Method and device for identifying fraudulent behavior
By acquiring and analyzing rights and interests, and using the decision tree model to identify fraud, the problem of poor identification of fraud in the existing technology is solved, the identification accuracy is improved, and user rights and interests are protected.
Patent Information
- Application Number
- CN202410927650.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-11
- Publication Date
- 2024-11-22
AI Technical Summary
The prior art has poor accuracy in identifying fraudulent behavior, resulting in wasting user rights and resources, making it difficult to effectively prevent fraudulent behavior.
Obtain information and user information by obtaining the rights and interests of the target resources, divide behavior characteristics, and classify behavior characteristics and user information based on the decision tree classification model. When the behavior classification results match the preset fraudulent behavior strategy, determine the target fraud user and behavior , combine multi-dimensional analysis of behavioral characteristics and decision tree model for identification.
It improves the accuracy of identifying fraudulent behaviors in users' collection of rights and interests, reduces the waste of rights and interests, and ensures the security of user rights and interests.
Smart Images

Figure CN119004277A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of data security technology, and in particular to a method and device for identifying fraudulent behavior. Background Art
[0002] With the popularity of big data technology in all walks of life, there are endless cases of using fake identities on the Internet to make huge profits in major online marketing activities. Among them, in order to ensure the normal operation of online marketing activities, to achieve the effectiveness and security of protecting user rights and interests, and to avoid the waste of rights and interests resources, it is usually necessary to exclude fraudulent behavior. At present, the channels for issuing coupons as rights and interests resources are restricted in quantity and amount, so that they can be identified as fraudulent behavior when the limit is exceeded. However, when setting quantity limits or amounts to be collected, a large number of user rights and interests resources will still be defrauded by fraudulent behavior, resulting in a large waste of rights and interests resources and the inefficiency of data resource use. Summary of the invention
[0003] In view of this, the present invention provides a method and device for identifying fraudulent behavior, the main purpose of which is to solve the problem that the accuracy of existing fraudulent behavior identification is poor, resulting in a large amount of equity resources being wasted.
[0004] According to one aspect of the present invention, a method for identifying fraudulent behavior is provided, comprising:
[0005] Acquire rights and interests acquisition information and user information of the target resource, wherein the rights and interests acquisition information is used to represent the content of obtaining rights and interests information of the target resource from different business channels;
[0006] Performing behavioral feature division on the rights acquisition information to obtain at least one behavioral feature, and performing behavioral classification on the behavioral feature and the user information based on a decision tree classification model to obtain a behavioral classification result;
[0007] When the behavior classification result matches at least one fraudulent behavior in the preset fraudulent behavior strategy, a target fraudulent user and a target fraudulent behavior are determined.
[0008] Furthermore, the performing of behavioral feature classification on the equity acquisition information to obtain at least one behavioral feature includes:
[0009] Analyze at least one of the number of behaviors, the content of the benefits, and the amount of resources in the benefit acquisition information;
[0010] The behavior knowledge graph is used to query the behavior features that match at least one of the number of behaviors, the rights and interests content, and the number of resources. The behavior knowledge graph stores behavior features corresponding to different numbers of behaviors, different rights and interests content, and different numbers of resources.
[0011] Furthermore, before the behavior classification is performed on the behavior features and the user information based on the decision tree classification model and the behavior classification result is obtained, the method further includes:
[0012] Creating an initialized decision tree classification model and obtaining a behavior classification sample set, wherein the behavior classification sample set includes behavior feature samples corresponding to different behavior classification labels and user samples;
[0013] Training the initialized decision tree classification model based on the behavior classification sample set to obtain the decision tree classification model that has completed model training;
[0014] Among them, each classification node in the decision tree classification model that has completed training is marked with the inventory of classified behavior feature samples and user samples, as well as the trained classification tree path.
[0015] Furthermore, before determining the target fraudulent user and the target fraudulent behavior, the method further includes:
[0016] If the target classification tree path corresponding to the behavior classification result is an untrained classification tree path, then the behavior feature sample stock and the user sample stock in all classification nodes corresponding to the target classification tree path are retrieved;
[0017] If the behavior feature sample stock and the user sample stock are greater than the behavior feature sample stock and the user sample stock of the classification node in the reference classification tree path, then the step of matching the behavior classification result with the preset fraud behavior strategy is performed;
[0018] If the behavior feature sample stock and the user sample stock are less than or equal to the behavior feature sample stock and the user sample stock of the classification node in the reference classification tree path, the behavior in the behavior classification result is determined to be non-fraudulent behavior.
[0019] Furthermore, after the behavior characteristics and the user information are classified based on the decision tree classification model and the behavior classification result is obtained, the method further includes:
[0020] Comparing the behavior classification result with at least one fraudulent behavior in the preset fraudulent behavior strategy;
[0021] If the behavior classification result does not match all the fraudulent behaviors in the preset fraudulent behavior strategy, a non-fraudulent user and a non-fraudulent behavior are determined.
[0022] Furthermore, before the behavior classification is performed on the behavior features and the user information based on the decision tree classification model and the behavior classification result is obtained, the method further includes:
[0023] Matching the behavior characteristics with business behavior rules, wherein the business behavior rules include at least one of a fraud group identification rule, a business anomaly identification rule, and a fraud individual identification rule;
[0024] If the behavior feature matches the business behavior rule, at least one of a fraudulent group of users, a fraudulent individual user, and a business abnormality user is determined, and user information of at least one of the fraudulent group of users, the fraudulent individual user, and the business abnormality user is used as user information for executing the behavior classification step.
