Abstract
Evaluation is a key factor to reflect the quality of a recommender system algorithm. Traditional recommenders pose the problem as an optimization task where they seek to minimize the error in predicted rating for an item or predicted top-n items of interest with respect a user. However, these predictions do not often translate to a well-perceived recommendation. In this work, instead of the typical rating prediction task, we predict the amount of interaction an item would receive through a social network. In particular, we propose a simple and efficient model to generate a ranked list of tweets of a user in the order of expected user interaction that they would receive on Twitter, which is expressed in terms of retweets and favorites. We evaluate our proposed model on an extended version of the MovieTweetings dataset, which contains tweets that are generated when users rate movies on IMDb (using the IMDb iOS app), and show that the proposed model performs better compared to the baselines.