A data processing method for targeted analysis and prediction of e-commerce platforms

Abstract

This invention relates to the field of e-commerce technology, specifically a data processing method for targeted analysis and prediction of e-commerce platforms. The method first automatically captures real-time data during the live stream, then trains a convolutional neural network algorithm on the real-time data to obtain clustered data with relevant features. The clustered data undergoes data processing, eliminating outliers while simultaneously performing custom extraction and formatted data aggregation. Finally, a big data analysis model is built on the formatted data to predict and evaluate the direction of the live stream. This invention solves the problems of low efficiency and overly simplistic manual user tagging in traditional live streams, which relies on convolutional neural networks to capture dynamic user features and capture data.

Description

Technical Field

[0001] This invention relates to the field of e-commerce technology, with IPC classification number G06Q10 / 04, specifically a data processing method for targeted analysis and prediction of e-commerce platforms. Background Technology

[0002] Currently, online live streaming platforms have gradually become a new product promotion model applied in people's lives. However, online live streaming platforms have the problem of indirect communication between users and merchants, making it difficult to understand each other's real needs and preferences. As a result, merchants cannot promote their products more effectively and improve the efficiency of live streaming sales, while users who buy products are not able to get the products they want.

[0003] Patent CN201410601571 provides an information interaction method and transaction system for e-commerce platforms. By setting up a horizontal e-commerce platform and a vertical e-commerce platform, it enables interactive operations for order placement and payment, thereby improving the user's payment efficiency and user experience. However, the transaction system described in this patent is an alternative payment platform to a third-party transaction platform, and is itself a form of third-party payment platform. Therefore, it fundamentally changes the user's payment mode and does not involve targeted processing and optimization of live broadcast information.

[0004] Patent CN202110354790 provides a data processing method and information service platform in a big data business scenario. This patent statistically collects and analyzes live streaming data from multiple e-commerce platforms, trains the data on various data types to obtain systematic data fragments, and then summarizes and analyzes these data fragments to obtain targeted live streaming feature values ​​for e-commerce platforms. However, this patent uses a training mode under big data, and the training results obtained are general and universal, unable to better process and target data for specific problems in a single product or live stream.

[0005] Therefore, in view of the problems existing in the analysis methods of e-commerce platforms, there is an urgent need to introduce a data processing method for targeted analysis and prediction of e-commerce platforms, so as to mine the characteristics and preferences of logged-in users in real time during the live broadcast, thereby enabling more targeted promotion and prediction of the live broadcast. Summary of the Invention

[0006] To address the aforementioned problems, this invention provides a data processing method for targeted analysis and prediction on e-commerce platforms, such as... Figure 2 As shown, the data processing method specifically includes:

[0007] S1. First, live data is collected by automatically capturing real-time data during the live broadcast. A convolutional neural network algorithm is then established based on the real-time data to train the real-time data and obtain clustered data with relevant features.

[0008] S2. Process and extract data from the clustered data, remove outliers from the clustered data, and perform custom extraction and formatted data summarization on the data.

[0009] S3. Establish a big data analysis model for formatted data to predict and evaluate the direction of live streaming.

[0010] Preferably, the data collection specifically includes live stream profile data collection, live stream data collection, and live stream audience profile collection.

[0011] Specifically, the live stream profile data collection includes product profile data and streamer tag data; wherein the product profile data is used to classify and display the characteristics of the products in the current live stream; and the streamer tag data is used to analyze the streamer's personal characteristics based on historical streamer data for targeted analysis.

[0012] Preferably, the live streaming data collection specifically includes comment interaction data collection and monitoring indicator data collection; wherein the comment interaction data collection includes data required for real-time comment maintenance and data required for real-time comment characteristics.

[0013] Specifically, such as Figure 1 As shown, the live streaming data collection also includes the collection of live streaming monitoring metrics data, which specifically includes GPM data monitoring and analysis (transactions per thousand live stream views = transaction amount / number of user visits * 1000), UV value data monitoring and analysis (average transaction amount generated per unique visitor = transaction amount / number of unique visitors), ROI data monitoring and analysis (transaction amount generated by traffic promotion / traffic promotion cost), and historical and current fan count monitoring and analysis.

