Abstract: After the ending of Netflix Prize contest in 2009，the Ensemble learning method for recommendation including Bagging and Stacking attracts much attention because the top teams wined the prize with this kind of algorithms.However，almost nobody cares the Boosting algorithms for personalized recommendation.This paper first analyzes the reason why Boosting framework can be successfully applied in the classification and points out its drawbacks when used in recommendation.By simplifying and decomposing the user-rating matrix，the original recommendation problem is transformed into a simple classification problem.And then，the Boosting algorithms can be applied into recommendation problems.Thus，a new combined algorithm is proposed，called RankBoost*，which boosts the multiple KNN algorithms using several different similarity measures to improve the final predication performance.Experimental results show the effectiveness of Boosting framework to improve the single learning algorithm for personalized recommendation.

Key words: personalized recommendation, boosting, weak hypothesis, collaborative filtering

摘要： 在2009年结束的Netflix推荐大赛中，由于顶级参赛小组均使用集成学习算法，使得基于Bagging和Stacking的Ensemble方法得到了广泛的关注，而基于Boosting的集成学习方法相对来说却无人问津。首先分析了基于Boosting的集成学习算法在分类问题中的优势，以及在推荐问题上的缺陷。通过对用户评分矩阵的简化和分解，将问题转换为简单的分类问题，使得Boosting的集成学习算法能够应用到推荐问题中，提出了基于KNN的集成学习推荐算法，通过集成多个不同的相似度计算方法来提高最终的推荐准确率。在大规模真实数据集上的实验说明，基于Boosting的学习框架可以较大提升单个推荐算法的性能。

关键词: 个性化推荐, 集成学习, 弱学习器, 协同过滤