Recommending News Based on Hybrid User Profile, Popularity, Trends, and Location

Reading the news is a favorite hobby for many people anywhere in the world. With the popularity of the Internet and social media, users are constantly provided, or even bombarded, with the latest news around the world. With numerous sources of news, it has become a real challenge for users to follow the news that they are interested. Previous work used user profile to recommend personalized news; and used RSS feeds and latest tweets to provide popular, trendy news. In this work we combine these two methods with three enhancements. First, to personalize news recommendation we used a hybrid approach, which involved the analyses of click through, user tweets, and user Twitter friends list to build user profile, this method significantly improves the accuracy of user profile. Second, to address the importance of temporal dynamics, we add a unique new feature of location preference to the news recommendation system. Third, we allow users to choose the ratio of popular news vs. trendy news they desire. The resulting system is then evaluated based on user satisfaction and accuracy. The results show that the average user satisfaction increases from 8.6 to 9.4 when location preference is added, while the accuracy of the recommendation system is around 92-95%. We believe that the proposed system is a successful example of incorporating temporal dynamics to recommendation systems; the combination of using hybrid user profile, popularity, trends and location would have significant impact on other recommendation systems in the future.