中文摘要：

在实际生活中,许多产品具有时效性,人们对于新产品和旧产品的选择通常会基于不同的理由.因此,在推荐中应该考虑这种差异.然而,目前的推荐算法中并没有考虑这种差异性.文中在分析产品时效性的基础上,提出一种时效性感知的个性化推荐方法,它采用了联合矩阵分解的算法,将产品按时效性划分为多个矩阵,再将其联合训练.这样既考虑到了时效性,又克服了产品的稀疏性,并且联合训练又可以得到产品的某种特征向量,挖掘划分成多个部分的时效性产品之间的内部联系.并将流行度作为正则化项.实验表明,该方法可以得到具备良好的推荐性能.

英文摘要：

Timeliness is a very important factor for many products in market.Generally,people choice new products and outofdated products based on different reasons.Therefore,these differences should be considered in recommendations.Unfortunately,most current recommendation approaches don′t take this factor into consideration.Based on the analysis of product timeliness,a timelinesssensitive recommendation approach is proposed.It is based on the collective matrix factorization algorithm,which divides the products into several matrices,sharing parameters among features when an entity involves in multiple relations.Then,using the gradient descent algorithm to train them together and measures the error between the real values and predictions.This approach not only considers the inner relations between products with different timeliness,but also overcomes the problem of sparse in other recommendation algorithms.Simultaneously,it can find the implicit relations between the divided product sets through discovering characteristic vectors in a more comprehensive way.The approach uses the characteristic of products as regularization,which can cope with over-fitting problem which arises during model training on the sample data.Moreover,the approach also uses popularity as another regularization in the model,which helps make the result to be better and more accurate.The experimental results show that this approach achieves better performances against other ones.Therefore,the timeliness-sensitive recommendation approach achieves higher prediction accuracy than factoring each matrix separately and demonstrates its superiority,as well as the benefit of considering the timeliness of products.