graph convolutional neural network of the classification model to process the probability of cross-category consumption by the user and the current consumption features to obtain current graph convolutional features includes: The probability of a user making cross-category purchases is regularized to obtain a regularization matrix; In the graph convolutional neural network of the classification model, the current consumption feature is used as the initial feature vector, and the regularization matrix guides the feature to propagate and aggregate layer by layer to obtain the current graph convolutional feature.

[0063] In one implementation, the category determination module 22 is specifically used for: Acquire several consumer behavior data of the user, including the current consumer behavior data; Feature extraction is performed on each of the consumer behavior data to be analyzed to obtain several consumer features to be analyzed. Construct a similarity matrix among the consumption features to be analyzed; The consumption characteristics to be analyzed corresponding to the current consumption behavior data are fused with the similarity matrix to obtain fused features; The current user category is determined based on the fusion features and the probability of the user consuming across categories.

[0064] In one implementation, the probability of a user making cross-category consumption exists in the form of a coexistence probability matrix, and the device further includes a matrix generation module for: Obtain the consumption behavior data of the aforementioned user group; Statistical analysis is performed on the consumption behavior data of the user group to divide the user group into several user categories and construct a user category topology graph; wherein, the nodes of the user category topology graph represent the user category, and the lines between the nodes represent the probability that the user categories of the connected nodes coexist. Extract the coexistence probability matrix from the user category topology graph.

[0065] In one embodiment, the device further includes a risk control module for: When a user's current consumption is abnormal, the consumption system is triggered to perform risk control operations.

[0066] In one implementation, the risk control operation includes a risk alert operation and / or a secondary verification operation.

[0067] In one implementation, the consumption system is an operator system.

[0068] In one implementation, the current consumption behavior data refers to the consumption behavior data generated by the user on the operator's online mall.

[0069] In one embodiment, the anomaly detection module 23 is further configured to: determine that the user's current consumption is normal when the current user category is the same as the reference user category.

[0070] It is worth noting that the specific working process of the user consumption anomaly detection device described in the embodiments of the present invention can refer to the working process of the user consumption anomaly detection method described in any of the above embodiments, and will not be repeated here.

[0071] See Figure 5 This invention also provides a user consumption anomaly detection device, including a processor 31, a memory 32, and a computer program stored in the memory 32 and configured to be executed by the processor 31. When the processor 31 executes the computer program, it implements the steps described in the above-described user consumption anomaly detection method embodiments, for example... Figure 1 The steps S11 to S13 described above; or, when the processor 31 executes the computer program, it implements the functions of each module in the above-described device embodiments.

[0072] For example, the computer program can be divided into one or more modules, which are stored in the memory 32 and executed by the processor 31 to complete the present invention. The one or more modules can be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in the user consumption anomaly detection device. For example, the computer program can be divided into multiple modules. The specific working process of each module can be referred to the working process of the operation and maintenance processing scheme model described in the above embodiments, and will not be repeated here.

[0073] The user consumption anomaly detection device can be a desktop computer, laptop, handheld computer, or cloud server, etc. The user consumption anomaly detection device may include, but is not limited to, a processor 31 and a memory 32. Those skilled in the art will understand that the user consumption anomaly detection device may also include input / output devices, network access devices, buses, etc.

[0074] The processor 31 can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. The processor 31 is the control center of the user consumption anomaly detection device, connecting all parts of the device via various interfaces and lines.

[0075] The memory 32 can be used to store the computer program and / or modules. The processor 31 implements various functions of the user consumption anomaly detection device by running or executing the computer program and / or modules stored in the memory 32 and calling the data stored in the memory 32. The memory 32 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as image playback function), etc.; the data storage area may store data created based on the use of the mobile phone, etc. In addition, the memory 32 may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0076] If the module integrated into the user consumption anomaly detection device is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by the processor 31, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0077] This invention also provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the user consumption anomaly detection method as described in any of the above embodiments.

[0078] Compared with existing technologies, the user consumption anomaly detection method, apparatus, device, storage medium, and product provided in this invention first acquire the user's current consumption behavior data, the probability of the user's cross-category consumption, and a reference user category. The reference user category is determined based on the user's historical consumption behavior data and the probability of the user's cross-category consumption, and the probability of the user's cross-category consumption is obtained through statistical analysis of user group consumption behavior data. Then, the current user category is determined based on the current consumption behavior data and the probability of the user's cross-category consumption. When the current user category differs from the reference user category, the user's current consumption is deemed abnormal. Therefore, this invention, by using the probability of the user's cross-category consumption and the user's historical consumption behavior data as a basis to determine whether the user's current consumption behavior is abnormal, improves the accuracy of identifying abnormal consumption behavior, reduces the probability of reasonable consumption being misjudged, and effectively improves the user experience.

