Towards Automatic Image Understanding and Mining via Social Curation

摘要: The amount and variety of multimedia data such as images, movies music available on over social networks are increasing rapidly. However, the ability to analyze exploit these unorganized remains inadequate, even with state-of-the-art media processing techniques. Our finding in this paper is that emerging curation service a promising information source for automatic understanding mining images distributed exchanged via media. One remarkable virtue datasets they weakly supervised: content manually collected, selected maintained by users. This very different from other sources, we can utilize characteristics without expensive In present machine learning system predicting view counts first step image evaluation. experiments confirm simple features extracted corpus much superior terms count prediction than gold-standard computer vision research.