[0025] Furthermore, after determining the target fraudulent user and the target fraudulent behavior, the method further includes:
[0026] The rights and interests information obtained by the target fraud user is cancelled, and the target fraud user is marked, so as to reissue the rights and interests information of the target resource to a normal user.
[0027] According to another aspect of the present invention, there is provided a device for identifying fraudulent behavior, comprising:
[0028] An acquisition module, used to acquire rights acquisition information and user information of a target resource, wherein the rights acquisition information is used to represent the content of obtaining rights information of the target resource from different business channels;
[0029] A classification module, configured to classify the rights acquisition information into behavior characteristics to obtain at least one behavior characteristic, and to classify the behavior characteristics and the user information into behavior characteristics based on a decision tree classification model to obtain a behavior classification result;
[0030] The determination module is used to determine a target fraud user and a target fraud behavior when the behavior classification result matches at least one fraud behavior in a preset fraud behavior strategy.
[0031] Furthermore, the classification module is specifically used to parse at least one of the number of behaviors, the rights and interests content, and the number of resources in the rights and interests acquisition information; and to query the behavior features that match at least one of the number of behaviors, the rights and interests content, and the number of resources according to the behavior knowledge graph, and the behavior knowledge graph stores the behavior features corresponding to different numbers of behaviors, different rights and interests content, and different numbers of resources.
[0032] Furthermore, the device also includes:
[0033] A creation module is used to create an initialization decision tree classification model and obtain a behavior classification sample set, wherein the behavior classification sample set includes behavior feature samples corresponding to different behavior classification labels and user samples;
[0034] A training module, used to train the initialized decision tree classification model based on the behavior classification sample set to obtain the decision tree classification model that has completed model training;
[0035] Among them, each classification node in the decision tree classification model that has completed training is marked with the inventory of classified behavior feature samples and user samples, as well as the trained classification tree path.
[0036] Furthermore, the device further comprises: a calling module, an execution module,
[0037] The retrieval module is used to retrieve the behavior feature sample stocks and user sample stocks in all classification nodes corresponding to the target classification tree path if the target classification tree path corresponding to the behavior classification result is an untrained classification tree path;
[0038] The execution module is used to execute the step of matching the behavior classification result with the preset fraud behavior strategy if the behavior feature sample inventory and the user sample inventory are greater than the behavior feature sample inventory and the user sample inventory of the classification node in the reference classification tree path;
[0039] The determination module is also used to determine that the behavior in the behavior classification result is non-fraudulent behavior if the behavior feature sample stock and the user sample stock are less than or equal to the behavior feature sample stock and the user sample stock of the classification node in the reference classification tree path.
[0040] Furthermore, the device further comprises: a comparison module,
[0041] The comparison module is used to compare the behavior classification result with at least one fraudulent behavior in the preset fraudulent behavior strategy;
[0042] The determination module is further configured to determine a non-fraudulent user and a non-fraudulent behavior if the behavior classification result does not match all the fraudulent behaviors in the preset fraudulent behavior strategy.
[0043] Furthermore, the device further comprises: a matching module,
[0044] The matching module is used to match the behavior characteristics with business behavior rules, wherein the business behavior rules include at least one of a fraud group identification rule, a business anomaly identification rule, and a fraud individual identification rule;
[0045] The determination module is also used to determine at least one of a fraudulent group of users, a fraudulent individual user, and a business abnormality user if the behavior feature matches the business behavior rule, and use the user information of at least one of the fraudulent group of users, the fraudulent individual user, and the business abnormality user as the user information for executing the behavior classification step.
[0046] Furthermore, the device also includes:
[0047] The marking module is used to cancel the rights and interests information obtained by the target fraud user and mark the target fraud user so as to reissue the rights and interests information of the target resource to a normal user.
[0048] According to another aspect of the present invention, a storage medium is provided, wherein at least one executable instruction is stored in the storage medium, and the executable instruction enables a processor to execute operations corresponding to the above-mentioned fraudulent behavior identification method.
[0049] According to another aspect of the present invention, there is provided a terminal, comprising: a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus;
[0050] The memory is used to store at least one executable instruction, and the executable instruction enables the processor to execute operations corresponding to the above-mentioned fraudulent behavior identification method.
[0051] By means of the above technical solution, the technical solution provided by the embodiment of the present invention has at least the following advantages:
[0052] The present invention provides a method and device for identifying fraudulent behavior. Compared with the prior art, the embodiment of the present invention obtains rights acquisition information and user information of a target resource, wherein the rights acquisition information is used to characterize the content of obtaining rights information of the target resource from different business channels; the rights acquisition information is divided into behavioral characteristics to obtain at least one behavioral characteristic, and the behavioral characteristics and the user information are behaviorally classified based on a decision tree classification model to obtain a behavior classification result; when the behavior classification result matches at least one fraudulent behavior in a preset fraudulent behavior strategy, a target fraudulent user and a target fraudulent behavior are determined, and identification is performed through multi-dimensional analysis of behavioral characteristics combined with a decision tree model, thereby improving the accuracy of identifying fraudulent behavior in users receiving rights information.
