Digital Enterprise Seminar Series - Professor Debashish Roy
- Date
- February 20, 2019
- Time
- 1:00 PM EST - 2:00 PM EST
- Location
- TRS 2-004
- Open To
- Students, Staff, Faculty of Ryerson
Title: Multiple Source Based Recommender Systems
Abstract
The problem of information overload has become prominent owing to the rapid growth in the amount of available digital information and the increasing number of visitors to the different web resources. Because of the huge volume of information, it is hard for the users to find items of interest in time. To help users find items of their interest, recommender systems are widely used. Depending on a user profile, recommender systems can determine whether a particular user will like an item or not. To build a recommender system, various approaches have been developed, including collaborative filtering, content-based filtering and hybrid filtering. Collaborative filtering (CF) is most commonly used and it recommends items by identifying other users with similar interest; it uses users’ opinion to develop a rating matrix to recommend items to an active user. Content-based (CT) technique recommends items based on content similarity; they are based on a description of the item and a profile of the user’s preferences. Hybrid filtering technique combines the CF and CT based filtering techniques in order to increase the accuracy of the recommender systems. Some examples of items to be recommended are: a movie of interest, a related item in an online shopping environment, point-of-interest (POI) based on check-ins from the friends, music of interest, a favorite tourist spot etc.
Most of the recommender systems are designed based on usage data collected on their own web sites. However, sometimes it could be helpful if we could get information about recommended items from multiple different sources, providing multiple perspectives for users to make their decisions. For example, for movie recommendation, we can build a movie profile by collecting different movie data such as reviews from different movie review websites and from social media data; trailer view history from YouTube, revenue data from box office information etc. So, with the inclusion of multiple data sources we can build a strong content profile. We are working to build a hybrid recommendation algorithm utilizing multiple data sources and we want to make this recommender system generic enough so that it can be easily adapted to different domains of information.