Key points are not available for this paper at this time.
With the maturation of e-commerce platform, online shopping has become an easy and preferable mode of shopping. As one of the largest e-commerce platforms worldwide, Amazon enjoy numerous user communities. Volumes of user-generated data of users' preferences and opinions towards products, usually for specific aspects of a commodity, popped up every day. Although loaded with information, these texts are often unstructured data that requires a thorough analysis for both consumers and manufactures to extract meaningful and relevant information. Traditional lexicon-based sentiment analysis considers polarity score of words but ignores the differences among aspects. Document level topic modeling help overcome these lacunae. In this paper, we claim that the aspects should also be weighted for highlighting significance of various aspects appropriate to a domain. Thus, manufacturers can understand what potential consumers may want as improvement in the forthcoming products. To showcase our framework, more than 400,000 Amazon unlocked phone reviews were collected as training data. LDA models were used to cluster topic words with their corresponding probability values. Based on the machine learning framework results, a corpus of nearly 1,000 Amazon reviews of a new mobile phone mode, iPhone X, was tested using this framework to perform topic labeling and sentiment analysis. Performance analysis was done using Confuse Matrix and F-measure.
Yiran et al. (Fri,) studied this question.