Evaluating folksonomy information sources for genre prediction
Deepa Anand · 2014
Automatic genre identification is a task which plays a crucial role in many domains such as automatic storytellers, recommender systems and web page topic detectors. Genre classification is especially interesting in the domain of narrative content which is characterized by a large number of ambiguous and overlapping categories. The rise in popularity of social tagging systems forms a rich source of input information which could be harnessed for this task. In this paper we investigate two different information folksonomy sources for the movie domain namely: keywords and tags, the first of which is user annotated and expert monitored whereas the latter is non-monitored. A comparison is performed to assess the efficacy of both sources in solving this multi-label classification problem and it is found that the in spite of being expert monitored and better structured, keywords are worse predictors of the genres of movies than tags in most cases.