A bank transaction record classification method

Abstract

This application discloses a method for classifying bank transaction records, which relates to the field of transaction classification technology. The method constructs a regional corpus corresponding to the consumption area by capturing consumption places within the consumption area, then trains a regional word segmentation model for the consumption area, and then determines the customer's corresponding consumption area based on the historical full transaction records of each customer within a predetermined historical period. The regional word segmentation model of the consumption area is used to segment the transaction description text of the customer's bank transaction record, and then inputs it into the transaction classification model to obtain the classification result of the bank transaction record. The method constructs different regional word segmentation models based on the consumption places and bank business nouns in the consumption area, thereby better meeting the merchant information and bank business needs of the consumption area. Compared with the practice of word segmentation based on traditional general corpus, the word segmentation logic is optimized, making the word segmentation results more accurate, thereby making the classification of bank transaction records more accurate.

Description

Technical Field

[0001] The present application relates to the technical field of transaction classification, and in particular to a method for classifying bank transaction records. Background Art

[0002] At present, online transactions have fully penetrated into people's lives, and the number of online transaction records generated has exploded. By classifying these transaction records and analyzing the classified transaction records using technologies such as big data and artificial intelligence, a series of valuable new data can be derived, such as the transaction preferences of terminal merchants or consumers, so that targeted services can be provided to different terminal merchants or consumers.

[0003] The accuracy of transaction record classification directly affects the effectiveness of subsequent data analysis. Existing transaction record classification methods are mainly based on the transaction description information in the transaction records. However, the correlation between many keywords in the transaction description information and common transaction categories is low, which often leads to problems such as difficulty in classification or incorrect classification, resulting in low transaction classification accuracy. Summary of the Invention

[0004] In response to the above problems and technical needs, the applicant has proposed a bank transaction record classification method. The technical solution of this application is as follows:

[0005] A bank transaction record classification method, the method comprising:

[0006] For any consumption area, capture the consumption places within the consumption area to build a regional corpus of the consumption area;

[0007] Based on the regional corpus of consumption areas, the regional word segmentation model of consumption areas is obtained by training the transaction description text of the entire historical transaction records of the consumption areas;

[0008] Determine the customer's corresponding consumption region based on the customer's historical full transaction records within a predetermined historical period, and use the consumption region's regional segmentation model to segment the transaction description text in the customer's bank transaction records;

[0009] The word segmentation results of each customer's bank transaction records are input into a pre-trained transaction classification model to obtain the classification results of the bank transaction records.

[0010] A further technical solution is that the method for determining the consumption area corresponding to each customer includes:

[0011] Obtain the customer's historical full transaction records within a predetermined historical period and mark the transaction locations, cluster the transaction locations of all historical full transaction records, and determine the customer's corresponding consumption area based on the clustering results.

[0012] A further technical solution is that the method for marking the transaction location of the customer's historical full transaction records includes:

[0013] The customer's basic address range is determined based on the customer's personal information, and the transaction locations of the customer's historical full transaction records are marked within the basic address range.

[0014] Its further technical solution is to determine the customer's basic address range based on the customer's personal information, including taking the area within the predetermined coverage area where the customer's residential address, work address and household registration address recorded in the customer's personal information are located as the customer's basic address range.

[0015] A further technical solution is that the method for determining the consumption area corresponding to the customer based on the clustering results includes:

[0016] The transaction locations of all historical transaction records are clustered using the maximum density clustering method to obtain several clusters. The consumption area to which the boundary area of the largest cluster belongs is determined to obtain the consumption area corresponding to the customer.

[0017] Its further technical solution is that the method further comprises:

[0018] Summarize the classification results of the customer's bank transaction records within a predetermined time interval to obtain statistical data corresponding to each transaction category of the customer. The statistical data of each transaction category includes the number of transactions and transaction amounts of the bank transaction records belonging to the transaction category within the predetermined time interval.

[0019] Its further technical solution is that the method further comprises:

[0020] Calculate the ratio of the number of transactions of each transaction category of the customer to the total number of transactions in all bank transaction records within a predetermined time interval to obtain the proportion of transactions of each transaction category;

[0021] Calculate the ratio of the customer's transaction amount for each transaction category to the total transaction amount of all bank transaction records within a predetermined time interval to obtain the transaction amount ratio for each transaction category;

[0022] The preference value of each transaction category is obtained by weighted calculation of the proportion of transaction numbers and transaction amounts in each transaction category of the customer. The higher the preference value, the higher the customer's preference for the transaction category.