[0014] Preferably, the comment interaction data collection involves filtering the comment interaction data during the live broadcast interaction process. The specific filtering method is as follows:

[0015] A1. Establish a dual-stage comment interaction data filtering database. First, filter based on sensitive keywords, and then filter based on characteristic keywords.

[0016] A2. Match and train the comment interaction data during the live interaction with the keywords in the filter database. Use sensitive keywords as feature values ​​and the user IDs corresponding to the sensitive keywords as target values. Extract the target values ​​obtained from the training and then block them. After blocking, they will not appear in the live interaction.

[0017] A3. Based on A2, filter the feature keywords in the filtering database, use an unsupervised learning-based clustering algorithm to summarize the same features, and use them for subsequent data processing.

[0018] Preferably, the dual-review interaction data filtering database stores evaluation keywords with product tag information, establishes a data list of evaluation keywords for a single product under a single user ID, and summarizes the data list into a data table. By filtering the characteristic keywords, a user demand analysis model is established for a single product.

[0019] Preferably, the live stream audience profile collection is achieved by establishing a dynamic human feature capture algorithm based on convolutional neural networks to automatically capture and classify the human tags corresponding to the users participating in the live stream, thereby automatically segmenting the targeted audience.

[0020] Specifically, the collection of live stream audience profiles includes audience purchase preference data, content preference data, age, gender, and region data, life stage data, consumer group segmentation data, and active time period data, wherein the collection of live stream audience profiles includes the content described in this invention.

[0021] Preferably, the big data analysis model specifically includes outputting implementation plans and schemes, predicting the results of the implementation plans, and regressing and correcting the actual results.

[0022] Preferably, the output implementation plan and scheme specifically includes a targeted audience delivery plan, outputting suitable product schemes based on the live broadcast room and different anchor tags, predicting anchor tags based on live broadcast room and product data, and providing quantitative and qualitative overviews of comments and interactions for the live broadcast room, anchor, and products.

[0023] Specifically, the targeted audience targeting plan can target specific live streams and product content according to the targeting method, speed, time, time period, target audience, network environment, and platform. Simultaneously, the targeted audience targeting plan uses the live stream optimization target, daily budget (in yuan), and bid as control weighting factors for targeted targeting, performing data constraint processing. The bid is the price willing to pay for each conversion for the optimization target; the price level affects conversion efficiency.

[0024] Preferably, the targeted audience targeting plan can implement a single behavior analysis mode for targeted targeting, or a combination of three behavior analysis modes for targeted targeting, based on the characteristics of the promoted product and by setting specified weight indicators.

[0025] Preferably, the actual results are then re-corrected by using probabilistic analysis methods to correct the targeted delivery strategy in the big data model in real time, thereby improving the accuracy of live data prediction for the next moment.

[0026] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0027] (1) This invention establishes a dynamic character feature capture algorithm based on convolutional neural networks, and automatically captures and classifies user character tags through deep learning, thereby finding targeted audiences. This solves the problems of low efficiency in the manual creation of traditional live streaming user character tags and the fact that manual classification is too one-sided and cannot specifically grasp the actual information of users watching the live stream.

[0028] (2) Based on (1), this invention establishes a dual-comment interaction data filtering database. First, it monitors and blocks malicious comments during the live broadcast to better overcome the problem of malicious comment spamming in traditional live broadcast rooms and ensure good interaction in the live broadcast room. At the same time, it extracts and classifies keywords in users' real-time comments to quickly grasp users' viewing experience of the live broadcast, the host and the products, quickly understand users' needs, and establish a user needs analysis model on this basis to improve merchants' understanding of users' psychology.

[0029] (3) Based on (1) and (2), this invention determines the differentiated needs of the target sales audience based on the data collected during the live broadcast, designs a targeted delivery strategy and behavior analysis model, wherein the behavior analysis model is divided in detail to assess user preferences and needs, and a clustering algorithm is used to cluster the user behavior analysis model, and precise delivery is carried out based on the clustering results.

[0030] (4) Based on (3), this invention implements a single behavior analysis mode targeted delivery strategy based on behavior analysis mode, and sets specified weight indicators based on the characteristics of the promoted products to establish a three-behavior analysis mode combined targeted delivery strategy, thereby dynamically optimizing and correcting the delivery data in the targeted delivery process, avoiding the problem of insignificant delivery effect caused by deviation of the delivery strategy due to external factors during the delivery process.

