融入上下文相互关系的概率矩阵分解推荐

Probabilistic matrix factorization recommendation with context correlation

上下文感知推荐系统在传统的推荐算法中加入了上下文信息,从而有效地提高了推荐效果.上下文感知推荐算法将上下文融入推荐生成过程的不同阶段分成三类.大部分算法虽整合了上下文信息,但忽略了上下文之间的相互关系.针对这种情况,提出一种推荐算法.首先在众多的上下文信息中,通过统计方法,提取具有显著不同的上下文特征,从而降低了数据的维度和稀疏度;然后计算上下文信息之间的修正余弦相似度,并与概率矩阵分解模型结合,从而有效地将上下文相互关系融入到了概率矩阵分解中.实验结果表明,该方法可以有效利用上下文的相互关系,提高推荐的准确度.

Context-aware recommender systems that incorporate context into traditional recommendation algorithms can effectively improve recommendation accuracy. Usually sontext-aware recommendation algorithms can be classified into three strategies.Usually,the algorithms integrate contextual information into systems but ignore the correlations of context.To solve this problem,we proposed a new algorithm.First,among the most contextual information,we used a statistical method to extract contextual features with significant difference to reduce the dimensions and the sparsity of data.Then we incorporated the context correlation by calculating the adjusted cosine similarity into probabilistic matrix factorization.Experimental results show that this technique can effectively utilize contextual correlations and improve the recommendation accuracy.