Data processing methods, apparatus, equipment and storage media
By generating user operation tags using N clustering algorithms within the P-layer transaction statistics period, the problem of single user characteristics in existing technologies is solved, and the service platform achieves accuracy in user risk assessment and targeted services.
Patent Information
- Application Number
- CN202310410353.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-04-17
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2043-04-17
AI Technical Summary
In existing technologies, the methods for generating user profile tags are highly homogeneous, quantitatively subjective, and semantically ambiguous, which makes it impossible for service platforms to obtain complete user characteristics and affects the accuracy of user risk assessment.
By acquiring multiple sets of user transaction characteristics, N clustering algorithms are used to cluster users within the P-layer transaction statistics period to generate user operation tags. Combining the clustering results with the business scenario, the similarity between user transaction characteristics and clusters is determined, resulting in richer user characteristics.
This improves the accuracy of the service platform in providing targeted services based on user characteristics. By progressively transforming the transaction statistics time period from coarse to fine, the tagging system is enriched, user-side tags are refined, and richer user characteristics are obtained.
Smart Images

Figure CN116485519B_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the field of computer technology, and in particular relates to a data processing method, apparatus, device and storage medium. Background Technology
[0002] With the development of internet technology, user profiling has been widely applied in various fields. User profiling involves abstracting each specific characteristic of a user into tags, using these tags to concretize the user's image, so that service platforms can provide targeted services based on these tags.
[0003] In related technologies, service platforms can use tags pre-associated with monitored businesses to reverse-engineer user characteristics, such as transaction data, corresponding to those tags, thereby determining a user's credit transaction permissions on the platform. However, due to limitations in the scope of monitored businesses, the business tags are homogeneous and restrictive, resulting in relatively singular user characteristics obtained by the service platform, which affects the platform's risk assessment of users. Summary of the Invention
[0004] This application provides a data processing method, apparatus, device, and storage medium that can solve the problem in related technologies where complete user characteristics cannot be obtained, affecting the accuracy of user evaluation by the service platform.
[0005] In a first aspect, embodiments of this application provide a data processing method, which may include:
[0006] Acquire multiple sets of user transaction features, each set of user transaction features including the target user transaction features on the user side;
[0007] According to the clustering dimension, each group of user transaction features is clustered using N clustering algorithms to obtain clustering results. The clustering dimension includes the P-layer transaction statistics time period, where the duration of each layer of transaction statistics time period is different. The clustering results include the set of clusters for each group of user transaction features within the p-th layer of transaction statistics time period of the P-layer transaction statistics time period and corresponding to the n-th clustering algorithm among the N clustering algorithms.
[0008] Based on the clustering results, user management tags corresponding to the business scenario are determined. The user management tags include the degree of similarity between the target user transaction features in each group of user transaction features and the clusters corresponding to the nth clustering algorithm within the transaction statistics time period of the p-th layer.
[0009] Secondly, embodiments of this application provide a data processing apparatus, which may include:
[0010] The acquisition module is used to acquire multiple sets of user transaction features. Each set of user transaction features includes the target user's transaction features on the user side.
[0011] The clustering module is used to cluster each group of user transaction features according to the clustering dimension and using N clustering algorithms to obtain the clustering results. The clustering dimension includes P-level transaction statistical time periods, each of which has a different duration. The clustering results include the set of clusters for each group of user transaction features that are within the p-th level of the P-level transaction statistical time period and correspond to the n-th clustering algorithm among the N clustering algorithms.
[0012] The determination module is used to determine user operation tags corresponding to the business scenario based on the clustering results. The user operation tags include the degree of similarity between the target user transaction features in each group of user transaction features and the cluster corresponding to the nth clustering algorithm within the transaction statistics time period of the p-th layer.
[0013] Thirdly, embodiments of this application provide a computer device, which includes: a processor and a memory storing computer program instructions;
[0014] When the processor executes computer program instructions, it implements the data processing method as described in the first aspect.
[0015] Fourthly, embodiments of this application provide a computer storage medium storing computer program instructions, which, when executed by a processor, implement the data processing method as described in the first aspect.
[0016] Fifthly, embodiments of this application provide a chip, which includes a processor and a communication interface. The communication interface and the processor are coupled, and the processor is used to run programs or instructions to implement the data processing method as shown in the first aspect.
[0017] In a sixth aspect, embodiments of this application provide a computer program product stored in a storage medium, which is executed by at least one processor to implement the data processing method as described in the first aspect.
[0018] The data processing method, apparatus, device, and storage medium of this application embodiment can acquire multiple sets of user transaction features, each set of user transaction features including target user transaction features on the user end. Next, according to the clustering dimension, each set of user transaction features is clustered using N clustering algorithms to obtain clustering results; wherein, the clustering dimension includes a P-layer transaction statistical time period, where the duration of each layer of the P-layer transaction statistical time period is different, and the clustering result includes the set of clusters corresponding to the nth clustering algorithm among the N clustering algorithms, within the p-th layer of the P-layer transaction statistical time period. Then, based on the clustering results, user management tags corresponding to the business scenario are determined, and user management tags include those used to characterize the similarity between the target user transaction features in each set of user transaction features within the p-th layer of the transaction statistical time period and the clusters corresponding to the nth clustering algorithm. In this way, by using N clustering algorithms to cluster each group of user transaction features according to the clustering dimension and the transaction statistical time period of each layer of the P-layer transaction statistical time period, a more reasonable cluster structure is generated. Combining the similarity between user transactions in the p-layer transaction statistical time period and the clusters corresponding to the n-th clustering algorithm, it is beneficial to obtain fuzzy membership information of target user transaction features with respect to various clusters. This helps to further refine the user-end tags, thereby enabling the service platform to obtain richer user features. This allows the platform to use machine learning methods to mine implicit transaction patterns of users in different scenarios and different layers of transaction statistical time periods. By progressively transforming the transaction statistical time periods from coarse to fine, the platform can find effective feature combinations from large-scale transaction patterns that help understand users' transaction preferences in different scenarios, enrich the construction logic of the tag system, and improve the accuracy of the service platform in providing targeted services to users based on tags. Attached Figure Description
[0019] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0020] Figure 1 A flowchart illustrating a data processing method provided in an embodiment of this application;
[0021] Figure 2 This is a schematic diagram of the structure of a data processing apparatus provided in one embodiment of this application;
[0022] Figure 3 This is a schematic diagram of the structure of a computer device provided in one embodiment of this application. Detailed Implementation
[0023] The features and exemplary embodiments of various aspects of this application will be described in detail below. To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain this application and not to limit it. For those skilled in the art, this application can be implemented without some of these specific details. The following description of the embodiments is merely to provide a better understanding of this application by illustrating examples.
[0024] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes the element.
[0025] With the development of internet technology, user profiling has been widely applied in various fields. User profiling involves abstracting each specific characteristic of a user into tags, using these tags to concretize the user's image, so that service platforms can provide targeted services based on these tags.
[0026] While current user profile tags have reached a certain scale and hierarchy, most are based on a series of first-order statistics (such as maximum / minimum, mean, percentage, and comprehensive values) derived from one or more combinations of basic transaction flow characteristics (time, location, amount, frequency, channel, merchant). These first-order statistics are then divided according to preset conditions to generate tags. However, this method of generating tags is insufficient to reflect the diverse user profiles, resulting in highly homogeneous, quantitatively subjective, and semantically ambiguous tags, affecting the accuracy of services provided by service platforms. High homogeneity means that a single first-order statistic cannot reflect differences and cannot reflect the objective differences in scenario construction and consumption levels across different regions. Quantitative subjectivity means that the division of consumption amount ranges reflecting user contributions depends on preset conditions, and a fixed statistical method is not conducive to adaptive adjustments to business strategies. Semantically ambiguous means that different users with the same tags lack differences in the degree of belonging to the concepts corresponding to those tags, affecting the accuracy of services provided by service platforms.
[0027] The root cause of these three problems lies in the fact that the current tagging system prioritizes the compliance of transactions across various channels and the overall effectiveness of user outreach. In practical applications, it fails to accurately and specifically define the user profile, resulting in tags that cannot reflect the user's profile or allow for the reverse filtering of structured user characteristics based on the tags.
[0028] In related technologies, tags can be generated through the following two technical approaches, and structured user characteristics can be filtered in reverse based on the tags.
