E-commerce sales volume fraud identification method based on deep learning model

Abstract

The invention provides an e-commerce sales volume fraud identification method based on a deep learning model, and belongs to the field of data analysis. The method comprises the steps: analyzing the sales volume data change of an e-commerce platform, building a model, judging whether a given shop has a sales volume fraud behavior or not, helping a user to timely identify potential bogus transactions, and improving the user experience. And the fairness and transparency of the market are ensured. In addition, the method not only can be used for e-commerce sales volume fraud detection, but also can be expanded and applied to the field of anomaly detection of other time series data.

Description

Technical Field

[0001] The present invention relates to the fields of data analysis, machine learning and deep learning, and in particular to an e-commerce sales fraud identification method based on a deep learning model. Background Art

[0002] With the rapid development of mobile technology and smart devices, the e-commerce industry has gradually become a key driving force in the retail market. Although e-commerce platforms have driven a significant increase in industry revenue, in an environment driven by sales data, sales fraud has become increasingly prominent. Sales fraud is the use of fictitious transactions to conduct false commercial propaganda in an attempt to deceive and mislead consumers. For example, some e-commerce operators will take advantage of platform rules, fabricate the number of commodity transactions, and increase the "full network sales" data on the platform through methods such as return orders to improve the influence and competitiveness of online stores. This behavior not only misleads consumers, but also undermines the fair competition order of e-commerce operators, and also affects the statistics of e-commerce data. However, traditional fraud detection methods are often unable to effectively deal with complex and hidden fraud methods. A more efficient and intelligent solution is urgently needed. Therefore, the development of effective e-commerce sales fraud detection technology is crucial for the entire e-commerce industry. Summary of the invention

[0003] In order to solve the above technical problems, the present invention provides an e-commerce sales fraud identification method based on a deep learning model, which aims to help consumers identify similar behaviors in a timely manner, and at the same time help companies to more accurately count e-commerce sales, making the data more authentic, and improving the overall integrity of the e-commerce platform, thereby promoting the healthy development of the e-commerce industry.

[0004] The technical solution of the present invention is:

[0005] A method for identifying e-commerce sales fraud based on deep learning models.

[0006] The first is to use crawler technology to obtain sales-related data from e-commerce platforms in real time, including product ID, sales volume (how much sales increase per unit time), user reviews, product prices, return rates (the data displayed on each platform is different and can be calculated based on indicators such as evaluation scores) and other information.

[0007] The second is to obtain sales information of e-commerce stores through channels, and use distributed data processing technology to standardize the collected data, remove outliers, process missing data, and finally store the cleaned data in the database.

[0008] Step 3: Establish a simple fraud monitoring system

[0009] (1) Model training module: Use the processed data to train the LSTM model.

[0010] (2) Real-time monitoring module: The real-time monitoring module is a key part of the entire system. It is responsible for deploying trained models for real-time detection in the daily operation of the e-commerce platform and performing real-time analysis of the sales data of each product.

[0011] (3) Abnormal analysis module: It issues notifications when fraudulent behavior is detected. For major abnormal behaviors, SMS or instant messages can be sent to ensure timely response.

[0012] (4) Intelligent analysis report module: Regularly generate sales analysis reports for the e-commerce platform. If a store is found to have abnormal order-brushing behavior, all sales increase data during the abnormal period will be eliminated, and the sales data during normal time will be analyzed to intelligently obtain the normal sales data of the store.

[0013] Furthermore,

[0014] Input the processed e-commerce sales data into the LSTM model for training; adjust the model's input, output layer, and hidden layer parameters.

[0015] Input Layer

[0016] The sales volume, user reviews, and return rate characteristics of each product are input into the model in time series. The time window of the input layer can be set according to business needs to ensure that the sales volume change characteristics of different periods can be captured;

[0017] Hidden Layer

[0018] It consists of several memory units with memory and forgetting mechanisms, retaining long-term information related to sales fluctuations and discarding useless data;

[0019] Output Layer

[0020] A binary classification output is used to determine whether the current input sales data is abnormal; then, based on the degree of deviation between the calculated input data and the normal sales pattern, a probability value is output, indicating the possibility that the sales of the product are fake.

[0021] Identify whether there are abnormal sales fluctuations of goods within a specific time period. By monitoring the data of the e-commerce platform in real time, detect abnormal phenomena such as sudden increase or frequent changes in sales, and mark them as potential fake orders or false transactions based on preset thresholds.

[0022] Based on the long-term sales data processed and the outliers obtained from the processing, we can eventually identify abnormal sales fluctuations and successfully identify abnormal behaviors such as fake orders.

[0023] The beneficial effects of the present invention are

[0024] The present invention utilizes the sales data disclosed in the e-commerce platform, and the data acquisition cost is low and the reliability is high. By processing multi-dimensional data such as historical sales, user evaluation and estimated return rate of different commodities, and using the deep learning LSTM model for training and prediction, it is possible to effectively identify abnormal fluctuations in sales data, thereby determining whether there is sales fraud in a specific time period. Since the e-commerce platform covers a wide range and has a variety of commodity types, the method of the present invention can ensure the comprehensiveness of the data and the accuracy of the prediction results. At the same time, the present invention realizes the more accurate identification of abnormal transaction behaviors in the e-commerce platform based on low-cost and efficient data analysis than traditional methods, which provides great help for consumers and statisticians to obtain more realistic e-commerce data. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is a schematic diagram of the implementation process of the present invention. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0027] The present invention combines various data from e-commerce platforms and uses deep learning models to detect sales fraud. This method can collect sales data, sales volume per unit time and other information from e-commerce platforms, build a model to identify sales fraud, and help consumers and data statisticians identify fraudulent behaviors such as fake orders.

