Method for Novelty Recommendation Using Topic Modelling

Content-based filtering methods fall short in situations where there are many similar items to recommend from, for instance when recommending articles from multiple news portals. To deal with this problem, we can consider the novelty of recommendations. Detecting novelty is usually implemented as finding the most dissimilar articles. We propose a method that uses topic modelling to find the novelty of articles. Our method ranks topics by their importance and novelty to the user and recommends articles according to their topics. We evaluate our method and compare it to other approaches to novelty recommendation and also to a method that doesn’t take novelty into account. The results show that our method was more successful than the other approaches to novelty detection in recommending relevant articles that the users were interested in. It also showed a better click-through rate than the method that didn’t incorporate novelty, although the order of its recommendations was less optimal.