Metadata based multi-labelling of YouTube videos

It is a challenging task to find video of interests on YouTube due to huge size of its repository. Multiple labels, if provided, can make search faster. This paper describes a two level automated mechanism to generate multiple labels for videos using their text based meta-data features. The first level of classification categorize videos into 5 harassment categories and then a second level generate a positive or negative label i.e. harassment or non-harassment. There has been no multi level classification of YouTube videos. Previous works have classified videos on a single level only whereas our work brings novelty to the approach by classifying videos into multi labels. Such a work can be useful for law enforcement and intelligence agencies to identify the unwanted and malicious videos on the Internet. The proposed approach has successfully generated multiple labels for unlabelled test videos.