[0029] Specifically, technical approach one focuses on risk assessment and strategy methods for credit users of financial institutions. This method consists of two parts: financial user classification and target management strategies. Financial user classification uses supervised learning to score the risk of multiple features in historical user data to predict the financial behavior and lending risk of target users. Target management strategies consist of a predefined set of quadrant-related management strategies, with different quadrants divided by thresholds to represent different categories of user credit. Thus, these two parts are logically sequential, aiming to personalize credit plans for different users before credit is granted, thereby increasing high-quality credit and reducing bad credit. However, this method has the following drawbacks: First, it only considers a single credit scenario, and the constructed features have limited reference value for analyzing users' other financial activities, insufficient to reflect the value of transfer and clearing data; second, the data involved can only utilize historical records within the field, which is insufficient to objectively assess the credit level of users at different levels, and incomplete transaction records may lead to biases in the model output when applied to the actual risk estimation of users; third, the assessment quadrants are determined based on thresholds, which relies heavily on subjective business experience.
[0030] Technical approach two focuses on monitoring methods for financial institutions' business products. This method extracts statistical values from transaction data over historical periods (e.g., daily, weekly, monthly) and summarizes these statistical features as pixels in a convolutional neural network, constructing an input matrix. After training the convolutional neural network, it compares the consistency of subsequent transaction statistics with the final business label. If the output matches the final business label category, the business product is considered normal; otherwise, abnormal transactions have occurred. However, this method has the following drawbacks: First, the feature hierarchy is coupled with the statistical frequency of the business (i.e., using the same unit of time), making user features insufficient to support joint data analysis that may be related to the business (e.g., marketing). Second, features are generated solely based on statistics without considering the differences in the units of measurement, making basic features insufficient to reflect customer information. Third, the use of a convolutional neural network to construct the model makes the output data untraceable, sacrificing the interpretability of the model's input data.
[0031] Based on this, to address the aforementioned problems, embodiments of this application provide a data processing method for constructing user management tags. This method aims to automatically generate tags with rich semantic connotations and geared towards user management analysis tasks. Furthermore, considering the importance of business scenarios to user management, embodiments of this application propose a method for filtering the generated multi-level user management tags. This method can not only construct differentiated user management tags for different scenarios but can also be used for filtering user management tags after joint intersection in vertical federated learning.
[0032] Therefore, for the stage of generating user management tags in data processing, a hierarchical user management tag data processing method is proposed. This method generates hierarchical features of user transaction data statistics based on user-card relationships and transaction flow. Specifically, it selects several sets of user transaction features and, according to P-layer transaction statistical time periods (e.g., each layer corresponds to year, month, week, and day, and each layer can also include sub-transaction statistical time periods, such as the day corresponding to the hour in the layer's transaction statistical time period), it performs multiple clustering operations on the target user transaction features in different regions (e.g., different regions, provinces, cities, etc., based on user card holdings) layer by layer, according to the clustering dimension. Then, layer by layer, it merges the results of N clustering algorithms into a single clustering result to characterize the similarity between the clusters corresponding to the nth clustering algorithm within the p-th layer transaction statistical time period in a specific region and time period, using these clusters as user management tags.
[0033] Thus, based on the distribution of user transaction data across at least one of the following platforms—card issuer, service acceptor, service provider, and merchant—both card-based and account-based characteristics are considered. Aside from the user operation tag generation and screening methods, no additional technical modifications are required, reducing reliance on business experience. Furthermore, by comprehensively utilizing clustering dimensions, each layer of user transaction characteristics can leverage the advantages of multiple clustering dimensions, generating a more reasonable cluster structure. This, combined with the similarity between the target user's transaction characteristics within the p-th layer transaction statistical time period and the cluster corresponding to the n-th clustering algorithm, allows for further refinement of user group selection.
[0034] Furthermore, the data processing method provided in this application also proposes a hierarchical screening process for user management tags. Based on each layer of user management tags generated in the above process, this process constructs both absolute and relative semantic tags for the user's scenario preferences within the same time period, using business scenarios as units. Then, the stability of each dimension tag is evaluated according to commonly used model stability indicators. When a group of user transaction features is insufficiently distinguishable from the defined transaction preferences, it is determined whether to replace the coarse-grained user transaction features with fine-grained level user transaction features from the same group. Finally, user management tags with business scenario differences are determined. Thus, the multi-layered user management tags generated by screening based on the dual-quantitative preference characteristics of business scenarios help form regionally differentiated tag combinations. With richer expressive power and volume, user management tags not only supplement the existing tag system but also enhance the effectiveness of business scenario-based and cross-institutional collaborative modeling.
[0035] Based on this, embodiments of this application provide a data processing method, apparatus, computer device, and storage medium. The following will describe in conjunction with the accompanying drawings. Figures 1 to 3 This application describes in detail the data processing methods, apparatus, computer equipment, and storage media of the embodiments. It should be noted that these embodiments are not intended to limit the scope of this application.
[0036] First, the data processing method provided in this application embodiment will be described in detail. Here, the data processing method provided in this application embodiment is a multi-level feature mapping and filtering method for improving user management. This method aims to use machine learning methods to mine implicit transaction patterns of users in different regions and time periods regarding different scenarios. Through a coarse-to-fine hierarchical transformation, it seeks effective feature combinations from large-scale transaction patterns that help understand users' transaction preferences in different scenarios, enriching the construction logic of the tag system and providing decision support for subsequent targeted marketing activities. The data processing method provided in this application embodiment is divided into two parts: a user management tag generation stage and a user management tag filtering stage. The user management tag generation part is executed in an unsupervised learning manner to obtain a search solution space for user management tags in different scenarios. The user feature filtering stage is executed in a supervised learning manner, evaluating user management tags for each scenario based on the input model stability metric to determine multiple sets of effective user management tags within the region and time period to be analyzed.
[0037] Based on this, the following will be combined with Figure 1 The present application provides an example of a data processing method for generating user management tags and an example of a data processing method for filtering user management tags, as detailed below.
[0038] Figure 1This is a flowchart of a data processing method provided in an embodiment of this application.
[0039] like Figure 1 As shown, this data processing method can be applied to the server side, and the data processing method can specifically include the following steps:
[0040] Step 110: Obtain multiple sets of user transaction features, each set including target user transaction features on the user side; Step 120: Cluster each set of user transaction features according to the clustering dimension using N clustering algorithms to obtain clustering results; wherein, the clustering dimension includes P-layer transaction statistical time periods, each layer of transaction statistical time periods in the P-layer transaction statistical time period has a different duration, and the clustering results include the set of clusters corresponding to the nth clustering algorithm among the N clustering algorithms within the p-th layer transaction statistical time period of the P-layer transaction statistical time period; Step 130: Based on the clustering results, determine user management tags corresponding to the business scenario, the user management tags including the degree of similarity between the target user transaction features in each set of user transaction features within the p-th layer transaction statistical time period and the clusters corresponding to the nth clustering algorithm.
[0041] In this way, by using N clustering algorithms to cluster each group of user transaction features according to the clustering dimension and the transaction statistical time period of each layer of the P-layer transaction statistical time period, a more reasonable cluster structure is generated. Combining the similarity between user transactions in the p-layer transaction statistical time period and the clusters corresponding to the n-th clustering algorithm, it is beneficial to obtain fuzzy membership information of target user transaction features with respect to various clusters. This helps to further refine the user-side tags, thereby enabling the service platform to obtain richer user features. This allows the platform to use machine learning methods to mine implicit transaction patterns of users in different scenarios and different layers of transaction statistical time periods. By progressively transforming the transaction statistical time periods from coarse to fine, the platform can find effective feature combinations from large-scale transaction patterns that help understand users' transaction preferences in different scenarios. This enriches the construction logic of the tag system and improves the accuracy of the service platform in providing targeted services to users based on tags.
[0042] The steps described above are explained in detail below.
[0043] First, regarding step 110, in one or more possible embodiments, the target user transaction characteristics include card-based characteristics and account-based characteristics, wherein the card-based characteristics include transaction flow characteristics of at least one payment card, and the account-based characteristics include transaction flow characteristics of at least one transaction account. Based on this, step 110 may specifically include steps 1101 and 1102, as shown below.
[0044] Step 1101: Obtain business-related data from the target end. The target end includes at least one of the card issuing end, service acceptance end, service end, and merchant end. The business-related data includes data stored on the target end regarding various types of entities involved in user transaction transfer and settlement for the user end, as well as the transaction itself.
[0045] For example, multiple sets of business attention data Ta with operational analysis semantics are obtained from at least one of the card issuer, service acceptance, service, and merchant. The multiple sets of business attention data Ta with operational analysis semantics include characteristic information of various types of entities involved in user transaction transfer and settlement, as well as the transaction itself. Typically, it includes one or more of the following seven dimensions: amount, frequency, acquiring institution, card issuer, channel, scenario, and merchant.