[0031] (5) Based on the data captured from real-time comment interaction information and operational information, establish a big data analysis model to predict and evaluate corresponding product sales plans, product improvement directions, and live streaming directions. Attached Figure Description

[0032] Figure 1 A flowchart of a data processing method for targeted analysis and prediction of e-commerce platforms;

[0033] Figure 2 This is a module diagram of a data processing method for targeted analysis and prediction of e-commerce platforms. Detailed Implementation

[0034] Example 1:

[0035] This embodiment describes a data processing method for targeted analysis and prediction on e-commerce platforms. The product profile data collected in the live-streaming room specifically includes: product brand; product category (levels 1-3); product price range; product material (cotton, linen, wool); product color scheme; product functional category (sunscreen and anti-wrinkle functions); product seasonal characteristics; product silhouette characteristics (shape, fit, and length); product design features (lace, bows, ribbons, embroidery); product style characteristics; and product tagging characteristics (celebrity-endorsed items and trending products).

[0036] The aforementioned streamer tag data includes the degree of fit between the streamer and product attributes, the degree of fit between the streamer and consumer attributes, and the streamer's personal characteristics. The streamer's personal characteristics include the streamer's personal style, age, personality traits, and historical sales categories and sales volume characteristics.

[0037] The data collected during the live stream, specifically the data required for real-time comment maintenance, includes malicious comments, repeated spam by paid commenters, repeated comment control by paid commenters, and inappropriate comments.

[0038] The aforementioned livestream audience profile collection specifically includes: purchase preferences (including women's clothing, children's clothing, parent-child clothing, snacks and nuts, beauty and skincare, and makeup); content preferences (including casual videos, variety shows, fashion, and parent-child content); life stage data (including single, dating, pregnant, and parent-child); and consumer group segmentation data (including young people in small towns, those under 35 years old in fourth-tier cities and below, middle-aged and elderly people in small towns, those over 35 years old in fourth-tier cities and below, Generation Z, young people under 24 years old in third-tier cities and above, sophisticated mothers, white-collar women aged 25-35 in third-tier cities and above who are preparing for pregnancy or have already given birth, emerging white-collar workers, white-collar workers, IT and finance professionals aged 25-35 in third-tier cities and above, experienced middle class, white-collar workers, IT and finance professionals aged 36-50 in third-tier cities and above, urban blue-collar workers, those with lower to middle spending power aged 25-35 in third-tier cities and above, and urban senior citizens, those over 50 years old in third-tier cities and above).

[0039] The targeted audience targeting plan includes the following methods: cost-controlled targeting, prioritizing cost control to achieve targets and maximizing advertising budget utilization; volume-based targeting, where costs may fluctuate based on prioritizing budget targets; targeting speed, including rapid deployment and concentrated budget allocation if suitable traffic is available; and even deployment; targeting duration, including long-term deployment, deployment with specified start and end dates, and fixed-duration deployment; targeting time periods, including unlimited deployment and deployment within specified time periods; targeted audience targeting, based on different regions, genders, ages, and user behavioral and interest intentions, where behavioral intentions include matching user reading, searching, and viewing behaviors; interest intentions include matching potential user interests; targeted network environments, including 2G, 3G, 4G wireless and wired environments; and targeted platforms, including web and client applications, with client applications including iOS and installed applications.

[0040] The targeted audience targeting plan includes targeted targeting of livestream influencers, where livestream influencers specifically refer to a particular account or category within the livestreaming field, and then matching their followers with people who have similar behaviors or interactions. The livestreaming platform's selected areas include targeted promotional directions and targeted exclusion directions. Targeted promotional directions include promotion-sensitive audiences, audiences with preferences for similar product categories, and general audiences. Targeted exclusion audiences include audiences with high order cancellation and return rates, audiences with high follower counts, highly active audiences, and livestreaming account followers.

[0041] In addition, the targeted audience delivery plan also includes creative categorization, which selects the appropriate creative category based on the ad content to obtain more suitable traffic; and creative tags, which manually add appropriate tags to the ad content to help obtain more suitable traffic.

