Event Identification in Social Media.

Abstract

Social media sites such as Flickr, YouTube, and Facebook host substantial amounts of user-contributed materials (e.g., photographs, videos, and textual content) for a wide variety of real-world events. These range from widely known events, such as the presidential inauguration, to smaller, community-specific events, such as annual conventions and local gatherings. By identifying these events and their associated user-contributed social media documents, which is the focus of this paper, we can greatly improve local event browsing and search in state-of-the-art search engines. To address our problem of focus, we exploit the rich “context” associated with social media content, including user-provided annotations (e.g., title, tags) and automatically generated information (e.g., content creation time). We form a variety of representations of social media documents using different context dimensions, and combine these dimensions in a principled way into a single clustering solution—where each document cluster ideally corresponds to one event—using a weighted ensemble approach. We evaluate our approach on a large-scale, real-world dataset of event images, and report promising performance with respect to several baseline approaches. Our preliminary experiments suggest that our ensemble approach identifies events, and their associated images, more effectively than the state-of-the-art strategies on which we build. 1.