Skip Navigation
Oklahoma State University
Research Week

Home of World Class Research

Hybrid Recommender System to provide suggestions using user ratings and reviews

Hybrid Recommender System to provide suggestions using user ratings and reviews

Name:
Ravi Subramanian

Department:
Business Analytics

Abstract:
If you have ever shopped on Amazon, Pandora or Netflix, you have probably experienced recommendation systems in action. These systems analyze historical buying behavior and make real time recommendations while you are shopping. The back end of these systems contain data mining models that make predictions about the product relevant to you. We plan to build a similar hybrid recommender system to suggest restaurants. We intend to combine content from Yelp reviews, user’s profile, their ratings/reviews for each restaurant visited, restaurant details and tips provided by the user. To implement our idea, we downloaded 2.2M reviews and 591K tips by 552,000 users at the Yelp website. The dataset for 77,000 restaurants contain information such as user profile information. Traditional systems utilize only user’s ratings to recommend new restaurants. However, the system we propose will use both user’s reviews or content and ratings to provide recommendations. The content based system is modeled by identifying the preferences for each user and associating them with key words such as cuisine, inexpensive, cleanliness and so on by constructing concept links and association rules based on their past reviews. The collaborative based system is modeled through k-means clustering by aggregating a particular user with other peer users based on the ratings provided for restaurants.