Multi-Label Text Classification And Topic Mining Based On Live E-Commerce Return Reviews

In recent years,with the rapid development of the Internet,online shopping has gradually become one of the mainstream consumption methods for users,providing people with more convenient and fast services.With the recent years,short video APP(such as Tiktok,Kwai)through the way live with goods to join the e-commerce team,consumers can more comprehensively through the anchor to explain the way to understand the goods,but such platforms do not yet have a more mature quality assurance system for goods.This requires merchants and platforms in the operation time,not only need to master the operational methods,but also pay attention to the study and treatment of after-sales issues.In order to protect consumer rights and stimulate consumption,merchants provide a seven-day no-reason return service,but need to enter the reason for refund when initiating a refund.Quality refund means that the consumer initiates a refund because of the quality of the goods.Currently,although e-commerce platforms keep statistics on the reasons for quality refunds,they do not study them in depth due to the large volume of text.It is important to study and analyze the reasons for product refunds to urge merchants and platforms to target product problems,improve consumers’ shopping experience.The main work includes the following aspects:(1)We crawl the return reviews data of Kwai’s live e-commerce platform by means of web crawler,and pre-process the obtained data,including data cleaning,mechanical compression to remove words,Chinese word separation,and training word vectors,etc.(2)We implement a multi-label text classification model based on return reviews,including CNN,LSTM,and CNN-LSTM based on classification models.The model parameter settings are explored through experiments,and the experimental results show that the CNN-LSTM model has the highest accuracy rate of 80.9%.Based on the classification results,suggestions are made to the platform.(3)We implement LDA topic model building based on product return reviews,finds the key reasons for high product return rate through wordcloud graph analysis and topic mining,visualizes the topic model,and thus makes suggestions to platforms and merchants.