引入用户情感偏好矩阵的ItemCF算法研究

传统基于物品的协同过滤算法（ItemCF）利用物品间的相似性为用户提供个性化推荐，然而该方法仅考虑了评分数据，而未关注用户情感偏好。基于从评论中的emoji表情提炼出的用户情感偏好，提出一种引入用户情感编号矩阵的ItemCF算法。该方法利用物品共现矩阵与用户情感偏好矩阵计算物品相似度，进而进行个性化推荐。根据某在线互联网教育实证数据集的实验结果表明，该方法相比于传统基于物品的协同过滤算法，在准确率和召回率上都有所提升，在Top1-5推荐均值上分别提高了0.02和0.03。

Research on ItemCF Algorithm with User Affective Preference Matrix

The traditional item-based collaborative filtering （ItemCF） algorithm makes use of similarity between items to make individual recommendation to users. However， ItemCF only considers rating data and does not pay attention to users’emotional preferences. Based on user emotional preferences extracted from emoji expressions in reviews， an ItemCF algorithm with user emotional numbering matrix is proposed. This method uses the item co-occurrence matrix and the user’s affective preference matrix to calculate the item similarity， and then carries on the personalized recommendation. Experimental results on an online Internet education data set show that the proposed method outperforms the traditional item-based collaborative filtering algorithm in terms of accuracy and recall， and improves by 0.02 and 0.03 respectively on the average of Top1-5 recommendation.