Intelligent classification management system for electronic transaction data based on artificial intelligence

Abstract

The present application relates to the field of data processing technology, and proposes an intelligent classification management system for electronic transaction data based on artificial intelligence, including: constructing the data correlation of transaction data according to the size of the element values ​​in the adjacent matrix corresponding to different walking steps of each type of transaction data; constructing the transaction impact ratio according to the different relevance of each type of transaction data to the detection of abnormal transaction data; and obtaining the detection results of abnormal transaction data in group purchase transaction data according to the data correlation and transaction impact ratio. The present application uses the data correlation and transaction impact ratio to construct the metric distance and inter-cluster similarity in the classification algorithm, avoiding the problem of misjudgment caused by judging abnormal transactions through the distribution of transaction data in different time periods during the group purchase period, and improving the detection accuracy of transaction data and the efficiency of classification management.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to an intelligent classification management system for electronic transaction data based on artificial intelligence. Background Art

[0002] With the development of Internet technology and the change of people's consumption concept, group buying has spread rapidly in major cities as a new consumption mode. The general group buying process is as follows: group buying personnel organize and share to determine the name of the group buying product and the group leader, the group leader signs a group buying contract with the supplier, and the group buying platform initiates the purchase demand to the supplier. Consumers log in to the group buying platform, select the group buying product and pay the fee.

[0003] The group buying process involves a large amount of consumer personal information data, product information data, transaction data and other types of data information. The transaction data includes transaction data between group buyers and group buying platforms, transaction data between group buying platforms and suppliers, and transaction data between group leaders and group buying platforms. In the process of transactions between different objects, they will be affected by network fluctuations, malicious purchases and other situations, resulting in abnormal transaction data. Therefore, the group buying platform needs to perform anomaly detection from a large amount of group buying personnel's transaction data, and send the detected transaction data to suppliers and group leaders. The group leaders will conduct subsequent group buying processes based on the personal transaction data of group buyers.

[0004] At present, abnormal data detection algorithms in the field of artificial intelligence are usually used to detect and manage group purchase transaction data, such as the commonly used support vector machine SVM, outlier factor detection LOF, isolation forest, etc. Among them, SVM needs to select data features with higher discrimination when detecting abnormal data; and LOF needs to pre-set parameter values, which has a greater impact on the detection results; the isolation forest algorithm lacks a noise processing mechanism when performing abnormal data detection tasks, and the recognition accuracy of real abnormal points is low. This type of algorithm not only has high false positive and false negative rates, but also has difficulty adapting to the various types of transaction data generated by group purchasers during the group purchase process. Summary of the invention

[0005] This application provides an intelligent classification management system for electronic transaction data based on artificial intelligence to solve the problem of difficulty in classification management of various types of transaction data of existing group purchase personnel. The technical solutions adopted are as follows:

[0006] An embodiment of the present application provides an electronic transaction data intelligent classification management system based on artificial intelligence, which includes the following modules:

[0007] The transaction data collection module is used to obtain the transaction data of all group purchasers in the group purchase process and construct a transaction data matrix using the transaction data of all group purchasers;

[0008] A data association calculation module is used to obtain a preset number of adjacent matrices according to the transaction data matrix and different walking step lengths, obtain the sociability index of each group purchaser for each type of transaction data according to the preset number of adjacent matrices and the transaction data, and obtain the data association of each group purchaser's transaction data according to the difference in the sociability index of each type of transaction data between each group purchaser and the remaining group purchasers;

[0009] A transaction impact ratio calculation module is used to obtain the adjacent coefficients of the two group purchasers with respect to each type of transaction data according to the corresponding element values ​​of the two group purchasers in the adjacent matrix corresponding to each type of transaction data and the maximum value of the elements in the adjacent matrix, and to divide each type of transaction data of all group purchasers into a first similar transaction data set and a second similar transaction data set according to the classification algorithm, and to obtain the transaction impact ratio of each type of transaction data according to the first similar transaction data set and the second similar transaction data set;

