Group Members

Jing Zhou, Linwei Chen, Zhe Xu, Kerem Kazan


With a bunch of chaotic information that promised a lot of potential, we wanted to analyze the patterns and algorithmically increase the reliability of the reviews. Then we essentially would be able to provide business owners with strategic plans. Based on these goals, our application mainly implemented three features:
  • Customer relationship information: We analyzed the customers of businesses by using the review data: How one customers are related to one another. We then analyzed which businesses attract independent individuals and which businesses attract groups of friends. Using this analysis, we provided businesses with useful information about how to decorate or how to campaign.
  • Influential customers: We marked those customers with high number of followers. Business owners can treat these customers in specific ways to have favorable customer relationships.
  • Untrollification: The purpose was to detect the ratings given by trolls who act in certain patterns. One patters was giving 1 star to every business. Another was giving suspiciously low rates to all but giving 5 stars to a single business. We also marked those customers who failed to give sufficient number of ratings (could suggest bots or not core users). The above kinds of behaviors could suggest unserious users, bots, businesses buying positive ratings, aggressively giving low ratings due to competition and etc. We weighed them down and eliminated non-core users to provide companies with untrollified ratings.