[0053] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented according to the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are listed below. BRIEF DESCRIPTION OF THE DRAWINGS
[0054] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:
[0055] Figure 1 A flow chart of a fraudulent behavior identification method provided by an embodiment of the present invention is shown;
[0056] Figure 2 A fraud identification system architecture diagram provided by an embodiment of the present invention is shown;
[0057] Figure 3 A schematic diagram of a column separation point provided by an embodiment of the present invention is shown;
[0058] Figure 4 A block diagram of a fraud identification device provided by an embodiment of the present invention is shown;
[0059] Figure 5 A schematic diagram of the structure of a terminal provided by an embodiment of the present invention is shown. DETAILED DESCRIPTION
[0060] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.
[0061] The embodiment of the present invention provides a method for identifying fraudulent behavior, such as Figure 1 As shown, the method includes:
[0062] 101. Obtain rights acquisition information and user information for target resources.
[0063] In the embodiment of the present invention, resources are used to represent objects to be traded in different business transaction scenarios, including but not limited to commodity resources, service resources, project resources, comment resources, etc., and rights and interests information is used to represent the preferential content enjoyed when resources are expected to be traded, including but not limited to coupons, discounts, discounts, red envelopes, etc. The current execution subject is a service end or processor for users to receive resource rights and interests. After detecting that the rights and interests information of the target resource issued by the rights and interests issuance platform is issued, the current execution end obtains the resource acquisition information and user information of the target resource. Among them, the rights and interests acquisition information is used to represent the content of the rights and interests information acquisition of the target resource from different business channels, including but not limited to the number of times, time, frequency, etc. of obtaining the rights and interests information, and the user information may include the user name, user account, user address, integral flow table, transaction channel table, customer basic information, activity information table, login information, e-commerce information, positioning information, network information, etc. of the user who obtains the rights and interests information, which is not specifically limited in the embodiment of the present invention.
[0064] It should be noted that after the current execution end detects that different users have collected the rights and interests information of the target resources, the current execution end sends a suspension cancellation instruction to the rights and interests cancellation platform to instruct the rights and interests cancellation platform to determine whether to perform rights and interests cancellation again after the current execution end executes steps 101-103. The embodiment of the present invention does not make specific limitations. At this time, the current execution end can perform steps 101-103 as a fraud identification platform. In addition, different resource manufacturers can issue rights and interests on the rights and interests issuance platform. At this time, the issuance target, issuance time, issuance scope, coupon usage rules and other information can be set for the issued rights and interests information. For example, after the coupon is successfully issued, it will enter the user's account and be viewed and used on the My Coupon page. The embodiment of the present invention does not make specific limitations.
[0065] 102. Performing behavioral feature division on the rights acquisition information to obtain at least one behavioral feature, and performing behavioral classification on the behavioral features and the user information based on a decision tree classification model to obtain a behavioral classification result.
[0066] In the embodiment of the present invention, since the rights acquisition information may include the number of times, time, frequency, etc. of obtaining the rights information, the rights acquisition information is further characterized to determine the behavioral characteristics of different users in obtaining the rights information. Among them, the behavioral characteristics are used to characterize the characteristics of the user's receipt of the rights information, including but not limited to the time characteristics, location characteristics, frequency characteristics, terminal device characteristics, etc. of receiving the rights information. At this time, the behavioral characteristics can be divided by the data identifier in the rights acquisition information, and the embodiment of the present invention does not make specific limitations.
[0067] It should be noted that in the embodiment of the present invention, in order to accurately identify fraudulent behavior, the current execution end classifies the behavior characteristics and user information based on the decision tree classification model that has completed model training, thereby identifying whether there is any user behavior of fraudulently claiming rights for the target resource. Among them, the decision tree classification model can be constructed based on a binary tree or a non-binary tree, so that the decision tree model is trained based on a behavior classification sample set.
[0068] 103. When the behavior classification result matches at least one fraudulent behavior in a preset fraudulent behavior strategy, determine a target fraudulent user and a target fraudulent behavior.
[0069] In an embodiment of the present invention, after the current execution end obtains the behavior classification result, it determines whether the behavior classification result is a fraudulent behavior according to the preset fraud behavior strategy. Among them, the preset fraud behavior strategy includes fraudulent behaviors pre-classified by different behavior classification results, so as to make a comparative judgment. In addition, since there can be multiple or one user who can obtain the rights and interests information of the target resource, after classifying the behaviors of different users, the behavior classification results corresponding to multiple different users can be obtained, and then when matching based on the preset fraud behavior strategy, one target fraud user or multiple target fraud users of the target fraud behavior are determined, and the embodiment of the present invention does not make specific limitations.
[0070] In a specific implementation scenario, after the behavior classification results of different users are obtained, if the preset fraud behavior strategy is to identify users with the same IP address who have multiple fraud behaviors, then it is determined that the target user has individual fraud behavior, which is not specifically limited in the embodiment of the present invention. In addition, the determined target fraud users can be divided according to the fraud subject, including first-party fraud, second-party fraud, third-party fraud, etc., which is not specifically limited in the embodiment of the present invention.
[0071] It should be noted that in the embodiments of the present invention, fraudulent behavior may include but is not limited to three categories: financial credit fraud, Internet business fraud, and credit card fraud. According to the fraudulent behavior, it can be divided into three categories: financial credit fraud, Internet business fraud, and credit card fraud. If it is further specified in the scene, there are behaviors such as stealing, stealing, loan fraud, cashing out, brushing orders, and brushing good reviews. The corresponding behavior characteristics are different according to different fraud scenarios. The behavior characteristics can be learned and trained in advance, and the embodiments of the present invention do not make specific limitations.
