Diversity

Individual tweet scores can produce a repetitive feed when several high-scoring items come from the same author or use the same medium. This lesson adds a list-level rule to balance those repetitions with predicted engagement.

Why do you need diverse Tweets?

Let’s assume that you adopted the following modelling option for your Twitter feed ranking system. The Tweet selection component will select one-hundred thousand Tweets for user A’s Twitter feed. The stage one ranker will choose the top five-hundred most engaging Tweets for user A. The stage two ranker will then focus on assigning engagement probabilities to these Tweets with a higher degree of accuracy. Finally, the Tweets will be sorted according to the engagement probability scores and will be ready for display on the user’s Twitter feed.