Diversified Recommendation Based On Personalized Set Ranking

In recent years, personalized recommender systems have been widely used in many fields of Internet as an important tool solving the problem of information overload. For the personalized top-N recommendation task,apart from relevance of recommendation results, diversity is also very necessary. In general, users may have various individual ranges of interests.Users who have board interests may like all kinds of topics, while users with focused interests may be only interested in a few topics. Thus the combination of relevance and diversity in personalized top-N recommendation results would be desirable.For this purpose, this paper integrates the concept of set diversity into traditional matrix factorization model which is a state of art model in collaborative filtering. By using a set-oriented pairwise personalized collaborative ranking method to optimize this model, it could achieve personalized diversified top-N recommendation directly. Category information is also utilized explicitly in learning personalized diversity. Experimental results show that for the task of personalized top-N recommendation, our model outperforms traditional models in terms of both relevance and diversity and achieves better personalized diversity.