Social event detection using multimodal clustering and integrating supervisory signals

摘要: A large variety of features can be extracted from raw multimedia items. Moreover, in many contexts, like the case uploaded by users social media platforms, items may linked to metadata that very useful for a analysis tasks. Nevertheless, such are typically heterogeneous and difficult combine unified representation would suitable analysis. In this paper, we discuss problem clustering collections with purpose detecting events. order achieve this, novel multimodal algorithm is proposed. The proposed method uses known currently examined domain, supervise fusion procedure. It tested on MediaEval event detection challenge data compared spectral approach early fusion. By taking advantage explicit supervisory signal, it achieves superior accuracy additionally requires specification much smaller number parameters. has wider scope; not only applicable task detection, but other problems as well.