Fusing auxiliary information into ratings has shown promising performance for many recommendation tasks, such as age, sex, vocation of users or actors, director, genre, reviews of movies. However, all above auxiliary information is still sparse and not informative. For movie recommendations, besides the above information, there exists richer information in plot texts, exerting huge impacts on improving the recommendation accuracy. In this paper, we explore effective fusion of movie ratings and plot texts, we propose a deep plot-aware generalized matrix factorization for collaborative filtering model, which effectively combines both ratings and plot texts to implement a generalized collaborative filtering. To verify our proposal, we conduct extensive experiments on two popular datasets, and the results perform better than other state-of-the-art approaches in common recommendation tasks.

将辅助信息融合到评分中已显示出对许多推荐任务的良好表现，例如年龄，性别，用户或演员的职业，导演，体裁，电影评论。但是，上述所有辅助信息仍然很少，并且没有提供信息。对于电影推荐，除上述信息外，情节文本中还存在更丰富的信息，这对提高推荐准确性有巨大影响。在本文中，我们探索电影收视率和情节文本的有效融合，提出了一种深度情节感知的广义矩阵分解用于协作过滤模型，该模型有效地结合了收视率和情节文本以实现广义协作过滤。为了验证我们的建议，我们对两个流行的数据集进行了广泛的实验，