[0023] Its further technical solution is that the method further comprises:

[0024] Use the regional word segmentation model for each consumption area to segment the transaction description text of the consumption area's historical full transaction records, obtain the word segmentation results for each historical full transaction record, and label the transaction category to which each historical full transaction record belongs to construct a training sample set;

[0025] The word segmentation results of the entire historical transaction record are used as input and the transaction category is used as output. The training sample set is used to perform classification fitting based on the naive Bayes algorithm to train a transaction classification model.

[0026] The beneficial technical effects of this application are:

[0027] The present application discloses a method for classifying bank transaction records. The method constructs different regional word segmentation models based on the consumption places and banking business nouns in the consumption area, so as to be more consistent with the merchant information and banking business in the consumption area. Compared with the word segmentation method based on the traditional general corpus, the word segmentation logic is optimized, and the word segmentation results are more accurate, thereby making the classification of bank transaction records more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 This is a flow chart of a method for classifying bank transaction records according to an embodiment of the present application. DETAILED DESCRIPTION

[0029] The specific implementation of this application will be further described below with reference to the accompanying drawings.

[0030] This application discloses a bank transaction record classification method, please refer to Figure 1 The method includes the following steps:

[0031] Step 1: For any consumption area, capture the consumption venues within the consumption area to build a regional corpus of the consumption area. The coverage of each consumption area can be customized. For example, a prefecture-level city can be selected as a consumption area. Using the map API, capture the consumption area according to its location category to build a regional corpus.

[0032] Step 2: Based on the regional corpus of each consumption area, the transaction description text of the entire historical transaction records of the consumption area is used to train a regional word segmentation model for the consumption area.

[0033] Different consumption areas contain different consumption venues. The names of some consumption venues are not clearly directional and may be close to common words. When using ordinary word segmentation methods, they are prone to misidentification, resulting in word segmentation errors and ultimately subsequent classification errors. Therefore, it is necessary to build a regional word segmentation model based on the consumption venues within the consumption area. Some consumption venues are regional and may only exist in certain consumption areas. Therefore, different regional word segmentation models are needed for different consumption areas.

[0034] In addition, when training the regional word segmentation model, stop words and learning words are added based on banking business nouns, such as Alipay, Jianghu platform, and liquidated damages, so as to optimize the word segmentation logic based on banking business proper nouns.

[0035] Step 3: Determine the customer's corresponding consumption area based on the customer's historical full transaction records within a predetermined historical period, and use the regional segmentation model of the corresponding consumption area to segment the transaction description text of the customer's bank transaction record.

[0036] As described above, the regional word segmentation models for different consumption areas are constructed based on the information of consumption places within the corresponding consumption areas, and are consistent with the characteristics of the consumption areas. Therefore, when performing word segmentation processing on a customer's bank transaction records, it is necessary to determine which regional word segmentation model of the consumption area to use for word segmentation processing. In one embodiment, determining the consumption area corresponding to the customer includes the following steps:

[0037] (1) Obtain the customer's historical full transaction records within a predetermined historical period and mark the transaction locations. It is usually difficult to accurately mark the transaction locations based solely on the transaction description text of the historical full transaction records. Therefore, the customer's basic address range is determined based on the customer's personal information, including the area within the predetermined coverage range where the customer's residential address, work address, and household registration address recorded in the customer's personal information are located as the customer's basic address range. Then, the transaction locations of the customer's historical full transaction records within the basic address range are marked.

[0038] (2) Clustering the transaction locations of all historical full transaction records. In one embodiment, the transaction locations of all historical full transaction records are clustered using the density maximum clustering method to obtain several clusters.

[0039] (3) Determine the customer's corresponding consumption area based on the clustering results. Even if the span between the residential address and the work address is large, for example, the customer is not in the same prefecture-level city, their transaction locations are generally concentrated. Therefore, the clusters obtained by clustering will generally have a cluster with the largest number of historical transaction records. The consumption area to which the boundary area of the largest cluster belongs is determined to obtain the consumption area corresponding to the customer. The largest cluster obtained by clustering represents the core transaction area where the customer consumes most frequently, and the consumption area to which it belongs is regarded as the consumption area corresponding to the customer.

[0040] Step 4: Input the word segmentation results of each customer's bank transaction record into the pre-trained transaction classification model to obtain the classification results of the bank transaction record.

[0041] The transaction classification model is pre-trained. After training a regional word segmentation model for each consumption area, the regional word segmentation model for each consumption area is used to segment the transaction description text of the corresponding consumption area's full historical transaction records. The segmentation results for each historical full transaction record are then labeled with the transaction category to construct a training sample set. Using the segmentation results of the historical full transaction records as input and the transaction category as output, the training sample set is used to perform classification fitting based on the Naive Bayes algorithm to train the transaction classification model.

