Consumer reviews analysis on E-commerce platforms

• Main Author: Lin, Lixian
• Other Authors: Lihui Chen
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2022
• Subjects: Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence
• Online Access: https://hdl.handle.net/10356/158784
• Institution: Nanyang Technological University

In recent years, online shopping has become more dominant as a marketing channel. Consumers can interact with products in various ways online – view, purchase, return, comment, recommend etc. It enables a complete purchase cycle in a purely digital form, allowing the information to be more accessible and spreadable. To keep up with the market trends and consumer needs, brands view it as an opportunity to better understand consumer behaviors. Customer reviews on e-commerce platforms are the focus of analysis as they are public-accessible, diverse, and enormous. This project targets on the largest consumer market in Asia – China and apply a machine learning method to analyze Chinese reviews from the top e-commerce platform Tmall.com. It aims to understand the sentiment from Chinese text by identifying aspects of interest mentioned in review and clustering into pre-defined groups. In particular, an attention-based aspect extraction model is studied, implemented and tuned to fit a dataset of another language. A Chinese Beauty Corpus containing 200k Chinese reviews from mainstream makeup and skincare products across categories and brands is built to train the model. Language-specific pre-processing methods are studied and applied, following by the usage of word2vec model to generate meaningful embeddings for model to learn. First part of this report focuses on background knowledge where related models and methods are studied. Model understanding and implementation are elaborated in the second part where method and algorithm are explained in detail. The last part shows experiment and result.