融合评论文本和评分矩阵的电影推荐算法研究

Research on Movie Recommendation Algorithm Combining Review and Rating Matrix

针对电影推荐领域评论文本信息在传统推荐中未被充分利用,推荐准确度不高的问题,提出一种融合评论文本和评分矩阵的电影推荐算法.首先通过基于自注意力机制的双向门控循环单元神经网络对电影的高质量评论文本进行建模,提取评论文本中的特征,构建电影评论特征矩阵.同时使用隐语义模型对用户评分矩阵进行分解,得到用户潜在兴趣矩阵和电影潜在特征矩阵.最后通过改进的DeepFM融合电影评论特征矩阵和电影潜在特征矩阵得到电影综合得分并形成推荐列表,以达到推荐的目的.实验结果表明,与其他几种电影推荐算法相比,在AUC、F-Score、RMSE上平均提升分别约7.37%、9.32%、8.23%,最高提升分别为11.60%、15.22%、12.79%.

In the field of movie recommendation,focusing on the problems that the movie review information was underutilized in traditional movie recommendation which not led to high recommendation accuracy,a movie recommendation algorithm combining movie review and rating matrix was proposed.Firstly,the bidirectional gated recurrent unit neural network based on self-attention mechanism was used to model the high-quality movie reviews,extract the features from them,and construct the feature matrix of movie review.At the same time,the latent factor model was used to decompose the rating matrix to get the potential interest matrix of users and the potential feature matrix of movie.At last,improved DeepFM was used to combine the feature matrix of movie review and the potential feature matrix of movie.Through the above process,the movie comprehensive score was calculated and the movie recommendation list was formed.The experimental results show that compared with some other recommendation algorithms,the average increase in AUC,F-score and RMSE is about 7.37%,9.32%and 8.23%,and the highest increase is 11.60%,15.22%and 12.79%.