[0042] By using data processing methods that target and predict e-commerce platforms, a big data analysis model was established. This model was then used to predict and evaluate product sales plans, product improvement directions, and livestreaming strategies. The evaluation results were applied to livestream optimization. The optimized livestream evaluation data was then statistically analyzed, resulting in the data shown in Table 1. This evaluation data includes data on viewers entering the livestream room, product clicks, orders placed, transactions, fan growth, and comments.

[0043] The improved e-commerce platform's various evaluation data showed month-on-month growth as shown in Table 1. Specifically, the various evaluation data of the e-commerce platform were calculated by changing the ratio of each indicator over a two-month consecutive statistical period.

[0044] Table 1

[0045] Evaluation data April June August GMV 2.46％ 28.72％ 12.88％ New fans -28.57％ 82.22％ -9.76％ Total orders -17.02％ 40.38％ 8.68％ Total number of transactions -26.28％ 70.30％ 9.88％ Average order value -24.41％ 38.98％ 2.73％ unit price -8.31％ 23.48％ 3.87％ Total viewers -32.20％ 63.75％ 1.53％ Total viewers -28.97％ -0.82％ 60.53％

Claims

1. A data processing method for targeted analysis and prediction of e-commerce platforms, characterized in that, The data processing method specifically includes: S1. First, live data is collected by automatically capturing real-time data during the live broadcast. A convolutional neural network algorithm is then established based on the real-time data to train the real-time data and obtain clustered data with relevant features. S2. Perform data processing on the clustered data, and while eliminating outliers in the clustered data through data processing, perform custom extraction and formatted data summarization on the data. S3. Build a big data analysis model for formatted data to predict and evaluate the direction of live streaming; The data collection specifically includes live stream profile data collection, live stream data collection, and live stream audience profile collection. The aforementioned live stream audience profiling collection method establishes a dynamic human feature extraction algorithm based on convolutional neural networks to automatically extract and classify the human tags corresponding to users participating in the live stream, thereby automatically segmenting the targeted audience. The big data analysis model specifically includes outputting implementation plans and schemes, predicting the results of the implementation plans, and revising the actual results. The output implementation plan and scheme include, in particular, a targeted audience delivery plan, a product scheme adapted to the live room and different anchor tags, an anchor tag prediction based on live room and product data, and a quantitative and qualitative overview of comments and interactions for the live room, anchor, and products. The aforementioned targeted audience targeting plan utilizes behavioral analysis modes. It can employ a single behavioral analysis mode for targeted targeting strategies, or, based on the characteristics of the promoted product, set specified weight indicators to implement a combination of three behavioral analysis modes for targeted targeting strategies.

2. The data processing method for targeted analysis and prediction of e-commerce platforms according to claim 1, characterized in that, The aforementioned live streaming data collection specifically includes comment interaction data collection and monitoring indicator data collection; wherein the comment interaction data collection includes data required for real-time comment maintenance and data required for real-time comment characteristics.

3. The data processing method for targeted analysis and prediction of e-commerce platforms according to claim 2, characterized in that, The aforementioned comment interaction data collection involves filtering the comment interaction data during the live broadcast. The specific filtering method is as follows: A1. Establish a dual-stage comment interaction data filtering database. First, filter based on sensitive keywords, and then filter based on characteristic keywords. A2. Match and train the comment interaction data during the live interaction with the keywords in the filter database. Use sensitive keywords as feature values ​​and the user IDs corresponding to the sensitive keywords as target values. Extract the target values ​​obtained from the training and then block them. After blocking, they will not appear in the live interaction. A3. Based on A2, filter the feature keywords in the filtering database, use an unsupervised learning-based clustering algorithm to summarize the same features, and use them for subsequent data processing.

4. The data processing method for targeted analysis and prediction of e-commerce platforms according to claim 3, characterized in that, The dual-review interaction data filtering database stores evaluation keywords with product tag information. A data list is created for evaluation keywords proposed for a single product under a single user ID, and the data list is summarized into a data table. By filtering the characteristic keywords, a user demand analysis model is established for a single product.

5. The data processing method for targeted analysis and prediction of e-commerce platforms according to claim 1, characterized in that, The actual results are then regressed and corrected. By using probabilistic analysis methods to correct the targeted delivery strategy in the big data model in real time, the accuracy of live data prediction for the next moment is improved.

Images