The Multimedia Challenges in Social Media Analytics
Tat‐Seng Chua · 2014
With the popularity and wide acceptance of social networks, users are now sharing information on multiple aspects of their life and on a wide range of social networks. In the meantime, there is a huge amount of situational information generated by sensor devices, often as part of human activities. Thus for any given entity, we can now find a wide variety of social, device and structured information from multiple sources. The generation of reliable social media analytics with respect to any entity is hence a highly challenging (multimedia) task. The key challenges include the ability to: (a) gather --representative? data about an entity from multiple sources; (b) handle the increasing amount of non-textual media content; (c) detect and track sub-topics around the target entity, along with deep analysis tools such as named-entity extraction, visual concept detection and sentiment analysis; and (d) generate predictive and prescriptive analytics. This talk describes a live social observatory system that we have developed and our research efforts to tackle the above challenges. In particular, we outline our research to transform unstructured live social media streams into descriptive, predictive and prescriptive analytics.