[0072] A fraudulent behavior identification method provided in the implementation of the present invention is as follows: Figure 2As shown in the structure diagram of the identification system, the fraud identification platform executes the steps 101-103 based on the data module, feature engineering module and model module. Among them, the data module is used to collect and preprocess data, the feature engineering module is used to select, extract and construct behavioral features, and the model module is used to select the decision tree model and evaluate based on the behavioral features to determine whether there is fraud. Correspondingly, in the fraud identification platform, data judgment nodes such as fraud network map, business event abnormality, business event frequency, user behavior information, device terminal and network can also be embedded to achieve accurate and flexible identification of fraudulent behavior.
[0073] In another embodiment of the present invention, for further definition and explanation, the step divides the equity acquisition information into behavioral characteristics, and obtaining at least one behavioral characteristic includes:
[0074] Analyze at least one of the number of behaviors, the content of the benefits, and the amount of resources in the benefit acquisition information;
[0075] The behavior characteristics matching at least one of the number of behaviors, the equity content, and the number of resources are searched according to the behavior knowledge graph.
[0076] In order to accurately identify user behaviors in the rights acquisition information, thereby improving the accuracy of identifying fraudulent behaviors, the current execution end, when performing behavioral feature division, specifically, first analyzes at least one of the number of behaviors, rights content, and resource quantity in the rights acquisition information. Among them, the number of behaviors is the number of times the user performs the claiming behavior on the rights information, the rights content is the specific rights enjoyed for the target resource, including but not limited to discounts, red envelopes, cash back, etc., and the number of resources is the number of resources corresponding to the rights received, for example, for a commodity, the number of times it can be purchased after a discount, which is not specifically limited in the embodiment of the present invention. Then, according to the behavior knowledge graph, the behavior feature that matches at least one of the parsed number of behaviors, rights content, and resource quantity is queried. At this time, the behavior knowledge graph stores the behavior features corresponding to different number of behaviors, different rights content, and different resource quantities. Therefore, the corresponding behavior features can be queried based on the knowledge graph, and input into the decision tree classification model as a result of the initial or first behavior recognition, and further classified in combination with user information.
[0077] It should be noted that the behavior knowledge graph in the embodiment of the present invention is configured by the business personnel for behavior recognition according to the behavior of historical users with fraudulent behavior, so as to meet the one-time identification of fraudulent behavior.
[0078] In another embodiment of the present invention, for further definition and explanation, the step of classifying the behavior characteristics and the user information based on a decision tree classification model, before obtaining the behavior classification result, the method further includes:
[0079] Create an initialized decision tree classification model and obtain a behavior classification sample set;
[0080] The initialized decision tree classification model is trained based on the behavior classification sample set to obtain the decision tree classification model that has completed model training.
[0081] In order to accurately identify fraudulent behavior based on decision trees, the current execution end pre-trains the decision data classification model. Specifically, first, an initial decision tree classification model is constructed. Among them, a decision tree is a tree structure (can be a binary tree or a non-binary tree), each non-leaf node represents a test on a feature attribute, each branch represents the output of this feature attribute in a certain value range, and each leaf node stores a category. The process of using a decision tree to make a decision is to start from the root node, test the corresponding feature attributes in the item to be classified, and select the output branch according to its value until it reaches the leaf node, and take the category stored in the leaf node as the decision result. At the same time, a behavior classification sample set is obtained, and the behavior classification sample set includes behavior feature samples and user samples corresponding to different behavior classification labels, so as to train the initial decision tree classification model based on the behavior classification sample set, and obtain the tree classification model that completes the model training. In addition, each classification node in the trained decision tree classification model is marked with the inventory of classified behavior feature samples and user samples, as well as the trained classification tree path. Among them, the sample data in the behavior classification sample set includes: user information of good customers and fraudulent customers collected for one month within three months through under-sampling without replacement. The user information at this time includes but is not limited to point flow table, transaction channel table, customer basic information, activity participation information, login information, electronic channel information, merchant information, etc., which is not specifically limited in the embodiment of the present invention.
[0082] Specifically, in the behavior classification based on decision trees, the core idea of the decision tree algorithm is to measure attribute selection by information gain, and select the attribute with the largest information gain after splitting. In a specific implementation scenario, the algorithm is as follows: Let D be the classification of training tuples by category, then the entropy of D is expressed as:
[0083] Among them, pi represents the probability of the i-th category appearing in the entire training tuple, which can be estimated by dividing the number of elements belonging to this category by the total number of elements in the training tuple. The actual meaning of entropy is the average amount of information required for the class label of the tuple in D. If it is assumed that the training tuple D is divided according to attribute A, then the expected information of A on the division of D is:
[0084] The information gain is the difference between the two:
[0085] gain(A)=info(D)-info A (D);
[0086] At this point, the decision tree algorithm calculates the gain rate of each attribute each time a split is required, and then selects the attribute with the largest gain rate for splitting. For example, when constructing a decision tree, assume that the training set contains 10 elements, as shown in Table 1 below:
[0087] Table 1
[0088]
[0089]
[0090] Here, s, m, and l represent small, medium, and large, respectively. At this time, L, F, H, and R represent log density, friend density, whether a real avatar is used, and whether the account is real. The information gain of each attribute is calculated as follows:
[0091] info(D)=-0.7log 2 0.7-0.3log 2 0.3 = 0.7 * 0.51 + 0.3 * 1.74 = 0.879;
[0092]
[0093] gain(L)=0.879-0.603=0.276;
[0094] Therefore, the information gain of log density is 0.276. Using the same method, the information gains of H and F are 0.033 and 0.553 respectively. Since F has the largest information gain, F is selected as the split attribute for the first split. The results after splitting are as follows Figure 3 As shown:
[0095] Table 2
[0096] Log density Whether to use real avatar Does the account really exist? s no no m no no m no yes s yes no
[0097] exist Figure 2On the basis shown in , this method is recursively used to calculate the splitting attributes of the child nodes, and finally the classification result of the entire decision tree can be obtained, as shown in Table 2. In addition, in the embodiment of the present invention, in a specific application scenario, the log density and the friend density are continuous attributes, and for the feature attribute is a continuous value, a decision tree algorithm can be used, specifically: first sort the elements in D according to the feature attribute, then the middle point of every two adjacent elements can be regarded as a potential splitting point, starting from the first potential splitting point, split D and calculate the expected information of the two sets, the point with the minimum expected information is called the best splitting point of this attribute, and its information expectation is used as the information expectation of this attribute, which is not specifically limited in the embodiment of the present invention.