[0079] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for detecting abnormal user consumption, characterized in that, include: The system acquires the user's current consumption behavior data, the probability of the user making cross-category purchases, and a reference user category; wherein the reference user category is determined based on the user's historical consumption behavior data and the probability of the user making cross-category purchases, and the probability of the user making cross-category purchases is obtained through statistical analysis of user group consumption behavior data. The current user category is determined based on the current consumption behavior data and the probability of the user making cross-category purchases; When the current user category is different from the reference user category, the user's current consumption is determined to be abnormal.

2. The user consumption anomaly detection method as described in claim 1, characterized in that, The reference user category is generated in the following way: Obtain the user's historical consumption behavior data and the probability of the user making cross-category purchases; Extract the common features of the historical consumption behavior data; The reference user category of the user is determined based on the probability of the user consuming across categories and the common characteristics.

3. The user consumption anomaly detection method as described in claim 1, characterized in that, The reference user category is generated in the following way: Obtain the user's historical consumption behavior data and the probability of the user making cross-category purchases; The historical consumption behavior data and the probability of the user making cross-category purchases are input into a pre-trained classification model; The backbone network of the classification model is used to extract features from the historical consumption behavior data to obtain the common features of the historical consumption behavior data. The graph convolutional neural network of the classification model is used to process the probability of cross-category consumption by the user and the common features to obtain reference graph convolutional features; The reference user category is obtained by classifying the convolutional features of the reference image using the classifier of the classification model.

4. The user consumption anomaly detection method as described in claim 3, characterized in that, The graph convolutional neural network of the classification model is used to process the probability of cross-category consumption by the user and the common features to obtain reference graph convolutional features, including: The probability of a user making cross-category purchases is regularized to obtain a regularization matrix; In the graph convolutional neural network of the classification model, the common features are used as the initial feature vector, and the regularization matrix guides the features to propagate and aggregate layer by layer to obtain the reference graph convolutional features.

5. The user consumption anomaly detection method as described in claim 1, characterized in that, The step of determining the current user category based on the current consumption behavior data and the probability of the user making cross-category purchases includes: Input the current consumption behavior data and the probability of the user making cross-category purchases into a pre-trained classification model; The current consumption characteristics are obtained by extracting features from the current consumption behavior data using the backbone network of the classification model. The graph convolutional neural network of the classification model is used to process the probability of cross-category consumption by the user and the current consumption features to obtain the current graph convolutional features; The current user category is obtained by classifying the convolutional features of the current graph using the classifier of the classification model.

6. The user consumption anomaly detection method as described in claim 5, characterized in that, The graph convolutional neural network of the classification model is used to process the probability of cross-category consumption by the user and the current consumption features to obtain the current graph convolutional features, including: The probability of a user making cross-category purchases is regularized to obtain a regularization matrix; In the graph convolutional neural network of the classification model, the current consumption feature is used as the initial feature vector, and the regularization matrix guides the feature to propagate and aggregate layer by layer to obtain the current graph convolutional feature.

7. The user consumption anomaly detection method as described in claim 1, characterized in that, The step of determining the current user category based on the current consumption behavior data and the probability of the user making cross-category purchases includes: Acquire several consumer behavior data of the user, including the current consumer behavior data; Feature extraction is performed on each of the consumer behavior data to be analyzed to obtain several consumer features to be analyzed. Construct a similarity matrix among the consumption features to be analyzed; The consumption characteristics to be analyzed corresponding to the current consumption behavior data are fused with the similarity matrix to obtain fused features; The current user category is determined based on the fusion features and the probability of the user consuming across categories.

8. The user consumption anomaly detection method as described in any one of claims 1 to 7, characterized in that, The probability of a user making cross-category purchases exists in the form of a coexistence probability matrix, which is generated in the following way: Obtain the consumption behavior data of the aforementioned user group; Statistical analysis is performed on the consumption behavior data of the user group to divide the user group into several user categories and construct a user category topology graph; wherein, the nodes of the user category topology graph represent the user category, and the lines between the nodes represent the probability that the user categories of the connected nodes coexist. Extract the coexistence probability matrix from the user category topology graph.

9. The user consumption anomaly detection method as described in claim 1, characterized in that, Also includes: When a user's current consumption is abnormal, the consumption system is triggered to perform risk control operations.

10. The user consumption anomaly detection method as described in claim 9, characterized in that, The risk control operations include risk alert operations and / or secondary verification operations.

11. The user consumption anomaly detection method as described in claim 9, characterized in that, The consumption system is an operator system.

12. The user consumption anomaly detection method as described in claim 1, characterized in that, The current consumption behavior data refers to the consumption behavior data generated by the user on the operator's online mall.

13. The user consumption anomaly detection method as described in claim 1, characterized in that, Also includes: When the current user category is the same as the reference user category, the user's current consumption is determined to be normal.

