This project aims to develop a system that can collect and aggregate videos uploaded by users on social media. First, the videos related to any event are download (from Youtube, Facebook, etc.) by searching the related keywords. Then the videos are segmented into shots, such that each shot contains the same semantic information. Then, deep features are extracted from each shots, and the videos are clustered based on these features. At the end one video is selected as the representative one from each cluster, based on some quality related metrics.