[0046] Step 1102: Extract user-side card-based features and account-based features from business-focused data.
[0047] Specifically, user card-based features and account-based features can be extracted from business-related data based on the user's identity identifier.
[0048] For example, the transaction characteristics x of the payment card issued by the card issuer for the same user are determined based on the user's identity identifier, such as the user's identity credential number and payment card number. v ={x1,x2,…x v}, and, based on user-end identifiers such as the user's mobile phone number, user biometrics (such as fingerprints, iris scans, etc.), and account passwords, determine the transaction flow characteristics X of the same user's transaction account. u ={X1,X2,…X u}. Where v and u are positive integers.
[0049] It's important to note that the core of card-based features is the payment card number. The data is stored on the payment card, and the payment medium includes not only card swiping but also point-of-sale (POS) terminals, contactless payments, telephone payments, online banking payments, and online non-magnetic and PIN-free payments—all carriers that use card information for payment. The core of account-based features is real-name authentication and account password verification. Account passwords can be keys, numbers, fingerprints, or SMS messages. The data is stored in the transaction account, and the payment medium can be the account balance.
[0050] Next, regarding step 120, in one or more possible embodiments, based on the above steps 1101 and 1102, the clustering result includes a first clustering result corresponding to the cardinality feature and a second clustering result corresponding to the account base feature. Based on this, step 120 in the embodiments of this application may specifically include:
[0051] Based on the clustering dimension, the Kjeldahl features are subjected to N differential clustering operations using N clustering algorithms to obtain the first clustering result. This first clustering result includes the set of clusters corresponding to the Kjeldahl features within the p-th layer of the P-layer transaction statistics time period and aligned with the n-th clustering algorithm among the N clustering algorithms.
[0052] According to the clustering dimension, the account base features are clustered N times with differences using N clustering algorithms to obtain the second clustering result. The second clustering result includes the set of clusters of the account base features within the p-th layer transaction statistical time period of the P-layer transaction statistical time period and corresponding to the n-th clustering algorithm among the N clustering algorithms.
[0053] It should be noted that, in addition to time, the clustering dimension in this application embodiment may also include region. That is, the clustering dimension may include the P-layer transaction statistics time period and M regions. Based on this, the clustering result includes a set of clusters for each group of users in the m-th region of the M regions, whose transaction characteristics are within the p-th layer transaction statistics time period of the P-layer transaction statistics time period and correspond to the n-th clustering algorithm among the N clustering algorithms.
[0054] Alternatively, in this embodiment, the clustering dimension can include not only time but also business scenarios. That is, the clustering dimension can also include a P-layer transaction statistics time period and K business scenarios. Based on this, the clustering result includes a set of clusters corresponding to the nth clustering algorithm among the N clustering algorithms, for the kth business scenario among the K business scenarios, where each group of user transaction features falls within the p-th layer transaction statistics time period of the P-layer transaction statistics time period.
[0055] Alternatively, the clustering dimension in this application embodiment may include time, region, and business scenario. That is, the clustering dimension may include the P-layer transaction statistics time period, K business scenarios, and M regions. Based on this, the clustering result includes a set of clusters corresponding to the nth clustering algorithm among the N clustering algorithms, for the kth business scenario among the K business scenarios and the mth region among the M regions, where each group of user transaction features is within the p-th layer transaction statistics time period of the P-layer transaction statistics time period.
[0056] It should be noted that P is an integer greater than 1, p∈[1,P]; K is a positive integer, k∈[1,K]; M is an integer greater than 1, m∈[1,M]; and N is a positive integer, n∈[1,N]. Based on this, user transaction characteristics can be structured into a clear and detailed cluster structure according to the above clustering dimensions and clustering algorithms.
[0057] In another or more possible embodiments, the clustering dimensions include P-layer transaction statistics time periods, M regions, and K business scenarios. Based on this, step 120 may specifically include:
[0058] Step 1201: Taking the kth business scenario out of the K business scenarios as a unit, obtain the basic user transaction characteristics of the mth region in the M regions within the pth layer transaction statistics time period from each group of user transaction characteristics;
[0059] Step 1202: Statistically analyze the business attributes of the T groups of basic user transaction characteristics;
[0060] Step 1203: According to the p-th layer transaction statistics time period, the m-th region, and the k-th business scenario, N clustering algorithms are used to cluster the target user transaction features in each group of user transaction features, respectively, to obtain the first cluster set of the target user transaction features in each group of user transaction features that corresponds to the n-th clustering algorithm among the N clustering algorithms in the p-th layer transaction statistics time period, the m-th region, and the k-th business scenario; and according to the p-th layer transaction statistics time period, the m-th region, and the k-th business scenario, N clustering algorithms are used to cluster the t-th group of business attributes in T groups of business attributes and the target basic user transaction features corresponding to the t-th group of business attributes, respectively, to obtain the second cluster set of the target basic user transaction features that corresponds to the n-th clustering algorithm among the N clustering algorithms and the t-th group of business attributes in the p-th layer transaction statistics time period, the m-th region, and the k-th business scenario.
[0061] Step 1204: Determine the first cluster set and the second cluster set as the clustering results.
[0062] For example, the business scenario is taken as the unit (the kth business scenario is denoted as S). k Specifically, it can be distinguished by merchant category (i.e., MCC code), according to the transaction statistics time period of the p-th layer and the m-th region, respectively according to card base feature f kt (x u ) and accounting characteristics f kt (X v ) Calculate the t-th group of business attributes in user transaction records (e.g., number of transactions in the catering industry, transaction amount in the tourism industry, etc.). Determine N clustering algorithms, which can include k-means, spectral clustering, etc. (It should be noted that different parameter settings of the same clustering algorithm are also acceptable). Where f kt (x u ) can be used to characterize the statistical value of the account-based characteristics of cardholder u in the t-th group of business attributes in the k-th business scenario, and similarly f kt (X v This can be used to calculate the statistical value of the cardholder v's card base characteristics in the kth business scenario and the tth group of business attributes.
[0063] Based on this, and according to the conventions of the selected clustering algorithm, clustering operations are performed on the t-th group in the p-th layer transaction statistics time period and the m-th region, respectively, based on account-based features and card-based features, so that the t-th group in scenario S... k The basic clustering results were obtained under the corresponding algorithm parameter settings.
[0064] With N clustering algorithms and T groups of business attributes, if the t-th group forms q clusters, then the corresponding clustering results include the first cluster set and the second cluster set (n = 1, 2, ..., N), we have:
[0065] C k (n)=[C k1 (n),…,C kN [(n)], where the cluster C kt (n)={C kt (n) 1 ,…,C kt (n) q};
[0066] c k (n)=[c k1 (n),…,c kN [n], where the cluster c kt (n)={c kt (n) 1 ,…,c kt (n) q};
[0067] The first cluster set includes the first sub-cluster set C corresponding to the Karnaki features. k (n) and the second sub-cluster set c corresponding to the accounting base characteristics k (n). The second cluster set includes the third sub-cluster set C corresponding to the Kjeldahl features. kt (n) and the fourth sub-cluster set c corresponding to the accounting base characteristics kt (n). Furthermore, C k (n) can specifically represent the set of clusters obtained by clustering the account base features in the k-th scenario using the n-th clustering algorithm. Similarly, c k (n) can specifically represent the set of clusters obtained by the Kjeldahl feature in the k-th scenario and the n-th clustering algorithm. C kt (n) can specifically represent the set of clusters obtained by the clustering algorithm of the nth clustering of the account base features in the kth scenario and the tth group of business attributes. Similarly, c kt (n) Specifically, it can represent the set of clusters obtained by the clustering algorithm of the k-th scenario and the t-th group of business attributes.
[0068] Then, relating to step 130, in one or more possible embodiments, step 130 may specifically include:
[0069] Step 1301: Based on the clustering results, determine the average information entropy of the clusters. The average information entropy of the clusters includes the average information entropy of the target user's transaction features in the kth business scenario and the mth region, which is composed of the jth cluster corresponding to the nth clustering algorithm of the tth group of business attributes, relative to the cluster set corresponding to the n'th clustering algorithm of the same business attribute.
[0070] Step 1302: Calculate the basic cluster average information entropy of the target user's transaction features based on the average information entropy of the clusters. The basic cluster average information entropy includes the average information entropy of the cluster set corresponding to the n clustering algorithms of the target user's transaction features in the kth business scenario and the mth region, and the tth group of business attributes, relative to the cluster set corresponding to each of the N clustering algorithms for the same business attribute.
