See Clicks Differently: Modeling User Clicking Alternatively with Multi Classifiers for CTR Prediction

Many recommender systems optimize click through rates (CTRs) as one of their core goals, and it further breaks down to predicting each item's click probability for a user (user-item click probability) and recommending the top ones to this particular user. User-item click probability is then estimated as a single term, and the basic assumption is that the user has different preferences over items. This is presumably true, but from real-world data, we observe that some people are naturally more active in clicking on items while some are not. This intrinsic tendency contributes to their user-item click probabilities. Besides this, when a user sees a particular item she likes, the click probability for this item increases due to this user-item preference.

Therefore, instead of estimating the user-item click probability directly, we break it down into two finer attributes: user's intrinsic tendency of clicking and user-item preference. Inspired by studies that emphasize item features for overall enhancements and research progress in multi-task learning, we for the first time design a Multi Classifier Click Rate prediction model (MultiCR) to better exploit item-level information by building a separate classifier for each item. Furthermore, in addition to utilizing static user features, we learn implicit connections between user's item preferences and the often-overlooked indirect user behaviors (e.g., click histories from other services within the app). In a common new-campaign/new-service scenario, MultiCR outperforms various baselines in large-scale offline and online experiments and demonstrates good resilience when the amount of training data decreases.