[0098] In the embodiment of the present invention, before training the decision tree model, in order to accurately learn different fraudulent behaviors, the current execution end can also analyze the sample data of unlabeled fraudulent behaviors in the behavior classification sample set. Specifically, graph construction, community discovery, anomaly detection and other methods can be used. Specifically, the graph construction is to construct it according to the correlation between each behavior, and then based on the community discovery method, the Unicom subgraph algorithm is first run, and the Louvain community discovery algorithm is used to mine the gang in the non-isolated subgraph, so as to discover the algorithm mining gang; then, the community indicators are counted, such as the number of people in each gang, the number of activities per capita in the gang, and the proportion of fraudulent lists in the gang; then, abnormal communities are detected, that is, community indicators are converted (such as log, power) to approximate Gaussian distribution, and Gaussian anomaly detection algorithm finds abnormal gangs; finally, the list is verified, and the business personnel investigate and verify the list of abnormal gangs, and the verified list is fed back to the decision tree model for real-time training and iterative optimization, which is not specifically limited in the embodiment of the present invention.
[0099] In another embodiment of the present invention, for further definition and explanation, before determining the target fraud user and the target fraud behavior, the method further includes:
[0100] If the target classification tree path corresponding to the behavior classification result is an untrained classification tree path, then the behavior feature sample stock and the user sample stock in all classification nodes corresponding to the target classification tree path are retrieved;
[0101] If the behavior feature sample stock and the user sample stock are greater than the behavior feature sample stock and the user sample stock of the classification node in the reference classification tree path, then the step of matching the behavior classification result with the preset fraud behavior strategy is performed;
[0102] If the behavior feature sample stock and the user sample stock are less than or equal to the behavior feature sample stock and the user sample stock of the classification node in the reference classification tree path, the behavior in the behavior classification result is determined to be non-fraudulent behavior.
[0103] In order to accurately identify fraudulent behavior and avoid fraudulent behavior that has not been learned by the decision tree classification model, the current execution end extracts the classification tree path corresponding to the behavior classification result after obtaining the behavior classification result by performing behavior identification based on the decision tree classification model as the target classification tree path. Specifically, if this target classification tree path is an untrained classification tree path, the current execution end retrieves the behavior feature sample inventory and user sample inventory in all classification nodes corresponding to the target classification tree path, and further compares them with the behavior feature sample inventory and user sample inventory of the classification node in the preset reference classification tree path. Among them, the reference classification tree path is a path that has been trained, and each classification node stores the feature sample inventory and user sample inventory classified according to this node, that is, the number of samples, which is not specifically limited in the embodiment of the present invention. Furthermore, if the behavior feature sample inventory and user sample inventory are greater than the behavior feature sample inventory and user sample inventory of the classification node in the reference classification tree path, it means that the probability of this target classification tree path being classified as fraudulent behavior is large, so the step of matching the execution behavior classification result with the preset fraud behavior strategy is sufficient. At the same time, if the behavioral feature sample stock and the user sample stock are less than or equal to the behavioral feature sample stock and the user sample stock of the classification node in the reference classification tree path, it means that the probability of this target classification tree path being classified as fraudulent behavior is small, then it is sufficient to determine that the behavior in the behavior classification result is non-fraudulent behavior.
[0104] In another embodiment of the present invention, in order to further define and illustrate, the step of classifying the behavior characteristics and the user information based on a decision tree classification model, and after obtaining the behavior classification result, the method further includes:
[0105] Comparing the behavior classification result with at least one fraudulent behavior in the preset fraudulent behavior strategy;
[0106] If the behavior classification result does not match all the fraudulent behaviors in the preset fraudulent behavior strategy, a non-fraudulent user and a non-fraudulent behavior are determined.
[0107] In order to improve the accuracy of identifying fraudulent behavior, the current execution end makes a judgment based on the preset fraudulent behavior strategy after obtaining the behavior classification result. Specifically, the current execution end compares the behavior classification result with at least one fraudulent behavior in the preset fraudulent behavior strategy. At this time, the preset fraudulent behavior strategy includes specific strategies for different behavior classification results to be judged as fraudulent behavior, including but not limited to identity fraud, proxy IP fraud, personal fraud, gang fraud, etc., which are not specifically limited in the embodiment of the present invention. For example, if the behavior classification result is that the behavior characteristics of user A are classified as multiple excess rights and interests, then the corresponding matching personal fraud behavior in the preset fraudulent behavior strategy will determine this user as a personal fraud user and the corresponding behavior is a fraudulent behavior, which is not specifically limited in the embodiment of the present invention. In addition, if the behavior classification result does not match all the fraudulent behaviors in the preset fraudulent behavior strategy, a non-fraudulent user and non-fraudulent behavior are determined, indicating that the user has not committed fraudulent behavior.