[0071] Step 1303: Using the average information entropy of the basic clustering under the kth business scenario as the evaluation condition, evaluate the clustering results and obtain the target clustering results;
[0072] Step 1304: Based on the target clustering results, determine the user management tags corresponding to the business scenario.
[0073] For example, the average information entropy of the clustering results under each business scenario is used as a metric to evaluate the amount of cluster information obtained by each group of business attributes under each clustering algorithm setting.
[0074] The cluster information content characterizes the probability that the business attribute values contained in the clusters generated by the nth clustering algorithm will appear in the various clusters of another clustering algorithm.
[0075] For scenario S k The j-th cluster of business attributes formed by the nth clustering algorithm under the account-based features has an average information entropy Hn'(C) relative to the n'th clustering algorithm. kt (n) j It can be calculated using the following formulas (1) and (2):
[0076]
[0077] in,
[0078] Among them, C kt (n) j The j-th cluster is obtained by clustering the n-th cluster of the t-th group of business attributes in the k-th business scenario using the basic features; c kt (n) jThe cluster is the j-th cluster obtained by the clustering algorithm of the n-th business attribute in the k-th business scenario and the t-th business attribute; log is the logarithm symbol (base 10), Hn'(C kt (n) j ) represents the average information entropy of the j-th cluster obtained by the nth clustering algorithm of the nth business attribute in the k-th business scenario, relative to the cluster set corresponding to the n'th clustering cluster set of the same business attribute.
[0079] Similarly, for scenario Sk, under the Kj' feature, the j-th cluster of the t-th business attribute formed by the n-th clustering algorithm, its average information entropy Hn'(c) relative to the n'-th clustering algorithm is... kt (n) j It can be calculated using the following formulas (3) and (4):
[0080]
[0081] in,
[0082] Among them, Hn'(c kt (n) j The information entropy of the j-th cluster, obtained by the nth clustering algorithm of the k-th business scenario and the t-th group of business attributes, is relative to the cluster set corresponding to the n'-th clustering cluster set of the same business attribute.
[0083] Next, based on the above average information entropy Hn'(C kt (n) j ),Hn'(c kn (n) j ), calculate the average information entropy H(C) of the basic clusters under the nth clustering algorithm setting respectively. kt (n)) and H(c kt (n)), specifically as shown in formulas (5) and (6):
[0084]
[0085]
[0086] Among them, H(C kt (n) represents the average information entropy of the clustering algorithm of the nth clustering algorithm for the account base feature in the kth business scenario, relative to the cluster sets corresponding to the same business attribute in N clustering algorithms. Similarly, H(c kt (n) The average information entropy of the clustering algorithm of the nth clustering algorithm for the tth group of business attributes in the kth business scenario, relative to the cluster set corresponding to the same business attribute in N clustering algorithms.
[0087] Based on this, in one example, step 1303 above may specifically include:
[0088] Step 13031: Based on the average information entropy of the basic clustering, select the target cluster set from the clustering results;
[0089] Step 13032: Remove the cluster set from the clustering result set to obtain the current clustering result;
[0090] Step 13033: Determine the current clustering result as the clustering result, and repeat the process of determining the average information entropy of the clusters based on the clustering result, and calculating the basic cluster average information entropy of the target user's transaction characteristics based on the average information entropy of the clusters.
[0091] Step 1304: Until the number of repetitions meets the preset conditions, the current clustering result set corresponding to the number of repetitions meeting the preset conditions is determined as the target clustering result; or, until the difference between the average information entropy of the basic clusters in h consecutive rounds is less than or equal to a preset threshold, the current clustering result set corresponding to the last round of h consecutive rounds is determined as the target clustering result.
[0092] For example, based on the average information entropy of the basic clusters and a preset threshold (α, β), high-quality basic clustering results (H(C)) are retained. kt (n))<α or H(c) kt (n))<α), delete inferior basic clustering results (H(C)). kt (n))>β or H(c) kt If (n) > β), recalculate the average information entropy of the remaining basic clusters after deleting the inferior basic cluster results. Through multiple iterations, when the average information entropy of each basic cluster remains basically unchanged for several consecutive rounds (usually less than a very small number, such as 10), the average information entropy of the remaining basic clusters is calculated. -5 ), to obtain the account-based features and card-based features in business scenario S k The target clustering result CL k and cl k Based on this, continuing with the example above, there are T groups of business attributes, and the t-th group forms q' clusters. Then:
[0093] CL k =[CL k1, CL k2 ,…,CL kT ,], where CL kt =[CL kt 1 ,CL kt 2 ,…,CL kt q’ ];
[0094] cl k =[cl k1 ,cl k2 ,…,cl kT ], where cl kt =[cl kt 1 ,cl kt 2 ,…,cl kt q’ ]
[0095] It should be noted that α is a threshold for measuring the quality of the basic clusters that should be retained as high-quality basic clusters, and β is a threshold for measuring the quality of the basic clusters that should be deleted as low-quality basic clusters. k The clustering result of the account base features in the k-th scenario is given. Similarly, cl k This represents the clustering result of the Kjelda features in the k-th scenario. kt For the clustering result of the accounting features in the t-th group of business attributes under the k-th business scenario, similarly, cl kt The clustering ensemble result of the t-th group of business attributes in the k-th business scenario by Kaki. kt q’ For the clustering ensemble result of the t-th group of business attributes in the k-th business scenario, the q'-th cluster is, similarly, cl kt q’ In the clustering ensemble results of the t-th group of business attributes in the k-th scenario, the q'-th cluster is...
[0096] In another example, step 1304 above may specifically include:
[0097] Step 13041: Using a similarity calculation algorithm, calculate the similarity degree between the target user transaction features in each group of user transaction features and the cluster sets corresponding to the n clustering algorithms within the p-th layer transaction statistics time period, the m-th region, and the k-th business scenario, for the t-th group of business attributes.
[0098] Step 13042: The similarity of the target user transaction features in each group of user transaction features with the cluster sets corresponding to the n clustering algorithms within the transaction statistics time period of the p-th layer, the m-th region, and the k-th business scenario, for the t-th group of business attributes, is determined as the user operation tag.
[0099] For example, a similarity calculation method d is defined based on common similarity metrics (such as Euclidean distance), and this method is used to calculate the values of each business attribute of the account base and card base in scenario S. k The following relates to the similarity Φ of all clusters in the clustering algorithm. k and Ψ kThis characterizes the degree of membership of cardholders in various clusters (i.e., potential transaction patterns) based on account-based features and card-based features (e.g., under Euclidean distance, the smaller the distance, the higher the degree of membership; in this case, distance and degree of membership are negatively correlated). Assuming there are T groups of business attributes, and the t-th group forms q' clusters, then:
[0100] Φ k =[Φ k1, Φ k2 ,…,Φ kT ], where Φ kt =[d(f kt (X m ),CL kt 1 ),…,d(f kt (X m ),CL kt q’ )]
[0101] Ψ k =[Ψ k1, Ψ k2 ,……,Ψ kT ], where Ψ kt =[d(f kt (x m ),cl kt 1 ),…,d(f kt (x m ),cl kt q’ )]
[0102] Where d is the formula for calculating the similarity of each cluster in the clustering results from user transaction features, and Φ k For the account-based features, Ψ represents the user operation features in the k-th business scenario. Similarly, Ψ k Φ represents the user management characteristics of the k-th business scenario, defined by the cardiology feature. kn Let Ψ be the user management feature of the nth business attribute in the kth business scenario. ki Let KJ be the user management feature of the nth business attribute in the kth business scenario.
[0103] Based on this, the above process is repeated for business attributes in the same region at different times to generate user operation tags (i.e., tuples (Φk,Ψk), which are user operation features in the kth scenario).
[0104] In addition, after step 130, the data processing method provided in this application embodiment may also include a process of filtering user business tags, which can be referred to in steps 140 and 150 below.
[0105] Step 140: According to the kth business scenario among the K business scenarios, construct absolute semantic tags and relative semantic tags for the user terminal based on user operation tags; wherein, the absolute semantic tags are used to characterize the user terminal's transaction preferences relative to a preset threshold within the p-th layer transaction statistics period and the m-th region of M regions, and the relative semantic tags are used to characterize the user terminal's transaction preferences relative to M regions within the p-th layer transaction statistics period and the m-th region of M regions.
[0106] For example, taking a business scenario as a unit, for each user terminal that needs to be analyzed when operating a specific user, determine the absolute statistics of whether or not the user likes the business tag in a specific region and time period (e.g., the cumulative transaction amount exceeds 1,000 yuan); and the relative statistics of whether or not the user likes the business tag in a specific region and time period (e.g., the cumulative transaction amount accounts for more than 20%).
