Cost-Effective User Monitoring for Popularity Prediction of Online User-Generated Content

In this paper, we study on the popularity prediction of online user-generated contents, where high quality predictions give us much more flexibility and preparing time in deploying limited resources (such as advertising budget, monitoring capacity) into more popular contents. However the high retrieval cost of data used in prediction is a big challenge due to the large amount of users and contents involved. We propose a notion that higher popularity user-generated contents can be predicted by concentrating on fewer but informative users, as we notice the fact that contents generated by those users tend to become popular while that which are generated by the rest users do not. We develop a cost-effective popularity prediction framework to fulfil online prediction. It contains 3 modules: (a) online data retrieving, (b) informative users selection and (c) popularity prediction. A hybrid user selection algorithm and several popularity prediction algorithms/improvements are presented, and their performance are evaluated and compared using (a) the selected users' generated data and (b) all users' generated data, retrieved from Sina Weibo Microblogger. The best prediction algorithm reaches a 78% accuracy at the time of 24 hours after publishing time when level width N-l equals 500. And the best combination of prediction and selection algorithms performs only about 7% worse on dataset