Online video suggestion tool based on user interaction

Abstract

Online video watching systems have grown exponentially during the last few years. This development has given rise to the need for integration of the recommendation systems theory into such systems with the aim of offering users more relevant suggestions. This would result in a transfer of work load from the user to the machine thus making the user's viewing experience much more enjoyable. This dissertation proposes a system, similar to and based on YouTube, where users can view videos as recommended by the suggestion engine. Two distance/similarity matrices for both users and videos are stored centrally and updated each time a user views and rates a video. The suggestions are based mainly on user ratings and are split into three categories: content based, collaborative and YouTube related. These three types of suggestions are then compared to check which of the three gave best results and the advantages and disadvantages of each type of suggestion are analysed. Results of experiments conducted using the system suggest that both the content and collaborative based approaches performed better and were liked better by the user than the suggestions currently offered by YouTube.