[0107] Step 150: Correct the user management tags using absolute semantic tags and relative semantic tags to obtain target user management tags that correspond to the business scenario.
[0108] Specifically, in one example, step 150 may include:
[0109] Step 1501: According to the first target layer transaction statistics period in the P layer transaction statistics period that meets the preset requirements, evaluate the absolute semantic label and the relative semantic label through the preset input stability index to obtain the first evaluation result.
[0110] Step 15021: If the stability of the absolute semantic label and the relative semantic label of the first evaluation result in the transaction statistics period of the first target layer meets the preset conditions, the user operation label corresponding to the evaluation result is determined as the target user operation label.
[0111] or,
[0112] Step 15022: If the stability of the absolute semantic tag and the relative semantic tag in the first target layer transaction statistics period does not meet the preset conditions, the absolute semantic tag and the relative semantic tag are evaluated according to the second target layer transaction period through the preset model stability index to obtain the second evaluation result. The duration of the second target layer transaction period is shorter than the first target layer transaction period and is the transaction period of the adjacent layer of the first target layer transaction period. Based on the second evaluation result, it is determined as the target user operation tag.
[0113] or,
[0114] Step 15023: If the stability of the absolute semantic label and the relative semantic label represented by the second evaluation result does not meet the preset conditions in the second target layer transaction statistics period and the second target layer transaction statistics period is the terminal layer transaction statistics period in the P layer transaction statistics period, remove the user operation label corresponding to the second evaluation result from the user operation label to obtain the target user operation label.
[0115] or,
[0116] Step 15023: If the stability of the absolute semantic label and the relative semantic label in the first evaluation result does not meet the preset conditions in the first target layer transaction statistics period, and there is no second target layer transaction statistics period in the P layer transaction statistics period, remove the user operation label corresponding to the second evaluation result from the user operation label to obtain the target user operation label.
[0117] For example, based on the generated user management tags, starting from the longest historical time period, i.e., the first target layer transaction statistics time period (e.g., month), stability is evaluated sequentially according to the model entry stability indicators (e.g., KS, IV, PSI). If the model entry stability of the current time period is insufficient, user management tags from multiple more granular time periods within the same time period, i.e., the second target layer transaction statistics time period (e.g., the week corresponding to the month), are used to replace the original user management tags until the model entry stability requirement is met or no more granular time periods can be found. If the model entry stability requirement is met, the resulting user management tags for different scenarios are the filtered user management tags (business scenario S). k The features obtained from the next screening are denoted as (Φ′). k ,Ψ′ k (i.e., the user's business characteristics in the kth scenario after filtering by user business characteristics), otherwise discard the user's transaction characteristics in that group.
[0118] In one example, prior to performing step 150, the method may further include:
[0119] Using scenarios as units, and combining absolute and relative statistics, determine the binary classification label logic for each scenario preference (e.g., if either one is satisfied, it is considered a preference; otherwise, it is considered no preference).
[0120] It should be noted that: 1) The meaning of the card-person relationship in this application embodiment is to have the function of identifying multiple transaction flows generated by the same cardholder. In practice, the difference in association method and name should not be considered as having a substantial difference from the application embodiment. 2) Differences in the selection of specific clustering algorithms (k-means, spectral clustering) for user operation tags should not be considered as having a substantial difference from this patent. Relevant technical personnel should understand the essence of this application embodiment from the perspective of clustering of business attribute statistics. Similarly, similarity calculation should not be limited to the Euclidean distance of the above embodiments. 3) The design of multiple sets of business attributes in each business scenario is related to the transaction scenario. In actual application, the statistical caliber and / or composition of different business attributes adopted from the embodiments should not be considered as having a substantial difference from the application embodiment. 4) After selecting the business attribute, the user operation tag should consider both the card base contribution and the card's contribution relative to the account base in terms of composition. 5) After a new user transaction feature is generated, the clustering integration result of the generated account base and card base does not need to be recalculated. It is only necessary to calculate the similarity of the business attribute value with respect to this clustering integration result based on the similarity calculation formula d. 6) In the user management tag screening embodiments of this application, it is assumed that the target users have been determined in some way. 7) In the user management tag screening stage, the setting of absolute and relative statistics for each scenario, and how the two are combined to define the calculation method of scenario preferences, should not be limited to the specific values and implementation methods in the embodiments in practice, but should be understood as the elements for constructing binary transaction scenario preference tags.
[0121] In summary, the data processing method, apparatus, device, and storage medium of this application embodiment can acquire multiple sets of user transaction features, each set of user transaction features including target user transaction features on the user end. Next, according to the clustering dimension, each set of user transaction features is clustered using N clustering algorithms to obtain clustering results; wherein, the clustering dimension includes a P-layer transaction statistical time period, where the duration of each layer of the P-layer transaction statistical time period is different, and the clustering results include the set of clusters corresponding to the nth clustering algorithm among the N clustering algorithms, within the p-th layer of the P-layer transaction statistical time period. Then, based on the clustering results, user management tags corresponding to the business scenario are determined, and user management tags include those used to characterize the similarity between the target user transaction features in each set of user transaction features within the p-th layer of the transaction statistical time period and the clusters corresponding to the nth clustering algorithm. This approach leverages the advantages of clustering user transaction features using N clustering algorithms across each transaction statistical time period within the P-layer transaction statistical timeframe. This generates a more rational cluster structure. By combining the similarity between user transactions within the p-layer transaction statistical timeframe and the cluster corresponding to the n-th clustering algorithm, it facilitates the acquisition of fuzzy membership information of target user transaction features across various clusters. This helps to further refine user-side tags, enabling the service platform to obtain richer user features. This allows for the use of machine learning methods to uncover implicit transaction patterns across different scenarios and transaction statistical timeframes. Furthermore, by progressively transforming transaction statistical timeframes from coarse to fine, it identifies effective feature combinations from large-scale transaction patterns that influence user transaction preferences across different scenarios. This enriches the construction logic of the tag system and improves the accuracy of the service platform in providing targeted services based on tags. Additionally, the distribution of user transaction data across various channels in terms of basic consumption characteristics takes into account both card-based and account-based factors. Aside from the user operation tag generation and screening methods, no additional technical modifications are required, and reliance on business experience is reduced. By comprehensively utilizing multiple clustering strategies (cluster ensemble), each layer of transaction features can take into account the advantages of multiple clustering hypotheses, generating a more reasonable cluster structure. Combining user transaction information regarding the fuzzy affiliation of various clusters helps to further refine customer segmentation.
[0122] Furthermore, based on the user management tags, absolute and relative semantic tags are constructed for the user in the k-th business scenario out of K business scenarios. The absolute semantic tags characterize the user's transaction preferences relative to a preset threshold within the p-th layer of transaction statistics and the m-th region out of M regions. The relative semantic tags characterize the user's transaction preferences relative to the M regions within the p-th layer of transaction statistics and the m-th region out of M regions. These absolute and relative semantic tags are used to correct the user management tags, resulting in target user management tags corresponding to the business scenarios. Therefore, combining the dual-quantitative preference characteristics of transaction scenarios to generate multi-level tags helps form tag combinations with regional differences. With richer expressive power and volume, user management tags not only supplement the existing tag system but also facilitate scenario-based marketing and optimize the effectiveness of joint modeling in cross-institutional user collaborative management.
[0123] It should be noted that the data processing method provided in this application embodiment serves cardholders in a target-end mode. Its transaction channels and scenarios are essentially the same as the switching and clearing system involved in this application embodiment. Related user management strategies may be matched with the modeling results of cardholder transaction flows. Therefore, it can be applied to scenarios such as high-net-worth user discovery and dormant card activation for financial institutions. This makes the common technology for constructing user management tags, which is the focus of this application embodiment, potentially usable by financial institutions. Furthermore, since not all tags contribute to specific user management businesses, and user profile tag screening is a standardized process, the data processing method provided in this application embodiment can also be applied to hierarchical user management tag screening application scenarios.
[0124] Based on the same inventive concept, this application also provides a data processing device. (Specifically combined with...) Figure 2 Please provide a detailed explanation.
[0125] Figure 2 This is a schematic diagram of the structure of a data processing apparatus provided in one embodiment of this application.
[0126] In some embodiments of this application, Figure 2 The data processing device shown can be installed in the server of this application embodiment.