[0010] The transaction data classification module is used to obtain the measured distance between the transaction data of group purchasers according to the data correlation and transaction impact ratio of each type of transaction data, and obtain the clustering clusters of the transaction data of all group purchasers according to the classification algorithm;

[0011] The transaction data management module is used to use the transaction data of the group purchase personnel corresponding to the cluster center of each cluster as the representative data of the cluster, obtain the inter-cluster similarity of each representative data according to the DTW distance between the representative data of different clusters, use the transaction data in the cluster with the minimum inter-cluster similarity as abnormal transaction data, and store the normal transaction data and abnormal transaction data in different databases according to their classification;

[0012] The method for obtaining the sociability index of each group purchaser for each type of transaction data based on a preset number of adjacent matrices and transaction data is:

[0013] , is the sociability index of the j-type transaction data of the i-th group purchaser, K is the maximum walking step length, m is the number of people participating in the group purchase, is the element value of the i-th row and t-th column in the adjacent matrix obtained by the random walk algorithm when the walk step length is b. , are the normalized results of the j-th transaction data of the i-th group purchaser and the j-th transaction data of the t-th group purchaser in the transaction data matrix, respectively. It is the preset tuning factor.

[0014] Preferably, the method for obtaining a preset number of adjacent matrices according to the transaction data matrix and different walking step lengths is:

[0015] Each type of transaction data in the transaction data matrix is ​​taken as a node to form a connection graph as input, different walking step sizes are preset, and the k-path random walk algorithm is used to perform random walks on each type of transaction data to obtain the adjacent matrix.

[0016] Preferably, the method for obtaining the data association degree of each group purchaser's transaction data according to the difference in the sociability index of each type of transaction data between each group purchaser and the other group purchasers is:

[0017] Obtain the sociability index of each type of transaction data of each group purchaser respectively, and obtain the mean sociability index of each type of transaction data according to the sociability index of each type of transaction data of all group purchasers;

[0018] The natural constant is used as the base, and the difference between the sociability index of each category of transaction data of each group purchasing personnel and the mean of the sociability index of each category of transaction data is used as the index calculation result as the class correlation degree of each category of transaction data of each group purchasing personnel. The class correlation degrees of all categories of transaction data of each group purchasing personnel are accumulated and taken as the data correlation degree corresponding to the transaction data of each group purchasing personnel.

[0019] Preferably, the method for obtaining the adjacent coefficients of the two group purchasers for each type of transaction data according to the corresponding element values ​​of the two group purchasers in the adjacent matrix corresponding to each type of transaction data and the maximum value of the elements in the adjacent matrix is:

[0020] , is the adjacent coefficient between the i-th group purchaser and the t-th group purchaser in the j-th category transaction data, , are the element values ​​of the i-th row and t-th column and the t-th row and i-th column in the adjacent matrix obtained by the random walk algorithm when the walk step length is b. is the maximum value of the element in the adjacent matrix, and K is the maximum walking step length.

[0021] Preferably, the method of dividing each type of transaction data of all group purchase personnel into a first type of transaction data set and a second type of transaction data set according to a classification algorithm is:

[0022] Calculate the adjacent coefficients between any two group purchasers in each type of transaction data, and use the Otsu algorithm to obtain the segmentation threshold;

[0023] The transaction data whose adjacent coefficient is greater than the segmentation threshold is classified as the first category of data, and the set consisting of all the first category of data is recorded as the first similar data set; the transaction data whose adjacent coefficient is less than the segmentation threshold is classified as the second category of data, and the set consisting of all the second category of data is recorded as the second similar data set.

[0024] Preferably, the method for obtaining the transaction impact ratio of each type of transaction data according to the first similar transaction data set and the second similar transaction data set is:

[0025] For each type of transaction data, obtain the inter-class variance when the Otsu algorithm divides each type of transaction data into the first type of data and the second type of data;

[0026] The ratio of the inter-class variance to the intra-class variance of each class of transaction data is used as the transaction impact ratio of each class of transaction data.

[0027] Preferably, the intra-class variance of each type of transaction data is the sum of the distribution variance of a first data set of the same type and the distribution variance of a second data set of the same type for each type of transaction data.

