一种基于特征的混合推荐方法

摘要

This paper proposes a hybrid method based on features of movies in terms of user interests.In this paper, topics of the movie can be captured by LDA model with the documents of movie plots.Accommodating with all the above mentioned features, user profile which presents user interests can be created by the rated movies.Finally, the predict rating for user to the new movies depends on the user similarity calculated by the user profile and the affinity of the user and the movie.Experiments based on the MovieLens-1m dataset show that the proposed method is better than the existing collaborative filtering method.%论文给予用户关注度提出了基于特征的混合推荐方法F-Hybrid(Feature-Hybrid).通过LDA文本建模从电影摘要中抽取电影主题,并结合电影的其他基本信息构成电影的综合特征;然后根据用户评价过的电影的特征刻画出用户画像;最终,用户对新电影的预测评分由两个因素决定:一是由用户画像计算出的用户相似度;二是由变换用户评分矩阵得到的电影与用户关联度.基于MovieLens-1m数据集上的实验结果表明,F-Hybrid方法在准确率等指标上均优于一般的推荐方法.