[0127] like Figure 2 As shown, the data processing device 20 may specifically include:
[0128] The acquisition module 201 is used to acquire multiple sets of user transaction features, each set of user transaction features including the target user transaction features on the user side;
[0129] Clustering module 202 is used to cluster each group of user transaction features according to the clustering dimension using N clustering algorithms to obtain clustering results. The clustering dimension includes P-layer transaction statistical time periods, each layer of the transaction statistical time period in the P-layer transaction statistical time period has a different duration. The clustering results include the set of clusters for each group of user transaction features within the p-th layer of the P-layer transaction statistical time period and corresponding to the n-th clustering algorithm among the N clustering algorithms.
[0130] The determination module 203 is used to determine the user operation label corresponding to the business scenario based on the clustering results. The user operation label includes the degree of similarity between the target user transaction feature in each group of user transaction features and the cluster corresponding to the nth clustering algorithm within the transaction statistics time period of the p-th layer.
[0131] The data processing device 20 in the embodiments of this application will be described in detail below.
[0132] In one or more optional embodiments, the target user transaction characteristics include card-based characteristics and account-based characteristics. The card-based characteristics include transaction history characteristics of at least one payment card, and the account-based characteristics include transaction history characteristics of at least one transaction account. Based on this, the data processing device 20 in this embodiment may further include an extraction module; wherein...
[0133] The acquisition module 201 can also be used to acquire business-related data from the target end, which includes at least one of the card issuing end, service acceptance end, service end, and merchant end. The business-related data includes data on various types of entities and the transactions themselves involved in user transaction transfer and settlement for the user end stored on the target end.
[0134] The extraction module is used to extract user-side card-based features and account-based features from business-related data.
[0135] In another or more optional embodiments, the clustering result includes a first clustering result corresponding to the cardiology feature and a second clustering result corresponding to the account-based feature. Based on this, the clustering module 202 in this application embodiment can specifically be used to perform N differential clustering operations on the cardiology feature according to the clustering dimension using N clustering algorithms to obtain a first clustering result, wherein the first clustering result includes the set of clusters corresponding to the cardiology feature in the p-th layer transaction statistical time period of the P-th layer transaction statistical time period and the n-th clustering algorithm among the N clustering algorithms; and,
[0136] According to the clustering dimension, the account base features are clustered N times with differences using N clustering algorithms to obtain the second clustering result. The second clustering result includes the set of clusters of the account base features within the p-th layer transaction statistical time period of the P-layer transaction statistical time period and corresponding to the n-th clustering algorithm among the N clustering algorithms.
[0137] In another or more alternative embodiments, the clustering dimension further includes M regions; the clustering result includes a set of clusters corresponding to the nth clustering algorithm among the N clustering algorithms, for each group of user transaction characteristics within the mth region of the M regions, within the pth layer transaction statistical time period of the P layer transaction statistical time period.
[0138] In another or more alternative embodiments, the clustering dimension may further include K business scenarios;
[0139] The clustering results include the set of clusters corresponding to the nth clustering algorithm among the N clustering algorithms, for each group of user transaction characteristics in the kth business scenario out of the K business scenarios, within the pth layer transaction statistical time period of the P layer transaction statistical time period;
[0140] or,
[0141] The clustering results include a set of clusters corresponding to the nth clustering algorithm among the N clustering algorithms, for the kth business scenario among the K business scenarios and the mth region among the M regions, where the user transaction characteristics of each group are within the pth layer transaction statistical time period of the P layer transaction statistical time period.
[0142] In another or more alternative embodiments, the clustering dimensions include P-layer transaction statistical time periods, M regions, and K business scenarios; based on this, the data processing device 20 in this embodiment may further include a statistics module; wherein,
[0143] The acquisition module 201 can also be used to acquire the basic user transaction characteristics of the mth region in M regions within the pth layer transaction statistics time period, taking the kth business scenario in K business scenarios as a unit and extracting the basic user transaction characteristics of the mth region in each group of user transaction characteristics.
[0144] The statistics module is used to statistically analyze the T groups of business attributes representing basic user transaction characteristics.
[0145] The clustering module 202 is further configured to: cluster the target user transaction features in each group of user transaction features according to the p-th layer transaction statistical time period, the m-th region, and the k-th business scenario, using N clustering algorithms, to obtain a first cluster set corresponding to the n-th clustering algorithm among the N clustering algorithms for the target user transaction features in each group of user transaction features within the p-th layer transaction statistical time period, the m-th region, and the k-th business scenario; and cluster the t-th group of business attributes and the target basic user transaction features corresponding to the t-th group of business attributes in T groups of business attributes according to the p-th layer transaction statistical time period, the m-th region, and the k-th business scenario, using N clustering algorithms, to obtain a second cluster set corresponding to the n-th clustering algorithm among the N clustering algorithms and the t-th group of business attributes for the target basic user transaction features within the p-th layer transaction statistical time period, the m-th region, and the k-th business scenario.
[0146] The determination module 203 is also used to determine the first cluster set and the second cluster set as the clustering result.
[0147] In another or more alternative embodiments, the data processing device 20 in this application embodiment may further include a first calculation module and a first evaluation module; wherein,
[0148] The determining module 203 is further configured to determine the average information entropy of the clusters based on the clustering results. The average information entropy of the clusters includes the average information entropy of the target user's transaction features in the kth business scenario and the mth region, which is composed of the jth cluster corresponding to the nth clustering algorithm of the tth group of business attributes, relative to the cluster set corresponding to the n'th clustering algorithm of the same business attribute.
[0149] The first calculation module is used to calculate the basic cluster average information entropy of the target user's transaction features based on the average information entropy of the clusters. The basic cluster average information entropy includes the average information entropy of the cluster set corresponding to the target user's transaction features in the kth business scenario and the mth region, and the tth group of business attributes, relative to the cluster set corresponding to each of the N clustering algorithms for the same business attribute.
[0150] The first evaluation module is used to evaluate the clustering results and obtain the target clustering results by using the average information entropy of the basic clustering under the kth business scenario as the evaluation condition.
[0151] The determination module 203 can also be used to determine user management tags corresponding to the business scenario based on the target clustering results.
[0152] In another or more alternative embodiments, the data processing device 20 in this application embodiment may further include a filtering module, a first removal module, and an execution module; wherein,
[0153] The filtering module is used to filter the target cluster set from the clustering results based on the average information entropy of the basic clustering.
[0154] The first removal module is used to remove cluster sets from the clustering result set to obtain the current clustering result;
[0155] The execution module is used to determine the current clustering result as the clustering result, repeatedly execute the process of determining the average information entropy of the clusters based on the clustering result, and calculating the basic clustering average information entropy of the target user's transaction characteristics based on the average information entropy of the clusters; and, until the number of repeated executions meets a preset condition, determine the current clustering result set corresponding to the number of repeated executions meeting the preset condition as the target clustering result; or, until the difference between the basic clustering average information entropies of h consecutive rounds is less than or equal to a preset threshold, determine the current clustering result set corresponding to the last round of h consecutive rounds as the target clustering result.
[0156] In another or more alternative embodiments, the data processing device 20 in this application embodiment may further include a second computing module; wherein,
[0157] The second calculation module is used to calculate the similarity degree between the target user transaction features in each group of user transaction features and the cluster sets corresponding to n clustering algorithms within the p-th layer transaction statistical time period, the m-th region and the k-th business scenario, for the t-th group of business attributes.
[0158] The determination module 203 is further used to determine the similarity of the target user transaction features in each group of user transaction features to the cluster set corresponding to the n clustering algorithms within the transaction statistics time period of the p-th layer, the m-th region and the k-th business scenario, for the t-th group of business attributes, as user operation tags.
[0159] In another or more alternative embodiments, the data processing device 20 in this application embodiment may further include a construction module and a correction module; wherein,
[0160] The construction module is used to construct absolute semantic tags and relative semantic tags for the user terminal based on the user operation tags, according to the kth business scenario among K business scenarios. The absolute semantic tags are used to represent the user terminal's transaction preferences relative to a preset threshold within the p-th transaction statistics period and the m-th region among M regions, while the relative semantic tags are used to represent the user terminal's transaction preferences relative to M regions within the p-th transaction statistics period and the m-th region among M regions.
[0161] The correction module is used to correct user management tags using absolute and relative semantic tags to obtain target user management tags that correspond to the business scenario.
[0162] In another or more alternative embodiments, the data processing device 20 in this application embodiment may further include a second evaluation module; wherein,
[0163] The second evaluation module is used to evaluate the absolute semantic labels and relative semantic labels by means of a preset input stability index for the first target layer transaction statistical time period that meets the preset requirements in the P layer transaction statistical time period, and obtain the first evaluation result.
[0164] The determining module 203 is further configured to, when the stability of the absolute semantic label and the relative semantic label in the first evaluation result characterizes the transaction statistics period of the first target layer as a target user operation label, determine the user operation label corresponding to the evaluation result as the target user operation label.