[0028] Preferably, the method for obtaining the metric distance between the transaction data of group purchase personnel according to the data correlation degree and transaction impact ratio of each type of transaction data is:

[0029] , is the metric distance between the transaction data of the i-th group purchaser and the t-th group purchaser, , are the data correlations of the i-th group purchaser and the t-th group purchaser, is the transaction impact ratio of the jth type of transaction data, n is the number of transaction data categories in the group purchase process, , They are the normalized results of the j-th category transaction data of the ith group purchaser and the t-th group purchaser in the transaction data matrix respectively.

[0030] Preferably, the method for screening abnormal transaction data based on inter-cluster similarity and performing classified storage management is:

[0031] Using the k-means algorithm, all group purchasers are divided into a preset number of clusters based on the measurement between the transaction data of any two group purchasers;

[0032] The transaction data of group purchase personnel corresponding to the cluster center of each cluster is taken as the representative data, the mean of the DTW distance between each representative data and the representative data of all clusters is taken as the inter-cluster similarity of each representative data, and the transaction data corresponding to the cluster corresponding to the minimum inter-cluster similarity is taken as the abnormal transaction data.

[0033] The beneficial effects of the present application are as follows: in the present application, firstly, a transaction data matrix is ​​obtained based on various types of transaction data in the group purchasing process, and the correlation of the transaction data is constructed according to the size of the element values ​​in the adjacent matrix corresponding to each type of transaction data at different walking steps. The correlation reflects the strength of the correlation of the transaction data of each group purchaser through the consistency of the associated data, and solves the problem of misjudgment caused by judging abnormal transactions only by the size of several adjacent transaction data; secondly, an influence ratio is constructed for each type of transaction data on the detection of abnormal transaction data, and the significance of the influence of multiple types of transaction data of each group purchaser on the abnormal transaction data is considered; then, the metric distance in k-means clustering is constructed based on the correlation and influence ratio, which can accurately evaluate the similarity between multiple types of transaction data of different group purchasers; finally, the abnormal transaction data in the entire group purchase process is obtained by using the improved k-means algorithm, which improves the recognition accuracy of transaction data and facilitates the classification management of group purchase transaction data. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0035] Figure 1 A flowchart of an artificial intelligence-based intelligent classification management system for electronic transaction data provided in accordance with one embodiment of the present application. DETAILED DESCRIPTION

[0036] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0037] See also Figure 1 , which shows a flow chart of an artificial intelligence-based intelligent classification management system for electronic transaction data provided by an embodiment of the present application. The system includes: a transaction data collection module, a data correlation calculation module, a transaction impact ratio calculation module, a transaction data classification module, and a transaction data management module.

[0038] The transaction data collection module is used to obtain the transaction data of all group purchase personnel in the group purchase process and construct a transaction data matrix using the transaction data of all group purchase personnel.

[0039] Specifically, under normal circumstances, in the group purchase process, group purchasers place orders on the group purchase platform, and the group purchase platform provides the supplier with the transaction data of the group purchasers. Therefore, in the embodiment of the present application, all transaction data are obtained from the background data system of the group purchase platform. The transaction data is n types of numerical data, including but not limited to the specific transaction amount of each group purchaser, the number of purchased products, the transaction time, the settlement cycle of a single person, and the unit price of the goods. For any type of transaction data, in order to avoid the phenomenon of data loss during the data acquisition process, after obtaining each type of digital data in the transaction data, the data is preprocessed using the mean filling method. Mean filling is a well-known technology, and the specific process will not be repeated.

[0040] Furthermore, since different group purchasers have different demands for group purchase products, in order to eliminate the influence of extreme values, the transaction data of group purchasers after preprocessing is normalized, and the normalized transaction data is used to form a transaction data matrix A, the dimension of A is m*n, where It refers to the normalized result of the jth transaction data of the ith group purchaser, m is the number of people participating in the group purchase, and n is the number of categories of the transaction data. The normalization process includes but is not limited to the maximum value normalization and Z-score normalization, which are not specifically limited in this application. The normalization of digital data is a well-known technology, and the specific process will not be described in detail.

