View article

User-generated comments in online social media have recently attracted increasing attention with regard to the integration of end users' knowledge with the creation of new descriptive annotations for media resources (such as exploiting a comment written on a Flickr photo in order to create descriptive annotations for the photo). However, users have different levels of expertise and the content quality of comments varies from useful to useless. In this paper, we compare the characteristics of useful comments on different platforms (YouTube and Flickr) and we propose a method for inferring useful comments. We provide an analysis of the impact of different textual, semantic, and user features on the proposed inference method.