Predicting User Engagement in Twitter with Collaborative Ranking

Abstract

Collaborative Filtering (CF) is a core component of popular web-based services such as Amazon, YouTube, Netflix, and Twitter. Most applications use CF to recommend a small set of items to the user. For instance, YouTube presents to a user a list of top-n videos she would likely watch next based on her rating and viewing history. Current methods of CF evaluation have been focused on assessing the quality of a predicted rating or the ranking performance for top-n recommended items. However, restricting the recommender system evaluation to these two aspects is rather limiting and neglects other dimensions that could better characterize a well-perceived recommendation. In this paper, instead of optimizing rating or top-n recommendation, we focus on the task of predicting which items generate the highest user engagement. In particular, we use Twitter as our testbed and cast the problem as a Collaborative Ranking task where the rich features extracted from the metadata of the tweets help to complement the transaction information limited to user ids, item ids, ratings and timestamps. We learn a scoring function that directly optimizes the user engagement in terms of nDCG@10 on the predicted ranking. Experiments conducted on an extended version of the MovieTweetings dataset, released as part of the RecSys Challenge 2014, show the effectiveness of our approach.Comment: RecSysChallenge'14 at RecSys 2014, October 10, 2014, Foster City, CA, US

Spreading the news

“…GBT have shown to outperform other models in classification and regression tasks and have been used successfully for audience engagement prediction [6].…”

What and How long: Prediction of Mobile App Engagement

“…The third type of approach (c), adopted by the web-analytics community, has been studying user engagement through online behaviour metrics that assess users' depth of engagement within a site, e.g., the time spent on a webpage. Studies on user engagement in the contexts of desktopbased systems [56] and websites [13,65] have shown that simple metrics such as dwell time are meaningful and robust in modelling user engagement. For example, the time spent on a resource has been validated as an effective metric for measuring user engagement in the context of web search [1,3], and recommendation tasks [64].…”

“…One way user engagement has been measured at a large scale is by tracking how long users spend with content, e.g., the time spent on a webpage. Studies on user engagement in the contexts of desktop-based systems [56] and websites [13,65] have shown that simple metrics such as dwell time are meaningful and robust in modelling user engagement. More importantly, these studies have shown that with an awareness of engagement, users' experience with a system can be substantially improved which in turn leads to user growth, user retention, and increasing revenue streams.…”

Mining Affective Context in Short Films for Emotion-Aware Recommendation

“…Context-aware recommender systems (CARS) [9,10] improve recommendation quality by incorporating contextual information to the recommendation process -e.g., time [18], location [26], social ties [16], or mood [34] -which enables them to adapt to the specific user's situation. In this work we are particularly interested in emotional context and its impact in CARS.…”

“…We incorporate the emotional context into the recommender system by following a Collaborative Ranking (CR) approach [12,16] -other state-of-the-art alternatives for CARS include Factorization Machines [33] and N-dimensional Tensor Factorization [21]. The main idea behind this approach is to cast the recommendation problem into a (personalized) learning-to-rank task [25].…”