Kakodkar, Yogesh. Preference prediction through feature-based collaborative filtering of textual reviews. Retrieved from https://doi.org/doi:10.7282/T33N22R1
DescriptionText reviews are often used by users to decide whether to buy a product or watch a movie or dine in a restaurant. Most of these reviews are raw text and lack a formal structure. Computers cannot easily understand and interpret these reviews to analyze and aggregate them. Users have to manually read through these reviews to find the useful information about the concerned restaurant. We use the topical and sentimental information compiled from raw textual reviews to understand user preferences. We use these preferences to cluster similar users together and then predict users' topical feelings towards the restaurants for which they may be requesting information and to make suitable recommendations. Users have similarities in their preferences for particular topics under which the restaurants have been reviewed. Therefore, we can soft-cluster them using these similarities extracted from their reviewing history. These cluster membership probabilities help us make predictions about the user's sentiments in each topic for the target restaurant. Our results show our accuracy for predicting these sentiments and show that we can provide recommendations to users in most topics for the target restaurant.