[0108] In a specific implementation scenario, the preset fraud behavior strategy can further divide the user's wool-pulling behavior into multiple levels of fraud behavior, including but not limited to three types of wool-pulling behavior. For example, the first type is the behavior of individuals pulling wool manually; the second type is to establish a fake client to pull wool by cracking the background interface of the platform (hacker type); the third type uses plug-in programs to fully automate the wool-pulling process; the fourth type is a gang wool-pulling party, usually an organizer organizes gang members to pull wool, and the combination strategy between various behavioral characteristics is set as the judgment of which type of wool-pulling behavior is which. The embodiment of the present invention does not make specific limitations. For example, a group control feature judgment is set, that is, the behavioral features of the same transaction amount for each transaction, concentrated behaviors, high transaction frequency within a short period of time after the activity is carried out, similar behavioral sequences, a large number of failed transactions, etc., then it is determined that there is group fraud; a gang feature judgment is set, that is, there is financial transactions between gangs, shared equipment, mobile phone numbers, IPs, similar GPS, etc., then it is determined that there is gang fraud; a resource-side feature judgment is set, that is, proxy IP, forged GPS, forged device numbers, impersonation of ID cards, a large number of one-to-many situations, etc., then it is determined that there is resource-side fraud; a redemption-side feature judgment is set, that is, concentrated electronic voucher collection, collection of hard currency such as liquor, high similarity of delivery addresses, etc., then it is determined that there is redemption-side fraud, and the embodiments of the present invention do not make specific limitations.
[0109] In another embodiment of the present invention, for further definition and explanation, the step of classifying the behavior characteristics and the user information based on a decision tree classification model, before obtaining the behavior classification result, the method further includes:
[0110] Matching the behavior characteristics with business behavior rules;
[0111] If the behavior feature matches the business behavior rule, at least one of a fraudulent group of users, a fraudulent individual user, and a business abnormality user is determined, and user information of at least one of the fraudulent group of users, the fraudulent individual user, and the business abnormality user is used as user information for executing the behavior classification step.
[0112] In order to ensure the accuracy of fraudulent behavior identification, before the behavior characteristics and user information are used as inputs of the decision tree classification model for behavior classification identification, judgment can also be made in advance based on the business behavior rules. Among them, the business behavior rules include at least one of the fraud group identification rules, business anomaly identification rules, and fraud individual identification rules. The fraud group identification rules are used to determine whether there is a group fraud when multiple users receive rights and interests information. The business anomaly identification rules are used to determine whether the user has the right to receive rights and interests information. The fraud individual identification rules are used to determine whether the user has personal fraud when receiving rights and interests information. The embodiment of the present invention does not make specific restrictions. After the behavior characteristics are matched with the business behavior rules, if the behavior characteristics match at least one of the fraud group identification rules, business anomaly identification rules, and fraud individual identification rules in the business behavior rules, then according to the matched fraud group identification rules, business anomaly identification rules, and fraud individual identification rules, at least one of the fraud group users, fraud individual users, and business anomaly users is determined, and then the user information of at least one of the fraud group users, fraud individual users, and business anomaly users is used as the user information for executing the behavior classification step, so as to further identify the behavior of receiving rights and interests information in combination with the corresponding individual fraud, group fraud, and business anomaly.
[0113] In another embodiment of the present invention, for further definition and explanation, after determining the target fraud user and the target fraud behavior, the method further includes:
[0114] The rights and interests information obtained by the target fraud user is cancelled, and the target fraud user is marked, so as to reissue the rights and interests information of the target resource to a normal user.
[0115] In order to reduce the loss of rights and interests information, the current execution end, after determining the target fraud user and the target fraud behavior, cancels the rights and interests information that has been received. Specifically, the current execution end cancels the rights and interests information obtained by the target fraud user and marks the target fraud user to re-issue the rights and interests information of the target resources to the normal user.
[0116] It should be noted that in the embodiment of the present invention, the current execution end acts as a fraud identification platform. After determining the target fraud user and the target fraud behavior, it can directly cancel the received rights and interests information and instruct the rights and interests information distribution platform to re-distribute it. The embodiment of the present invention does not make specific limitations.
[0117] The embodiment of the present invention provides a method for identifying fraudulent behavior. Compared with the prior art, the embodiment of the present invention obtains rights acquisition information and user information of a target resource, wherein the rights acquisition information is used to characterize the content of obtaining rights information of the target resource from different business channels; the rights acquisition information is divided into behavioral characteristics to obtain at least one behavioral characteristic, and the behavioral characteristics and the user information are behaviorally classified based on a decision tree classification model to obtain a behavior classification result; when the behavior classification result matches at least one fraudulent behavior in a preset fraudulent behavior strategy, a target fraudulent user and a target fraudulent behavior are determined, and identification is performed through multi-dimensional analysis of behavioral characteristics combined with a decision tree model, thereby improving the accuracy of identifying fraudulent behavior in users receiving rights information.