[0165] In another or more alternative embodiments, the data processing device 20 in this application embodiment may further include a third evaluation module; wherein,
[0166] The third evaluation module is used to evaluate the absolute semantic label and the relative semantic label according to the second target layer transaction time period and through the preset model stability index when the stability of the absolute semantic label and the relative semantic label in the first evaluation result does not meet the preset conditions. The second evaluation result is obtained when the duration of the second target layer transaction time period is less than that of the first target layer transaction time period and is the transaction time period of the adjacent layer of the first target layer transaction time period.
[0167] The determination module 203 is also used to determine the target user operation tag based on the second evaluation result.
[0168] In another or more alternative embodiments, the data processing apparatus 20 in this application embodiment may further include a second removal module; wherein,
[0169] The second removal module is used to remove the user management label corresponding to the second evaluation result from the user management label when the stability of the absolute semantic label and the relative semantic label in the second evaluation result does not meet the preset conditions in the second target layer transaction statistics period and the second target layer transaction statistics period is the termination layer transaction statistics period in the P layer transaction statistics period, so as to obtain the target user management label.
[0170] In another or more alternative embodiments, the data processing apparatus 20 in this application embodiment may further include a third removal module; wherein,
[0171] The third removal module is used to remove the user management label corresponding to the second evaluation result from the user management label when the stability of the absolute semantic label and the relative semantic label in the first evaluation result does not meet the preset conditions in the first target layer transaction statistics period and there is no second target layer transaction statistics period in the P layer transaction statistics period, so as to obtain the target user management label.
[0172] Therefore, multiple sets of user transaction features can be obtained, each set including the target user transaction features on the user side. Next, according to the clustering dimension, each set of user transaction features is clustered using N clustering algorithms to obtain clustering results. The clustering dimension includes a P-layer transaction statistical time period, where each layer of the P-layer transaction statistical time period has a different duration. The clustering results include the set of clusters for each set of user transaction features within the p-th layer of the P-layer transaction statistical time period that correspond to the n-th clustering algorithm among the N clustering algorithms. Then, based on the clustering results, user management tags corresponding to the business scenario are determined. These user management tags include a measure of the similarity between the target user transaction features in each set of user transaction features within the p-th layer of the transaction statistical time period and the clusters corresponding to the n-th clustering algorithm. In this way, by using N clustering algorithms to cluster each group of user transaction features according to the clustering dimension and the transaction statistical time period of each layer of the P-layer transaction statistical time period, a more reasonable cluster structure is generated. Combining the similarity between user transactions in the p-layer transaction statistical time period and the clusters corresponding to the n-th clustering algorithm, it is beneficial to obtain fuzzy membership information of target user transaction features with respect to various clusters. This helps to further refine the user-side tags, thereby enabling the service platform to obtain richer user features. This allows the platform to use machine learning methods to mine implicit transaction patterns of users in different scenarios and different layers of transaction statistical time periods. By progressively transforming the transaction statistical time periods from coarse to fine, the platform can find effective feature combinations from large-scale transaction patterns that help understand users' transaction preferences in different scenarios. This enriches the construction logic of the tag system and improves the accuracy of the service platform in providing targeted services to users based on tags.
[0173] Based on the same inventive concept, this application also provides a computer device. (Specifically combined with...) Figure 3 Please provide a detailed explanation.
[0174] Figure 3 This is a schematic diagram of the structure of a computer device provided in one embodiment of this application.
[0175] like Figure 3As shown, the computer device may include at least one of the following as described in the embodiments of this application: an electronic device, a server. The computer device may include a processor 301 and a memory 302 storing computer program instructions.
[0176] Specifically, the processor 301 may include a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.
[0177] Memory 302 may include a large-capacity memory for data or instructions. For example, and not limitingly, memory 302 may include a hard disk drive (HDD), a floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, or a Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 302 may include removable or non-removable (or fixed) media. Where appropriate, memory 302 may be internal or external to the integrated gateway disaster recovery device. In a particular embodiment, memory 302 is non-volatile solid-state memory. In a particular embodiment, memory 302 includes solid-state storage (ROM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable ROM (PROM), an erasable PROM (EPROM), an electrically erasable PROM (EEPROM), an electrically rewritable ROM (EAROM), or flash memory, or a combination of two or more of these.
[0178] The processor 301 implements any of the data processing methods described in the above embodiments by reading and executing computer program instructions stored in the memory 302.
[0179] In one example, the computer device may also include a communication interface 303 and a bus 310. Wherein, as... Figure 3 As shown, the processor 301, memory 302, and communication interface 303 are connected through bus 310 and complete communication with each other.
[0180] The communication interface 303 is mainly used to realize communication between various modules, devices, units and / or equipment in the embodiments of this application.
[0181] Bus 310 includes hardware, software, or both, that couples components of a flow control device together. For example, and not limitingly, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count (LPC) bus, a memory bus, a Microchannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or other suitable buses, or combinations of two or more of these. Where appropriate, bus 310 may include one or more buses. Although specific buses are described and illustrated in embodiments of this application, any suitable bus or interconnect is contemplated herein.
[0182] The data processing device can execute the data processing method described in the embodiments of this application, thereby achieving the combination Figures 1 to 2 The data processing methods and apparatus described.
[0183] Furthermore, in conjunction with the data processing methods in the above embodiments, this application embodiment can provide a computer-readable storage medium for implementation. This computer-readable storage medium stores computer program instructions; when executed by a processor, these computer program instructions implement any of the data processing methods in the above embodiments.
[0184] It should be clarified that this application is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of this application is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of this application.
[0185] The functional blocks shown in the above block diagram can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this application are programs or code segments used to perform the required tasks. Programs or code segments can be stored on a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried on a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, etc. Code segments can be downloaded via computer networks such as the Internet, intranets, etc.
[0186] It should also be noted that the exemplary embodiments mentioned in this application describe methods or systems based on a series of steps or apparatus. However, this application is not limited to the order of the above steps; that is, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.
[0187] The above are merely specific embodiments of this application. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the protection scope of this application.
Claims
1. A data processing method, comprising: Acquire multiple sets of user transaction features, each set of user transaction features including target user transaction features on the user side; According to the clustering dimension, each group of user transaction features is clustered using N clustering algorithms to obtain clustering results; wherein, the clustering dimension includes P-layer transaction statistical time periods, each layer of transaction statistical time periods in the P-layer transaction statistical time periods has a different duration, and the clustering results include the set of clusters of each group of user transaction features within the p-th layer of the P-layer transaction statistical time periods and corresponding to the n-th clustering algorithm among the N clustering algorithms; Based on the clustering results, user management tags corresponding to the business scenario are determined. The user management tags include the degree of similarity between the target user transaction features in each group of user transaction features and the clusters corresponding to the nth clustering algorithm within the transaction statistics time period of the p-th layer. The target user transaction features include card-based features and account-based features. The card-based features include transaction flow features of at least one payment card, and the account-based features include transaction flow features of at least one transaction account.
2. The method according to claim 1, wherein, The acquisition of multiple sets of user transaction characteristics includes: Obtain business-related data from the target end, wherein the target end includes at least one of the card issuing end, service acceptance end, service end, and merchant end, and the business-related data includes data on various types of entities and the transactions themselves involved in user transaction transfer and settlement for the user end stored on the target end; Extract the user's card-based features and account-based features from the business-focused data.
3. The method according to claim 2, wherein, The clustering results include a first clustering result corresponding to the cardiology feature and a second clustering result corresponding to the account base feature; The process involves clustering each group of user transaction features according to a clustering dimension using N clustering algorithms to obtain clustering results, including: According to the clustering dimension, the Kjeldahl features are subjected to N differential clustering operations using N clustering algorithms to obtain the first clustering result. The first clustering result includes the set of clusters corresponding to the Kjeldahl features within the p-th layer of the P-layer transaction statistics time period and corresponding to the n-th clustering algorithm among the N clustering algorithms. According to the clustering dimension, the account base features are clustered N times with differences using N clustering algorithms to obtain the second clustering result. The second clustering result includes the set of clusters of account base features within the p-th layer transaction statistical time period of the P-th layer transaction statistical time period and corresponding to the n-th clustering algorithm among the N clustering algorithms.
4. The method according to claim 1, wherein, The clustering dimension also includes M regions; the clustering result includes a set of clusters corresponding to the nth clustering algorithm among the N clustering algorithms, for each group of user transaction features within the mth region of the M regions, within the pth layer transaction statistical time period of the P layer transaction statistical time period.
