Paper Title
Channel Recommendation for Youtube Ads Using Contextual Similarity Modeling

Video advertising has been growing progressively over the years. YouTube, being among the mediums that host the highest number of videos available on the internet, is growing in viewership year on year as well. With its ever-increasing audience pool, advertisers are slowly beginning to view it as a goldmine in terms of the sheer number of users they can reach out to on the website. This research attempts to tap into video advertising over YouTube to propose a context-driven algorithm that can be used to enhance ad targeting on the website. Current data-driven targeting strategies on the website mainly revolve around the use of user-level data, whereas, the context-driven strategies available to advertisers are mostly off-the-shelf strategies provided by YouTube. This paper proposes the use of geometric deep learning techniques to generate contextual similarity models based on the description of YouTube channels. This model is used to generate YouTube channel recommendations that are relevant for advertisers to serve their ads on. This paper then proposes a heuristic criterion to evaluate the recommendations generated against existing targeting strategies. Keywords - YouTube Advertising, Video Advertising, Geometric Deep Learning, Recommendation System, YouTube Channel Recommendations, Contextual Similarity Models.