[0041]

[0042] It should be noted that for transaction time, the normalization process is: obtain the transaction time data of all group purchase personnel on that day, first convert the transaction time of each group purchase personnel into specific minute data, for example, if the transaction time is 12:39, the corresponding minute data is 12×60+39=759. This conversion is based on a total of 1440 minutes per day. The ratio of the minute data corresponding to each group purchase personnel to 1440 is used as the corresponding normalized result. The earlier the transaction time, the smaller the normalized result, and the later the transaction time, the larger the normalized result.

[0043] The data association calculation module is used to obtain a preset number of adjacent matrices based on the transaction data matrix and different walking step sizes, obtain the sociability index of each group purchaser for each type of transaction data based on the preset number of adjacent matrices and transaction data, and obtain the data association of each group purchaser's transaction data based on the difference in the sociability index of each type of transaction data between each group purchaser and other group purchasers.

[0044] Specifically, abnormal transaction data is often transaction data generated when malicious order brushing occurs or is affected by network fluctuations. If a data in the transaction data matrix A or the transaction data of a group purchaser is detected separately, it is difficult to obtain a reliable detection result. For example, the transaction amount corresponding to a group purchaser is relatively large, and it is impossible to determine whether it has a large actual demand or malicious order brushing. Therefore, in this application, the correlation between each transaction data and the remaining transaction data is considered. The more similar the correlation between the transaction data of two group purchasers and the remaining transaction data is, the more similar the transaction data of the two group purchasers are. For example, group purchasers 1 and 2 are both malicious order brushing personnel, and group purchasers 3 and 4 are both normal transaction personnel. Only through the transaction data of group purchasers 1 and 2, it is impossible to accurately judge whether it is abnormal transaction data, and abnormal consumption data is manifested as outlier data compared to a large amount of normal transaction data, that is, the correlation between the transaction data of group purchaser 1 and group purchasers 3 and 4 is similar to the correlation between the transaction data of group purchaser 2 and group purchasers 3 and 4.

[0045] Since the demand of each group purchaser for group purchase goods is random, that is, the multiple transaction data of each group purchaser has a certain degree of randomness, in order to improve the analysis efficiency of transaction data and mine similar data features in each type of transaction data, in this application, each type of transaction data in the transaction data matrix A is taken as a node to form a connection graph as input, and the k-path random walk algorithm is used to perform random walks on each type of transaction data to obtain an adjacent matrix, which is recorded as the adjacent matrix Y of each type of transaction data. The step interval of the random walk is set to [2,7]. Each value in the adjacent matrix Y represents the probability of reaching the remaining data when the data contained in each type of transaction is walked. The element values ​​are non-zero and zero, respectively, indicating whether there is a connection between the corresponding two nodes in the input connection graph. Non-zero indicates that the two transaction data may have similar data features, and the element value is zero, indicating that the two transaction data do not have similar data features. The k-path random walk to obtain the adjacent matrix is ​​a well-known technology, and the specific process will not be repeated.

[0046] Based on the above analysis, the data correlation degree P is constructed here to characterize the correlation strength between the transaction data of the group purchase personnel and other group purchase personnel. Calculate the data correlation degree of the transaction data of the i-th group purchase personnel :

[0047]

[0048]

[0049] In the formula, is the sociability index of the j-th category transaction data of the i-th group purchaser, j is the j-th category transaction data, , are the normalized results of the j-th transaction data of the i-th group purchaser and the j-th transaction data of the t-th group purchaser in the transaction data matrix A, m is the number of people participating in the group purchase, b is the walking step length, Y is the adjacent matrix corresponding to the walking step length b, is the element value of the i-th row and t-th column in the adjacent matrix Y, K is the maximum walking step length, and the value of K is 7. is a control factor, which is used to avoid the denominator being 0. The empirical value is 0.001.