[0028] The present invention mainly includes three components:

[0029] The first is to use crawler technology to obtain sales-related data from e-commerce platforms in real time, including product ID, sales volume (how much sales increase per unit time), user reviews, product prices, return rates (the data displayed on each platform is different and can be calculated based on indicators such as evaluation scores) and other information.

[0030] The second is to obtain sales information of e-commerce stores through channels, and use distributed data processing technology to standardize the collected data, remove outliers, process missing data, and finally store the cleaned data in the database.

[0031] Step 3: Establish a simple fraud monitoring system

[0032] Model training module: Use the processed data to train the LSTM model.

[0033] Real-time monitoring module: The real-time monitoring module is a key part of the entire system. It is responsible for deploying trained models for real-time detection in the daily operation of the e-commerce platform and performing real-time analysis of the sales data of each product.

[0034] Abnormal analysis module: Issues notifications when fraudulent behavior is detected. For major abnormal behaviors, SMS or instant messages can be sent to ensure timely response.

[0035] Intelligent analysis report module: Regularly generate sales analysis reports for e-commerce platforms. If abnormal order-brushing is found in a store, all sales increase data during the abnormal period will be eliminated, and then the sales data during normal time will be analyzed to intelligently obtain the normal sales data of the store.

[0036] Implement process work processing such as Figure 1 As shown:

[0037] (1) Data collection:

[0038] The sales information collected regularly is recorded in the database. Some product data are as follows:

[0039] (2) System identification and prediction

[0040] Input the processed e-commerce sales data into the LSTM model for training. Adjust the model's input, output layer, and hidden layer parameters

[0041] a) Input layer

[0042] The sales volume, user evaluation, return rate and other characteristics of each product are input into the model in time series. The time window of the input layer can be set to daily, hourly, etc. according to business needs to ensure that the sales volume change characteristics in different time periods can be captured.

[0043] b) Hidden Layer

[0044] It is composed of multiple memory units and has a memory and forgetting mechanism. It retains long-term information related to sales fluctuations, discards useless data, and effectively processes e-commerce sales-related data.

[0045] c) Output layer

[0046] The binary classification output is used to determine whether the current input sales data is abnormal. Then, according to the deviation degree between the calculated input data and the normal sales pattern, a probability value is output, indicating the possibility of fake sales of the product.

[0047] d) Anomaly identification and prediction

[0048] The trained LSTM model can identify whether there are abnormal sales fluctuations in a specific time period. The model monitors the data of the e-commerce platform in real time, detects abnormal phenomena such as sudden increases and frequent changes in sales, and marks them as potential fake orders or false transactions based on the preset threshold.

[0049] e) System Optimization

[0050] Further optimize system performance to ensure more accurate forecast results in different time periods and product types.

[0051] Finally, based on the effectively processed long-term sales data and the outliers obtained through processing, abnormal sales fluctuations can be identified, and abnormal behaviors such as fake orders can be successfully identified.

[0052] The above description is only a preferred embodiment of the present invention, which is only used to illustrate the technical solution of the present invention, and is not used to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention are included in the protection scope of the present invention.

Claims

1. A method for identifying e-commerce sales fraud based on a deep learning model, characterized in that: By analyzing the changes in sales data on e-commerce platforms, a model is established to determine whether a given store has sales fraud, helping users to identify potential false transactions in a timely manner.

2. The method according to claim 1, characterized in that First, use crawler technology to obtain sales-related data from e-commerce platforms in real time; Second, the sales information of e-commerce stores obtained through channels is standardized using distributed data processing technology to remove outliers, process missing data, and finally store the cleaned data in the database; Third, establish a simple fraud monitoring system.

3. The method according to claim 2, characterized in that The data includes product ID, sales volume, user reviews, product price, and return rate information.

4. The method according to claim 2, characterized in that: The fraud monitoring system includes: Model training module: Use the processed data to train the LSTM model; Real-time monitoring module: The real-time monitoring module is a key part of the entire system. It is responsible for real-time detection of the deployed trained models in the daily operation of the e-commerce platform and real-time analysis of the sales data of each product. Abnormal analysis module: issues notifications when fraudulent behavior is detected, and sends SMS or instant messages for abnormal behavior; Intelligent analysis report module: Regularly generate sales analysis reports for the e-commerce platform. If abnormal behavior is found in a store, all sales increase data during the abnormal period will be eliminated, and then the sales data during normal time will be analyzed to intelligently obtain the normal sales data of the store.

5. The method according to claim 4, characterized in that Input the processed e-commerce sales data into the LSTM model for training; adjust the model's input, output layer, and hidden layer parameters.

6. The method according to claim 5, characterized in that a. Input layer The sales volume, user evaluation, and return rate characteristics of each product are input into the model in time series. The time window of the input layer can be set according to business needs to ensure that the sales volume change characteristics of different periods can be captured; b. Hidden Layer It consists of several memory units with memory and forgetting mechanisms, retaining long-term information related to sales fluctuations and discarding useless data; c. Output layer Use binary classification output to determine whether the current input sales data is abnormal; Then, based on the degree of deviation between the calculated input data and the normal sales pattern, a probability value is output, indicating the possibility that the sales volume of the product is inflated.

7. The method according to claim 6, characterized in that Identify whether there are abnormal sales fluctuations of goods within a specific time period, detect abnormal sales through real-time monitoring of e-commerce platform data, and mark them as potential fake orders or false transactions based on preset thresholds.

8. The method according to claim 7, characterized in that Based on the long-term sales data processed and the outliers obtained from the processing, abnormal sales fluctuations were finally identified and abnormal behavior was successfully identified.

Images