Unsupervised sports video particles annotation based on social latent semantic analysis

Large volumes of video particles on practically every major sports event are posted on social media. According to socialbakers.com [1], four of the top twenty Facebook pages focus on sports. These particles can be processed and automatically annotated with events, entities etc. Furthermore, several annotated particles referring to a different time interval of the same sports event, could be synchronized to accomplish annotation of full sports games. Towards this direction, in this paper an innovative scheme is proposed that performs unsupervised annotation of sports video particles, posted on social media. The scheme is based on an intelligent wrapper architecture that automatically gathers and segments content and on the newly introduced Social Latent Semantic Analysis. This paper forms an initial study of automatic sports video particles annotation and experiments indicate its promising performance.