[0050] in, The larger the value, the stronger the connectivity between the transaction data during the wandering process. and The more similar the data features are, the stronger the correlation between the two data is; on the other hand, The smaller the value, the more similar the demands of the i-th group purchaser and the t-th group purchaser for group purchase products are, that is, The larger the transaction data and The stronger the correlation of transaction data, the more likely it is that ... The larger the value, the more transaction data is reflected. The more consistent it is with the overall distribution of the m j-th category transaction data, the less likely it is to be an outlier that deviates from the majority of the j-th category transaction data.

[0051] is the data correlation of the transaction data of the i-th group purchaser, n is the number of categories of transaction data in the group purchase process, e is the natural base number, It is the mean of the sociability index of the j-th type of transaction data.

[0052] in, The larger the value is, the more normal the transaction data of the jth category of the i-th group purchaser is among all the transaction data of the jth category. The larger the value, the more normal the transaction data of the i-th group purchaser is, and the stronger the correlation with other normal transaction data is.

[0053] The transaction impact ratio calculation module is used to obtain the adjacent coefficients of the two group purchasers for each type of transaction data according to the corresponding element values ​​of the two group purchasers in the adjacent matrix corresponding to each type of transaction data and the maximum value of the elements in the adjacent matrix, and to divide each type of transaction data of all group purchasers into a first type of transaction data set and a second type of transaction data set according to the classification algorithm, and to obtain the transaction impact ratio of each type of transaction data according to the first type of transaction data set and the second type of transaction data set.

[0054] For each group purchaser, similar personnel with similar data features are obtained based on the adjacent matrix Y of each type of transaction data. If the transaction data of the i-th group purchaser and the t-th group purchaser have large element values ​​in the corresponding adjacent matrix and have a large correlation under multiple walking step sizes, it means that the t-th group purchaser can be regarded as a similar person to the i-th group purchaser in the corresponding transaction data. Using different walking step sizes, the similarity between different types of transaction data is used to characterize the association between group purchasers, such as two group purchasers are normal group purchases, have similar transaction data features, and the same adjacent transaction data. The higher the similarity between transaction data, the higher the possibility that different group purchasers have the same transaction features.

[0055] Based on the above analysis, the adjacent coefficient T is constructed here to characterize the possibility that two group purchasers in each type of transaction data are of the same type, and the adjacent coefficient between the i-th group purchaser and the t-th group purchaser in the j-th type of transaction data is calculated. :

[0056]

[0057] In the formula, , They are is the value of the element in the i-th row and t-th column and the t-th row and i-th column in the adjacent matrix Y. is the maximum value of the element value in the adjacent matrix Y, K is the maximum walking step length, and the empirical value of K is 7.

[0058] in, The larger the value, the more After a certain number of walks and transaction data The stronger the correlation, the The larger the value, the more After a certain number of walks and transaction data The stronger the correlation, The larger the value, the more transaction data there is in the jth category. , The more they should be classified into the same category.

[0059] Furthermore, the adjacent coefficients between all the j-th category transaction data are calculated, and the segmentation threshold is obtained using the Otsu algorithm. The j-th category transaction data with an adjacent coefficient greater than the threshold is divided into the first category data, and the set of all the first category data is recorded as the first category data set. , the transaction data with adjacent coefficients less than or equal to the threshold are divided into the second category of data, and the set of all the second category of data is recorded as the second category of data set .

[0060] Secondly, all n types of transaction data are traversed, and the adjacent coefficients and the optimal thresholds are obtained according to the same steps, and each type of transaction data is divided into two categories. The Otsu algorithm is a well-known technology, and the specific process is not repeated here.

[0061] Furthermore, among the n types of transaction data collected, each type of transaction data has a different impact on the abnormal transaction data. For example, the impact of transaction time data on the judgment of abnormal transaction data is smaller than that of transaction amount data and group purchase type data, because different group purchasers place orders at different times, and it is difficult to obtain the detection results of abnormal transaction data only from the differences in transaction time data. Therefore, it is considered to use the distribution characteristics of each type of transaction data to calculate the impact of each type of transaction data on the abnormal detection results.