[0118] Furthermore, as a response to the above Figure 1 The embodiment of the present invention provides a fraudulent behavior identification device, such as Figure 4 As shown, the device comprises:
[0119] An acquisition module 21 is used to acquire rights acquisition information and user information of a target resource, wherein the rights acquisition information is used to represent the content of obtaining rights information of the target resource from different business channels;
[0120] A classification module 22, configured to classify the rights acquisition information into behavior characteristics to obtain at least one behavior characteristic, and to classify the behavior characteristics and the user information into behavior characteristics based on a decision tree classification model to obtain a behavior classification result;
[0121] The determination module 23 is used to determine a target fraudulent user and a target fraudulent behavior when the behavior classification result matches at least one fraudulent behavior in a preset fraudulent behavior strategy.
[0122] Furthermore, the classification module is specifically used to parse at least one of the number of behaviors, the rights and interests content, and the number of resources in the rights and interests acquisition information; and to query the behavior features that match at least one of the number of behaviors, the rights and interests content, and the number of resources according to the behavior knowledge graph, and the behavior knowledge graph stores the behavior features corresponding to different numbers of behaviors, different rights and interests content, and different numbers of resources.
[0123] Furthermore, the device also includes:
[0124] A creation module is used to create an initialization decision tree classification model and obtain a behavior classification sample set, wherein the behavior classification sample set includes behavior feature samples corresponding to different behavior classification labels and user samples;
[0125] A training module, used to train the initialized decision tree classification model based on the behavior classification sample set to obtain the decision tree classification model that has completed model training;
[0126] Among them, each classification node in the decision tree classification model that has completed training is marked with the inventory of classified behavior feature samples and user samples, as well as the trained classification tree path.
[0127] Furthermore, the device further comprises: a calling module, an execution module,
[0128] The retrieval module is used to retrieve the behavior feature sample stocks and user sample stocks in all classification nodes corresponding to the target classification tree path if the target classification tree path corresponding to the behavior classification result is an untrained classification tree path;
[0129] The execution module is used to execute the step of matching the behavior classification result with the preset fraud behavior strategy if the behavior feature sample inventory and the user sample inventory are greater than the behavior feature sample inventory and the user sample inventory of the classification node in the reference classification tree path;
[0130] The determination module is also used to determine that the behavior in the behavior classification result is non-fraudulent behavior if the behavior feature sample stock and the user sample stock are less than or equal to the behavior feature sample stock and the user sample stock of the classification node in the reference classification tree path.
[0131] Furthermore, the device further comprises: a comparison module,
[0132] The comparison module is used to compare the behavior classification result with at least one fraudulent behavior in the preset fraudulent behavior strategy;
[0133] The determination module is further configured to determine a non-fraudulent user and a non-fraudulent behavior if the behavior classification result does not match all the fraudulent behaviors in the preset fraudulent behavior strategy.
[0134] Furthermore, the device further comprises: a matching module,
[0135] The matching module is used to match the behavior characteristics with business behavior rules, wherein the business behavior rules include at least one of a fraud group identification rule, a business anomaly identification rule, and a fraud individual identification rule;
[0136] The determination module is also used to determine at least one of a fraudulent group of users, a fraudulent individual user, and a business abnormality user if the behavior feature matches the business behavior rule, and use the user information of at least one of the fraudulent group of users, the fraudulent individual user, and the business abnormality user as the user information for executing the behavior classification step.
[0137] Furthermore, the device also includes:
[0138] The marking module is used to cancel the rights and interests information obtained by the target fraud user and mark the target fraud user so as to reissue the rights and interests information of the target resource to a normal user.
[0139] The embodiment of the present invention provides a device for identifying fraudulent behavior. Compared with the prior art, the embodiment of the present invention obtains rights acquisition information and user information of a target resource, wherein the rights acquisition information is used to characterize the content of obtaining rights information of the target resource from different business channels; the rights acquisition information is divided into behavioral characteristics to obtain at least one behavioral characteristic, and the behavioral characteristics and the user information are behaviorally classified based on a decision tree classification model to obtain a behavior classification result; when the behavior classification result matches at least one fraudulent behavior in a preset fraudulent behavior strategy, a target fraudulent user and a target fraudulent behavior are determined, and identification is performed through multi-dimensional analysis of behavioral characteristics combined with a decision tree model, thereby improving the accuracy of identifying fraudulent behavior in users receiving rights information.
[0140] According to an embodiment of the present invention, a storage medium is provided, wherein the storage medium stores at least one executable instruction, and the computer executable instruction can execute the fraudulent behavior identification method in any of the above method embodiments.
[0141] Figure 5 A schematic diagram of the structure of a terminal provided according to an embodiment of the present invention is shown. The specific embodiment of the present invention does not limit the specific implementation of the terminal.
[0142] like Figure 5 As shown, the terminal may include: a processor (processor) 302, a communication interface (Communications Interface) 304, a memory (memory) 306, and a communication bus 308.
[0143] The processor 302 , the communication interface 304 , and the memory 306 communicate with each other via the communication bus 308 .
[0144] The communication interface 304 is used to communicate with other devices such as clients or other servers.
[0145] The processor 302 is used to execute the program 310, and specifically can execute the relevant steps in the above-mentioned fraudulent behavior identification method embodiment.
[0146] Specifically, the program 310 may include program codes, which include computer operation instructions.
[0147] The processor 302 may be a central processing unit (CPU), or an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiment of the present invention. The one or more processors included in the terminal may be processors of the same type, such as one or more CPUs; or processors of different types, such as one or more CPUs and one or more ASICs.
[0148] The memory 306 is used to store the program 310. The memory 306 may include a high-speed RAM memory, and may also include a non-volatile memory, such as at least one disk memory.
