Exploiting user reviews for automatic movie tagging

Auto-tagging movies with apt keywords/tags is essential and indispensable for online video providers. Tagged keywords are beneficial for movie promotion and recommendation. Currently, a popular approach is to propagate tags between similar videos identified by checking image similarity, which in principle takes each video as a sequence of frames. This approach is applicable for short video clips, however it is inefficient in processing long videos such as commercial movies, because it is very rare for two commercial movies to share a large portion of similar frames even if they belong to the same genre. In this work, we propose a novel scheme to auto-tag movies with two major steps. In the first step, we only consider popular movies with tremendous amount of attentions from online users, and we tag them by extracting keywords from user reviews. In the second step, unpopular movies are tagged by propagating the tags of similar popular movies to them. The similarity is evaluated based on multiple quantified attributes, including the movie summary, the title, the country, the genre and the tags, instead of frames to avoid expensive computation cost. To evaluate the performance of our scheme, we conduct experiments using data crawled from Douban, one of the largest movie rating websites in China. Experiment results demonstrate the superiority of our scheme by significantly improving tagging performance (in terms of Precision, Recall and F-Score) in comparison to baseline schemes.