14. A device for detecting abnormal user spending, characterized in that, include: The data acquisition module is used to acquire the user's current consumption behavior data, the probability of the user's cross-category consumption, and the reference user category; wherein, the reference user category is determined based on the user's historical consumption behavior data and the probability of the user's cross-category consumption, and the probability of the user's cross-category consumption is obtained through statistical analysis of user group consumption behavior data; The category determination module is used to determine the current user category based on the current consumption behavior data and the probability of the user consuming across categories; The anomaly detection module is used to determine that the user's current consumption is abnormal when the current user category is different from the reference user category.

15. The user consumption anomaly detection device as described in claim 14, characterized in that, The category determination module is also used for: Obtain the user's historical consumption behavior data and the probability of the user making cross-category purchases; Extract the common features of the historical consumption behavior data; The reference user category of the user is determined based on the probability of the user consuming across categories and the common characteristics.

16. The user consumption anomaly detection device as described in claim 14, characterized in that, The category determination module is also used for: Obtain the user's historical consumption behavior data and the probability of the user making cross-category purchases; The historical consumption behavior data and the probability of the user making cross-category purchases are input into a pre-trained classification model; The backbone network of the classification model is used to extract features from the historical consumption behavior data to obtain the common features of the historical consumption behavior data. The graph convolutional neural network of the classification model is used to process the probability of cross-category consumption by the user and the common features to obtain reference graph convolutional features; The reference user category is obtained by classifying the convolutional features of the reference image using the classifier of the classification model.

17. The user consumption anomaly detection device as described in claim 16, characterized in that, The graph convolutional neural network of the classification model is used to process the probability of cross-category consumption by the user and the common features to obtain reference graph convolutional features, including: The probability of a user making cross-category purchases is regularized to obtain a regularization matrix; In the graph convolutional neural network of the classification model, the common features are used as the initial feature vector, and the regularization matrix guides the features to propagate and aggregate layer by layer to obtain the reference graph convolutional features.

18. The user consumption anomaly detection device as described in claim 14, characterized in that, The category determination module is specifically used for: Input the current consumption behavior data and the probability of the user making cross-category purchases into a pre-trained classification model; The current consumption characteristics are obtained by extracting features from the current consumption behavior data using the backbone network of the classification model. The graph convolutional neural network of the classification model is used to process the probability of cross-category consumption by the user and the current consumption features to obtain the current graph convolutional features; The current user category is obtained by classifying the convolutional features of the current graph using the classifier of the classification model.

19. The user consumption anomaly detection device as described in claim 18, characterized in that, The graph convolutional neural network of the classification model is used to process the probability of cross-category consumption by the user and the current consumption features to obtain the current graph convolutional features, including: The probability of a user making cross-category purchases is regularized to obtain a regularization matrix; In the graph convolutional neural network of the classification model, the current consumption feature is used as the initial feature vector, and the regularization matrix guides the feature to propagate and aggregate layer by layer to obtain the current graph convolutional feature.

20. The user consumption anomaly detection device as described in claim 14, characterized in that, The category determination module is specifically used for: Acquire several consumer behavior data of the user, including the current consumer behavior data; Feature extraction is performed on each of the consumer behavior data to be analyzed to obtain several consumer features to be analyzed. Construct a similarity matrix among the consumption features to be analyzed; The consumption characteristics to be analyzed corresponding to the current consumption behavior data are fused with the similarity matrix to obtain fused features; The current user category is determined based on the fusion features and the probability of the user consuming across categories.

21. The user consumption anomaly detection device as described in any one of claims 14 to 20, characterized in that, The probability of a user making cross-category purchases exists in the form of a coexistence probability matrix, and the device further includes a matrix generation module for: Obtain the consumption behavior data of the aforementioned user group; Statistical analysis is performed on the consumption behavior data of the user group to divide the user group into several user categories and construct a user category topology graph; wherein, the nodes of the user category topology graph represent the user category, and the lines between the nodes represent the probability that the user categories of the connected nodes coexist. Extract the coexistence probability matrix from the user category topology graph.

22. The user consumption anomaly detection device as described in claim 14, characterized in that, The device also includes a risk control module for: When a user's current consumption is abnormal, the consumption system is triggered to perform risk control operations.

23. The user consumption anomaly detection device as described in claim 22, characterized in that, The risk control operations include risk alert operations and / or secondary verification operations.

24. The user consumption anomaly detection device as described in claim 22, characterized in that, The consumption system is an operator system.

25. A device for detecting abnormal user spending, characterized in that, The device includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements the user consumption anomaly detection method as described in any one of claims 1 to 13.

26. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device on which the computer-readable storage medium is located to perform the user consumption anomaly detection method as described in any one of claims 1 to 13.

27. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the user consumption anomaly detection method as described in any one of claims 1 to 13.

Images