[0042] In one embodiment, transaction categories include dining and food, shopping and life, healthcare, transportation and travel, leisure and entertainment, recharge and payment, technology and life, hotel accommodation, platform transactions, financial transactions, and unclassifiable.

[0043] In step 5, after obtaining the classification results of each bank transaction record of each customer, the classification results of the bank transaction records of the same customer within a predetermined time interval can also be summarized to obtain statistical data corresponding to each transaction category of the customer. The statistical data of each transaction category includes the number of transactions and transaction amount of the bank transaction records belonging to the transaction category within the predetermined time interval.

[0044] Step 6: Based on this, the proportion of the number of transactions in each customer transaction category to the total number of transactions in all bank transaction records within the predetermined time interval can be calculated to obtain the transaction number share of each transaction category. Furthermore, the proportion of the transaction amount in each customer transaction category to the total transaction amount in all bank transaction records within the predetermined time interval can be calculated to obtain the transaction amount share of each transaction category. A weighted calculation is then performed on the transaction number share and transaction amount share of each customer transaction category to obtain a preference value for each transaction category. A higher preference value indicates a greater customer preference for that transaction category.

[0045] The above description is only a preferred embodiment of the present application, and the present application is not limited to the above embodiments. It is understood that other improvements and variations directly derived or imagined by those skilled in the art without departing from the spirit and concept of the present application should be considered to be included in the scope of protection of the present application.

Claims

1. A bank transaction record classification method, characterized in that: The method comprises: For any consumption area, capture the consumption places within the consumption area to construct a regional corpus of the consumption area; A regional word segmentation model for the consumption area is obtained by training the transaction description text of the entire historical transaction records of the consumption area based on the regional corpus of the consumption area; Determine the customer's corresponding consumption region based on the customer's historical full transaction records within a predetermined historical period, and perform word segmentation processing on the transaction description text of the customer's bank transaction records using the regional word segmentation model of the consumption region; Inputting the word segmentation results of each customer's bank transaction record into a pre-trained transaction classification model to obtain a classification result for the bank transaction record; The method for determining the consumption area corresponding to each customer includes: obtaining the customer's historical full transaction records within a predetermined historical period and marking the transaction locations, and clustering the transaction locations of all historical full transaction records, and determining the consumption area corresponding to the customer based on the clustering results.

2. The method according to claim 1, characterized in that Methods for marking the transaction locations of a customer's historical full transaction records include: The basic address range of the customer is determined based on the personal information of the customer, and the transaction locations of the customer's historical full transaction records are marked within the basic address range.

3. The method according to claim 2, characterized in that Determining the basic address range of the customer based on the personal information of the customer includes taking the area within a predetermined coverage range where the residential address, work address and registered address recorded in the personal information of the customer are located as the basic address range of the customer.

4. The method according to claim 1, wherein The method for determining the consumption area corresponding to the customer according to the clustering result includes: The transaction locations of all historical transaction records are clustered using the maximum density clustering method to obtain several clusters, and the consumption area to which the boundary area of the largest cluster belongs is determined to obtain the consumption area corresponding to the customer.

5. The method according to claim 1, characterized in that: The method further comprises: Summarize the classification results of the customer's bank transaction records within a predetermined time interval to obtain statistical data corresponding to each transaction category of the customer, where the statistical data of each transaction category includes the number of transactions and transaction amount of the bank transaction records belonging to the transaction category within the predetermined time interval.

6. The method according to claim 5, characterized in that The method further comprises: Calculating the ratio of the number of transactions of each transaction category of the customer to the total number of transactions in all bank transaction records within the predetermined time interval to obtain a proportion of the number of transactions of each transaction category; Calculating the ratio of the transaction amount of each transaction category of the customer to the total transaction amount of all bank transaction records within the predetermined time interval to obtain the transaction amount ratio of each transaction category; A preference value for each transaction category is obtained by performing a weighted calculation on the proportion of the number of transactions and the proportion of the transaction amount of each transaction category of the customer. The higher the preference value, the higher the customer's preference for the transaction category.

7. The method according to claim 1, characterized in that: The method further comprises: Using the regional word segmentation model for each consumption area, the transaction description text of the regional full-volume transaction records of the consumption area is segmented to obtain the word segmentation results of each historical full-volume transaction record. The transaction category to which each historical full-volume transaction record belongs is labeled to construct a training sample set. The word segmentation results of the entire historical transaction record are used as input and the transaction category is used as output. The training sample set is used to perform classification fitting based on the naive Bayes algorithm to train and obtain the transaction classification model.

Images