[0062] Based on the above analysis, the transaction impact ratio is constructed here. , which is used to characterize the influence of different types of transaction data on the abnormal transaction data detection results, and calculate the transaction impact ratio of the j-th type of transaction data :

[0063]

[0064] In the formula, is the transaction impact ratio of the j-th type of transaction data, is the inter-class variance when the j-th category of transaction data is divided into the first category of data and the second category of data, is the intra-class variance of the j-th class of transaction data, which is equal to the first class of data sets The variance of the second similar data set The sum of variances.

[0065] Among them, the intra-class variance in the j-th category of transaction data reflects the distribution between each transaction data and the same category of data. The smaller the distribution variance, the smaller the intra-class variance, the more concentrated the distribution of the j-th category of transaction data, and the less likely it is that there will be abnormal transaction data. The larger the value, the greater the influence of the corresponding j-th type of transaction data on the detection of abnormal transaction data, and the more obvious the abnormality of the corresponding transaction data is in the transaction data, while the influence ratio of transaction data with a smaller influence is close to 0. The beneficial effect of the transaction influence ratio is that the data classification results of different degrees of association in each type of transaction data are obtained through the adjacent matrix, and the influence ratio is obtained by using the data distribution in two similar data sets, taking into account the actual distribution of transaction data with different group purchase demands, and avoiding the influence of excessive or small group purchase demands of a few normal group purchase personnel on the overall abnormal transaction data.

[0066] The transaction data classification module is used to obtain the measured distance between the transaction data of group purchase personnel according to the data correlation and transaction impact ratio of each type of transaction data, and obtain the clustering clusters of the transaction data of all group purchase personnel according to the classification algorithm.

[0067] Furthermore, the transaction impact ratio is used to construct the metric distance D used in the k-means algorithm in this application to calculate the metric distance between the transaction data of the i-th group purchaser and the t-th group purchaser. :

[0068]

[0069] In the formula, , are the correlations between the i-th group purchaser and the t-th group purchaser, is the transaction impact ratio of the jth type of transaction data, n is the number of transaction data categories in the group purchase process, , They are the normalized results of the j-th transaction data of the ith group purchaser and the j-th transaction data of the t-th group purchaser in the transaction data matrix A, respectively.

[0070] in, The size of reflects the impact of the difference in the j-th type of transaction data of different group purchase personnel on the detection of abnormal data in the overall transaction data. The size of reflects the degree of association between the transaction data of the ith and tth group purchase personnel. After random walks of different step lengths, the more similar the transaction data associated with the transaction data of the ith and tth group purchase personnel are, the more similar the transaction data of the ith and tth group purchase personnel are, that is, the measurement distance The smaller it is, the more likely that the transaction data of the i-th and t-th group purchase personnel will be clustered into the same cluster during the clustering process.

[0071] Furthermore, the transaction data of m group purchase personnel are taken as samples, the size of the cluster number k is taken as an empirical value of 10, the metric distance D is taken as the metric distance from the sample point to the initial cluster center, the maximum number of iterations is set to 100, and the iteration stopping condition is that the cluster center no longer changes in the previous and subsequent clustering results. The k-means algorithm is a well-known technology, and the specific process is not repeated here.

[0072] The transaction data management module is used to use the transaction data of group purchase personnel corresponding to the cluster center of each cluster as the representative data of the cluster, obtain the inter-cluster similarity of each representative data according to the DTW distance between the representative data of different clusters, filter abnormal transaction data based on the inter-cluster similarity, and classify and store them for management.

[0073] Specifically, after obtaining the clustering results of the transaction data, the transaction data of the group purchase personnel corresponding to the cluster center of each cluster is taken as the representative data, which are recorded as , to , calculate representative data The inter-cluster similarity :

[0074]

[0075] In the formula, , are the representative data corresponding to the 1st and bth clusters respectively. is the DTW distance between two representative data, k is the number of clusters, and the empirical value of k is 10.

[0076] Furthermore, k representative data are traversed to obtain the inter-cluster similarity of each representative data, and they are sorted in ascending order. Next, the Top-N analysis method is used to obtain abnormal transaction data, and the transaction data of the group purchase personnel in the cluster where the first N representative data in the inter-cluster similarity sorting result are located are taken as abnormal transaction data. For example, when the value of N is 1, the transaction data in the cluster corresponding to the minimum inter-cluster similarity value is taken as abnormal transaction data.

