Improving Recommendation Diversity

Recommender systems offer users a more intelligent and personalised mechanism to seek out new information. Content-based recommender systems generally prefer to retrieve a set of items maximally similar to a users’ query and/or profile. We argue that as new types of recommendation domains and tasks emerge, this blind faith in the similarity assumption begins to seem flawed. We show that very often recommendation diversity is important and that traditional recommendation systems are marred by poor diversity characteristics. We evaluate a new class of diversity-preserving algorithm capable of addressing this without compromising similarity or