Sangati — A Social Event Web approach to Index Videos
Anjana Anilkumar, Anusha Sreenivasan, Animesh Sahay, Dilip Gurumurthy, M P Nirupama, Subramaniam Kalambur, Dinkar Sitaram, Ramesh Jain · 2018
In its inception, the World Wide Web was a means of sharing text documents primarily. As the Web has grown, the amount of videos and images has grown exponentially due to the desire of users to share their experiences to a wider audience. Video in particular has a very rich number of associations. For example, a goal scored at the last minute in a football game may have an association of “turning point of match” and should logically be hyperlinked to other similar events, as these associations are part of the experience the user would like to share and may be the reason for the user to take the video at that particular point. The amount of image and video content on the web far exceeds the text content, and there are applications, such as Facebook, Instagram, Snapchat and others which provide support for image and video sharing. However, hyperlinking of video content does not still capture the full range of hyperlinks latent in the video, making it difficult for users to share such experiences. In this paper, we model a video as a social event web, which is a complex graph with hyperlinks between events. The event web can be linked to the WWW in a natural way. We discover hidden hyperlinks within the video by mining for data using context from the event web and domain knowledge. Experiments on sports videos show the effectiveness of our framework and prove that our approach can provide richer hyperlinks between video segments, thereby extending the reach of the WWW to include the hidden content within videos and enabling users to share such links.