Anand, Konjengbam and Singh, Manish
(2020)
Exploration of Product Reviews.
PhD thesis, Indian Institute of Technology Hyderabad.
Abstract
E-commerce is a popular platform for trade of products and services through the Internet. Product reviews play a vital role in e-commerce by influencing the purchasing decision of customers. As many products have thousands of reviews, it is difficult to explore and extract useful information from them. We need review exploration systems to help users quickly explore and comprehend huge volumes of reviews. The current review exploration systems lack effective means of finding relevant review and summarizing them. In this thesis, we present three limitations of existing review exploration systems and propose solutions to address each of the three limitations. Existing aspect based review summarization systems do not show the semantic relations that exist between aspects, which is required for proper exploration of reviews. We address this limitation by showing users an aspect ontology tree, which is created in an unsupervised manner from reviews, to show the relationship between aspects and sub-aspects. We then allow users to navigate reviews according to the aspects of the ontology. Many of the existing review exploration systems summarize reviews by giving an average star rating of the reviews, or by finding the number of positive and negative reviews for each of the aspects. Such summarization methods do not show the actual opinion words of users, which is crucial to understand what other users like or don’t like. We address the problem of summarizing reviews by creating informative and readable tags. We present a novel unsupervised method to generate the top-k opinionated tags. We also address the problem of tag generation for cold products, which have only a limited number of reviews and that too, with very limited content. Finally, we study the problem of stance detection in comparative reviews using word-embeddings. Online debate sites are popular platforms for users to express and form opinions. Comparative reviews are very popular as they give a comparison of different aspects of two competing products. Standard aspect-based summarization approaches cannot be used for comparative reviews as we need to figure out the target preference of each of the aspects, which is often not explicitly available in the review text. We propose an unsupervised approach to summarize comparative reviews by detecting the stance of users from comparative reviews.
Actions (login required)
 |
View Item |
Altmetric
Altmetric