[0077] After obtaining the abnormal transaction data in the transaction data matrix A according to the above steps, all transaction information of the group purchase personnel corresponding to the abnormal transaction data is obtained from the data management system of the online shopping platform, and the staff of the group purchase platform cancels the abnormal order corresponding to the abnormal transaction data.

[0078] In order to meet the group purchase quantity required by the group purchase requirements and count the number of abnormal orders, the online shopping platform generates a new order link. First, the abnormal order and the relevant information of the group purchase personnel are sent to the group leader, and the group leader is prohibited from providing subsequent services; and the new order link is sent to the group leader at the same time, and the group leader looks for new group purchase personnel. The data management system of the online shopping platform saves the transaction data of the new group purchase personnel and performs anomaly detection until the transaction data of all group purchase personnel are normal transaction data and the group purchase quantity meets the requirements. After the group purchase is completed, the data management system classifies and stores the transaction data of the entire group purchase process, and stores the normal transaction data in the same database in the data management system, and the abnormal transaction data in the same database in the data management system.

[0079] It should be noted that the above-described embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

[0080] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

Claims

1. An intelligent classification management system for electronic transaction data based on artificial intelligence, characterized in that: The system includes the following modules: The transaction data collection module is used to obtain the transaction data of all group purchasers in the group purchase process and construct a transaction data matrix using the transaction data of all group purchasers; A data association calculation module is used to obtain a preset number of adjacent matrices according to the transaction data matrix and different walking step lengths, obtain the sociability index of each group purchaser for each type of transaction data according to the preset number of adjacent matrices and the transaction data, and obtain the data association of each group purchaser's transaction data according to the difference in the sociability index of each type of transaction data between each group purchaser and the remaining group purchasers; A transaction impact ratio calculation module is used to obtain the adjacent coefficients of the two group purchasers with respect to each type of transaction data according to the corresponding element values ​​of the two group purchasers in the adjacent matrix corresponding to each type of transaction data and the maximum value of the elements in the adjacent matrix, and to divide each type of transaction data of all group purchasers into a first similar transaction data set and a second similar transaction data set according to the classification algorithm, and to obtain the transaction impact ratio of each type of transaction data according to the first similar transaction data set and the second similar transaction data set; The transaction data classification module is used to obtain the measured distance between the transaction data of group purchasers according to the data correlation and transaction impact ratio of each type of transaction data, and obtain the clustering clusters of the transaction data of all group purchasers according to the classification algorithm; The transaction data management module is used to use the transaction data of group purchase personnel corresponding to the cluster center of each cluster as the representative data of the cluster, obtain the inter-cluster similarity of each representative data according to the DTW distance between the representative data of different clusters, filter abnormal transaction data based on the inter-cluster similarity, and perform classified storage management; The method for obtaining the sociability index of each group purchaser for each type of transaction data based on a preset number of adjacent matrices and transaction data is: , is the sociability index of the j-type transaction data of the i-th group purchaser, K is the maximum walking step length, m is the number of people participating in the group purchase, is the element value of the i-th row and t-th column in the adjacent matrix obtained by the random walk algorithm when the walk step length is b. , are the normalized results of the j-th transaction data of the i-th group purchaser and the j-th transaction data of the t-th group purchaser in the transaction data matrix, respectively. It is the preset tuning factor.

2. The electronic transaction data intelligent classification management system based on artificial intelligence according to claim 1 is characterized in that: The method for obtaining a preset number of adjacent matrices according to the transaction data matrix and different walking step lengths is: Each type of transaction data in the transaction data matrix is ​​taken as a node to form a connection graph as input, different walking step sizes are preset, and the k-path random walk algorithm is used to perform random walks on each type of transaction data to obtain the adjacent matrix.

