基于Pairwise排序学习的因子分解推荐算法

FACTORISATION RECOMMENDATION ALGORITHM BASED ON PAIRWISE RANKING LEARNING

针对基于内存的协同过滤推荐算法存在推荐列表排序效果不佳的问题,提出基于Pairwise排序学习的因子分解推荐算法(简称Pairwise-SVD推荐算法)。新算法将因子分解的预测结果作为排序学习算法的输入,把排序问题转化成分类问题使用排序学习理论进行排序产生推荐列表。实验结果表明相比基于内存的协同过滤推荐算法,Pairwise-SVD推荐算法的排序效果更佳。其在指标Kendall-tau上提高了近一倍,在指标MRR上提高了近30%,且在指标MAP上也有小幅提高。

Memory-based collaborative filtering recommendation algorithm has the problem of poor recommendation list ranking effect, in light of this, we presented a new kind of factorisation recommendation algorithm which is based on Pairwise ranking learning, namely the Pairwise-SVD recommendation algorithm. The new algorithm uses the predictive result of factorisation as the input of ranking learning algorithm, in this w a y , it converts the ranking problem into a classification problem and then uses the ranking learning theory to rank and generates a recommendation list. Experimental results showed that, compared with the memory-based collaborative filtering recommendation algorithm, the Pairwise-SVD recommendation algorithm had better ranking effect. W h a t ’s more,it almost doubled the index of Kendall-tau,and improved about 3 0 % in index of M R R , and reached a small improvement in the index of M A P as well.