[0149] The program 310 may be specifically configured to enable the processor 302 to perform the following operations:
[0150] Acquire rights and interests acquisition information and user information of the target resource, wherein the rights and interests acquisition information is used to represent the content of obtaining rights and interests information of the target resource from different business channels;
[0151] Performing behavioral feature division on the rights acquisition information to obtain at least one behavioral feature, and performing behavioral classification on the behavioral feature and the user information based on a decision tree classification model to obtain a behavioral classification result;
[0152] When the behavior classification result matches at least one fraudulent behavior in the preset fraudulent behavior strategy, a target fraudulent user and a target fraudulent behavior are determined.
[0153] Obviously, those skilled in the art should understand that the above modules or steps of the present invention can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, and optionally, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order than here, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.
[0154] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.
Claims
1. A method for identifying fraudulent behavior, characterized in that: include: Acquire rights and interests acquisition information and user information of the target resource, wherein the rights and interests acquisition information is used to represent the content of obtaining rights and interests information of the target resource from different business channels; Performing behavioral feature division on the rights acquisition information to obtain at least one behavioral feature, and performing behavioral classification on the behavioral feature and the user information based on a decision tree classification model to obtain a behavioral classification result; When the behavior classification result matches at least one fraudulent behavior in the preset fraudulent behavior strategy, a target fraudulent user and a target fraudulent behavior are determined.
2. The method according to claim 1, characterized in that The performing of behavioral feature classification on the equity acquisition information to obtain at least one behavioral feature includes: Analyze at least one of the number of behaviors, the content of the benefits, and the amount of resources in the benefit acquisition information; The behavior knowledge graph is used to query the behavior features that match at least one of the number of behaviors, the rights and interests content, and the number of resources. The behavior knowledge graph stores behavior features corresponding to different numbers of behaviors, different rights and interests content, and different numbers of resources.
3. The method according to claim 2, characterized in that Before performing behavior classification on the behavior features and the user information based on the decision tree classification model to obtain the behavior classification result, the method further includes: Creating an initialized decision tree classification model and obtaining a behavior classification sample set, wherein the behavior classification sample set includes behavior feature samples corresponding to different behavior classification labels and user samples; Training the initialized decision tree classification model based on the behavior classification sample set to obtain the decision tree classification model that has completed model training; Among them, each classification node in the decision tree classification model that has completed training is marked with the inventory of classified behavior feature samples and user samples, as well as the trained classification tree path.
4. The method according to claim 3, characterized in that Before determining the target fraud user and the target fraud behavior, the method further includes: If the target classification tree path corresponding to the behavior classification result is an untrained classification tree path, then the behavior feature sample stock and the user sample stock in all classification nodes corresponding to the target classification tree path are retrieved; If the behavior feature sample stock and the user sample stock are greater than the behavior feature sample stock and the user sample stock of the classification node in the reference classification tree path, then the step of matching the behavior classification result with the preset fraud behavior strategy is performed; If the behavior feature sample stock and the user sample stock are less than or equal to the behavior feature sample stock and the user sample stock of the classification node in the reference classification tree path, the behavior in the behavior classification result is determined to be non-fraudulent behavior.
5. The method according to claim 4, characterized in that After the behavior characteristics and the user information are classified based on the decision tree classification model to obtain the behavior classification result, the method further includes: Comparing the behavior classification result with at least one fraudulent behavior in the preset fraudulent behavior strategy; If the behavior classification result does not match all the fraudulent behaviors in the preset fraudulent behavior strategy, a non-fraudulent user and a non-fraudulent behavior are determined.
6. The method according to claim 1, characterized in that Before performing behavior classification on the behavior features and the user information based on the decision tree classification model to obtain the behavior classification result, the method further includes: Matching the behavior characteristics with business behavior rules, wherein the business behavior rules include at least one of a fraud group identification rule, a business anomaly identification rule, and a fraud individual identification rule; If the behavior feature matches the business behavior rule, at least one of a fraudulent group of users, a fraudulent individual user, and a business abnormality user is determined, and user information of at least one of the fraudulent group of users, the fraudulent individual user, and the business abnormality user is used as user information for executing the behavior classification step.
7. The method according to any one of claims 1 to 6, characterized in that: After determining the target fraud user and the target fraud behavior, the method further includes: The rights and interests information obtained by the target fraud user is cancelled, and the target fraud user is marked, so as to reissue the rights and interests information of the target resource to a normal user.
8. A fraudulent behavior identification device, characterized in that: include: An acquisition module, used to acquire rights acquisition information and user information of a target resource, wherein the rights acquisition information is used to represent the content of obtaining rights information of the target resource from different business channels; A classification module, configured to classify the rights acquisition information into behavior characteristics to obtain at least one behavior characteristic, and to classify the behavior characteristics and the user information into behavior characteristics based on a decision tree classification model to obtain a behavior classification result; The determination module is used to determine a target fraud user and a target fraud behavior when the behavior classification result matches at least one fraud behavior in a preset fraud behavior strategy.
9. A storage medium storing at least one executable instruction, wherein the executable instruction enables a processor to execute an operation corresponding to the method for identifying fraudulent behavior according to any one of claims 1 to 7.
10. A terminal, comprising: A processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus; The memory is used to store at least one executable instruction, and the executable instruction enables the processor to execute an operation corresponding to the fraudulent behavior identification method according to any one of claims 1-7.
Citation Information
Patent Citations
Merchant fraudulent transaction detection method based on graph
CN116245543A
Fraudulent user identification method and device, computer equipment and storage medium
CN116402506A