3. The electronic transaction data intelligent classification management system based on artificial intelligence according to claim 1 is characterized in that: The method for obtaining the data correlation of each group purchaser's transaction data according to the difference in the sociability index of each type of transaction data between each group purchaser and the other group purchasers is: Obtain the sociability index of each type of transaction data of each group purchaser respectively, and obtain the mean sociability index of each type of transaction data according to the sociability index of each type of transaction data of all group purchasers; The natural constant is used as the base, and the difference between the sociability index of each category of transaction data of each group purchasing personnel and the mean of the sociability index of each category of transaction data is used as the index calculation result as the class correlation degree of each category of transaction data of each group purchasing personnel. The class correlation degrees of all categories of transaction data of each group purchasing personnel are accumulated and taken as the data correlation degree corresponding to the transaction data of each group purchasing personnel.

4. The electronic transaction data intelligent classification management system based on artificial intelligence according to claim 1 is characterized in that: The method for obtaining the adjacent coefficients of the two group purchasers for each type of transaction data according to the corresponding element values ​​of the two group purchasers in the adjacent matrix corresponding to each type of transaction data and the maximum value of the elements in the adjacent matrix is: , is the adjacent coefficient between the i-th group purchaser and the t-th group purchaser in the j-th category transaction data, , are the element values ​​of the i-th row and t-th column and the t-th row and i-th column in the adjacent matrix obtained by the random walk algorithm when the walk step length is b. is the maximum value of the element in the adjacent matrix, and K is the maximum walking step length.

5. The electronic transaction data intelligent classification management system based on artificial intelligence according to claim 1 is characterized in that: The method of dividing each type of transaction data of all group purchase personnel into a first type of transaction data set and a second type of transaction data set according to the classification algorithm is as follows: Calculate the adjacent coefficients between any two group purchasers in each type of transaction data, and use the Otsu algorithm to obtain the segmentation threshold; The transaction data whose adjacent coefficient is greater than the segmentation threshold is classified as the first category of data, and the set consisting of all the first category of data is recorded as the first similar data set; the transaction data whose adjacent coefficient is less than the segmentation threshold is classified as the second category of data, and the set consisting of all the second category of data is recorded as the second similar data set.

6. The electronic transaction data intelligent classification management system based on artificial intelligence according to claim 1 is characterized in that: The method for obtaining the transaction impact ratio of each type of transaction data according to the first similar transaction data set and the second similar transaction data set is: For each type of transaction data, obtain the inter-class variance when the Otsu algorithm divides each type of transaction data into the first type of data and the second type of data; The ratio of the inter-class variance to the intra-class variance of each class of transaction data is used as the transaction impact ratio of each class of transaction data.

7. The electronic transaction data intelligent classification management system based on artificial intelligence according to claim 6 is characterized in that: The intra-class variance of each type of transaction data is the sum of the distribution variance of a first data set of the same type and the distribution variance of a second data set of the same type for each type of transaction data.

8. The electronic transaction data intelligent classification management system based on artificial intelligence according to claim 1 is characterized in that: The method for obtaining the measured distance between the transaction data of group purchase personnel according to the data association degree and transaction impact ratio of each type of transaction data is: , is the metric distance between the transaction data of the i-th group purchaser and the t-th group purchaser, , are the data correlations of the i-th group purchaser and the t-th group purchaser, is the transaction impact ratio of the jth type of transaction data, n is the number of transaction data categories in the group purchase process, , They are the normalized results of the j-th category transaction data of the ith group purchaser and the t-th group purchaser in the transaction data matrix respectively.

9. The electronic transaction data intelligent classification management system based on artificial intelligence according to claim 1 is characterized in that: The method for screening abnormal transaction data based on inter-cluster similarity and performing classified storage management is: Using the k-means algorithm, all group purchasers are divided into a preset number of clusters based on the measurement between the transaction data of any two group purchasers; The transaction data of group purchase personnel corresponding to the cluster center of each cluster is taken as the representative data, the mean of the DTW distance between each representative data and the representative data of all clusters is taken as the inter-cluster similarity of each representative data, and the transaction data corresponding to the cluster corresponding to the minimum inter-cluster similarity is taken as the abnormal transaction data.

Images