Discovering Items with Potential Popularity on Social Media

Predicting the future popularity of online content is highly important in many applications. Under preferential attachment influence popular items get more popular thereby resulting in long tailed distribution problem. Consequently, new items which can be popular (potential ones), are suppressed by the already popular items. This paper proposes a novel model which is able to identify potential items. It identifies the potentially popular items by considering the number of links or ratings it has received in recent past along with it's popularity decay. For obtaining an efficient model we consider only temporal features of the content, avoiding the cost of extracting other features. Prediction accuracy is measured on three industrial data sets namely Movielens, Netflix and Facebook wall post. We have found the recent gain in link formation are a good predictor for future link formation as compare to total links, in other words we can say people follow the recent behaviours of their peers, considering the fact that collective attention makes something popular. Experimental results show that compare to state-of-the-art model our model have better prediction accuracy.