Using User Historical Behavior to Predict Aggregate Ratings: Analysis and Model

Open Access
Alrizah, Mshabab
Graduate Program:
Computer Science and Engineering
Master of Science
Document Type:
Master Thesis
Date of Defense:
December 11, 2014
Committee Members:
  • Wang Chien Lee, Thesis Advisor
  • Aggregate Ratings
  • Prediction
  • Regression Analysis
  • E-commerce Products
  • Products Categories
Creditable information conveyed by aggregate ratings are valuable for e-commerce transactions. The considerable and fast expansion of Internet usage boosts the importance of aggregate ratings. However, acquisition of a sufficient number of reliable user ratings becomes critical requirement in building ranking service. Spamicity and inaccuracy are the primary issues causing unfair ratings, which impacts the precision of aggregate ratings. In order to find solution and eliminate unfair ratings, we perform an analysis to better understand the behavior of users and rating processes in rating systems. The regression coefficient, variance and heteroscedasticity of rating stream in the considered dataset are three investigated angles. Moreover, the user rating behaviors among products categories in e-commerce websites are analyzed as well. Many interesting observations that provide a picture of user behaviors in rating systems are provided. These findings provide a guidance to diminish the effects of unfair rating. Moreover, we propose a novel algorithm that leverages the products categories to predict aggregate ratings. The algorithm has shown statistically significant results in predicting aggregate ratings. This study shows an evidence of the benefit in using products categories.