5. The method according to claim 1 or 4, wherein, The clustering dimension also includes K business scenarios; The clustering result includes, for the kth business scenario among the K business scenarios, the set of clusters corresponding to the nth clustering algorithm among the N clustering algorithms within the pth layer transaction statistical time period of the P layer transaction statistical time period for each group of user transaction features; or, The clustering result includes a set of clusters corresponding to the nth clustering algorithm among the N clustering algorithms, for the kth business scenario among the K business scenarios and the mth region among the M regions, where each group of user transaction features is within the pth layer transaction statistical time period of the P layer transaction statistical time period.
6. The method according to any one of claims 1-5, wherein, The clustering dimensions include P-layer transaction statistics time periods, M geographical regions, and K business scenarios; the clustering results are obtained by clustering each group of user transaction features according to the clustering dimensions using N clustering algorithms, including: Taking the kth business scenario among the K business scenarios as a unit, the basic user transaction characteristics of the mth region among the M regions within the pth layer transaction statistical time period are obtained from each group of user transaction characteristics; Statistically analyze the T groups of business attributes representing the basic user transaction characteristics; Based on the p-th layer transaction statistics time period, the m-th region, and the k-th business scenario, N clustering algorithms are used to cluster the target user transaction features in each group of user transaction features, resulting in a first cluster set corresponding to the n-th clustering algorithm among the N clustering algorithms within the p-th layer transaction statistics time period, the m-th region, and the k-th business scenario. Additionally, based on the p-th layer transaction statistics time period, the m-th region, and the k-th business scenario, N clustering algorithms are used to cluster the t-th group of business attributes in the T groups of business attributes and the target basic user transaction features corresponding to the t-th group of business attributes, resulting in a second cluster set corresponding to the t-th group of business attributes within the p-th layer transaction statistics time period, the m-th region, and the k-th business scenario. The first cluster set and the second cluster set are determined as the clustering result.
7. The method according to claim 6, wherein, The step of determining user management tags corresponding to the business scenario based on the clustering results includes: Based on the clustering results, the average information entropy of the clusters is determined. The average information entropy of the clusters includes the average information entropy of the target user's transaction features in the kth business scenario and the mth region, which is composed of the jth cluster corresponding to the nth clustering algorithm of the tth group of business attributes, relative to the cluster set corresponding to the n'th clustering algorithm of the same business attribute. Based on the average information entropy of the clusters, the basic cluster average information entropy of the target user's transaction features is calculated. The basic cluster average information entropy includes the average information entropy of the cluster set corresponding to the n clustering algorithms for the target user's transaction features in the kth business scenario and the mth region, and the tth group of business attributes, relative to the cluster set corresponding to each of the N clustering algorithms for the same business attribute. Using the average information entropy of the basic clustering under the kth business scenario as the evaluation condition, the clustering results are evaluated to obtain the target clustering result; Based on the target clustering results, user management tags corresponding to the business scenario are determined.
8. The method according to claim 7, wherein, The evaluation of the clustering results using the average information entropy of the basic clustering under the k-th business scenario as the evaluation criterion, to obtain the target clustering result, includes: Based on the average information entropy of the basic clustering, a set of target clusters is selected from the clustering results; Remove inferior basic clustering results from the target cluster set to obtain the current clustering result. The inferior basic clustering results include clustering results whose average information entropy of basic clusters is greater than a preset threshold. The current clustering result is determined as the clustering result, and the process of determining the average information entropy of the clusters based on the clustering result and calculating the basic cluster average information entropy of the target user's transaction characteristics based on the average information entropy of the clusters is repeated. The current target cluster set corresponding to the number of repetitions meeting the preset conditions is determined as the target clustering result; or, the current target cluster set corresponding to the last round of the h consecutive rounds is determined as the target clustering result when the difference between the average information entropy of the basic clusters in h consecutive rounds is less than or equal to a preset threshold.
9. The method according to claim 7 or 8, wherein, The step of determining user management tags corresponding to the business scenario based on the target clustering results includes: The similarity calculation algorithm is used to calculate the similarity between the target user transaction features in each group of user transaction features and the cluster sets corresponding to the n clustering algorithms within the p-th layer transaction statistics time period, the m-th region, and the k-th business scenario, for the t-th group of business attributes. The similarity between the target user transaction features in each group of user transaction features and the cluster sets corresponding to the n clustering algorithms within the p-th layer transaction statistics time period, the m-th region, and the k-th business scenario, for the t-th group of business attributes, is determined as the user management tag.
10. The method according to claim 1, wherein, The method further includes: Based on the user management tag, and according to the kth business scenario out of K business scenarios, an absolute semantic tag and a relative semantic tag are constructed for the user terminal. The absolute semantic tag characterizes the user terminal's transaction preference relative to a preset threshold within the p-th layer transaction statistics period and the m-th region of M regions. The relative semantic tag characterizes the user terminal's transaction preference relative to the M regions within the p-th layer transaction statistics period and the m-th region of M regions. The user management tag is the user management tag obtained within the k-th business scenario, the m-th region, and the p-th layer transaction statistics period. The user management tags are corrected using the absolute semantic tags and relative semantic tags to obtain target user management tags corresponding to the business scenario.
11. The method according to claim 10, wherein, The step of correcting the user management tags using the absolute semantic tags and relative semantic tags to obtain target user management tags corresponding to the business scenario includes: According to the first target layer transaction statistics period whose duration meets the preset requirements in the P-layer transaction statistics period, the absolute semantic label and relative semantic label are evaluated by the preset model stability index to obtain the first evaluation result; If the stability of the absolute semantic tag and the relative semantic tag in the transaction statistics period of the first target layer, as characterized by the first evaluation result, meets the preset conditions, the user management tag corresponding to the evaluation result will be determined as the target user management tag.
12. The method according to claim 11, wherein, The step of correcting the user management tags using the absolute semantic tags and relative semantic tags to obtain target user management tags corresponding to the business scenario includes: If the stability of the absolute semantic label and the relative semantic label in the first target layer transaction statistics period does not meet the preset conditions, the absolute semantic label and the relative semantic label are evaluated according to the second target layer transaction period by the preset model stability index to obtain the second evaluation result. The duration of the second target layer transaction period is less than the first target layer transaction period and is the transaction period of the adjacent layer of the first target layer transaction period. Based on the second evaluation result, the target user's management tag is determined.
13. The method according to claim 12, wherein, The step of correcting the user management tags using the absolute semantic tags and relative semantic tags to obtain target user management tags corresponding to the business scenario includes: If the stability of the absolute semantic tag and the relative semantic tag in the second target layer transaction statistics period does not meet the preset conditions, and the second target layer transaction statistics period is the termination layer transaction statistics period in the P layer transaction statistics period, the user operation tag corresponding to the second evaluation result is removed from the user operation tag to obtain the target user operation tag.
14. The method according to claim 12, wherein, The step of correcting the user management tags using the absolute semantic tags and relative semantic tags to obtain target user management tags corresponding to the business scenario includes: If the stability of the absolute semantic tag and the relative semantic tag in the first target layer transaction statistics period does not meet the preset conditions, and there is no second target layer transaction statistics period in the P layer transaction statistics period, the user management tag corresponding to the first evaluation result is removed from the user management tag to obtain the target user management tag.
15. A data processing apparatus, comprising: The acquisition module is used to acquire multiple sets of user transaction features, each set of user transaction features including the target user transaction features on the user side; The clustering module is used to cluster each group of user transaction features according to a clustering dimension using N clustering algorithms to obtain clustering results. The clustering dimension includes a P-layer transaction statistical time period, where the duration of each layer of the P-layer transaction statistical time period is different. The clustering result includes the set of clusters for each group of user transaction features within the p-th layer of the P-layer transaction statistical time period and corresponding to the n-th clustering algorithm among the N clustering algorithms. The determination module is used to determine user operation tags corresponding to the business scenario based on the clustering results. The user operation tags include the degree of similarity between the target user transaction features in each group of user transaction features and the clusters corresponding to the nth clustering algorithm within the transaction statistics time period of the p-th layer. The target user transaction features include card-based features and account-based features. The card-based features include transaction flow features of at least one payment card, and the account-based features include transaction flow features of at least one transaction account.
16. A computer device, the device comprising: Processor and memory storing computer program instructions; When the processor executes the computer program instructions, it implements the data processing method as described in any one of claims 1-14.
17. A storage medium storing computer program instructions, which, when executed by a processor, implement the data processing method as described in any one of claims 1-14.
Citation Information
Patent Citations
Business data processing method and device, electronic equipment and medium
CN110457175A
User service demand identification method and device and computer readable storage medium
CN113256351A
