可提高多样性的基于重排序图书推荐算法研究

Improving Aggregate Recommendation Diversity of Books Using Ranking-Based Techniques

通过提供个性化推荐,推荐系统的重要性越来越受到读者的重视。现有推荐算法着重关注推荐的准确度,将读者引导到少数热点图书上,导致产生较多长尾图书的问题;并且读者的兴趣过于集中,不利于挖掘读者潜在的兴趣点。提出一种重排序的基于用户协同过滤算法,该算法通过对推荐列表TOP-N进行重排序来产生推荐列表。实验结果表明,该算法可以在一定精确度损失的条件下,大幅提高最终推荐列表的多样性有利于读者接触更多的未知领域及长尾图书的销售。

The importance of Recommender systems is becoming more and more to readers by providing personalized recommendations.The existing recommendation algorithms that focus on recommendation accuracy will misguide readers to a few hot books,thus creating many long-tail books.As a result,the excessive concentration of reader interest is unfavorable for excavation of potential points of interest.The paper proposed a reranking user-based collaborative filtering algorithm,which generated a new recommendation list via reranking of TOP-N on the original list.The experimental results showed that this algorithm could greatly improve the diversity of the final recommendation list at the sacrifice of certain accuracy.This algorithm helped readers to know more previously unknown fields as well as the borrowing of long-tail books.