Predicting User-specific Temporal Retweet Count Based on Network and Content Information

    Twitter generates a constant flow of quality news and mixed social content. While it is relative easy to separate large popularity news sources from personal messages, we address a more difficult question to predict the success of a single message among all messages of the same user. We describe a temporal evaluation framework to analyze which messages of which users will be retweeted the most. It turns out that global popularity depend mostly on the network character- istics of the user, while for a given user, the retweet count of a single message can be predicted best by using a variety of features of the content, including linguistic characteristics.

    Bálint Daróczy, Róbert Pálovics, Vilmos Wieszner, Richárd Farkas, András A. Benczúr
    INRA workshop, RecSys 2015