A Framework for Video Popularity Forecast Utilizing Metaheuristic Algorithms

Data available online is growing day by day that leads to a tough competition among data publishers to attract the largest possible audience. Obtaining reliable prediction related to future popularity of the online content becomes a concern to the publishers. In this paper, a novel nature-inspired metaheuristic framework is proposed that shortlists the prominent features of the video to participate in making predictions. Fitness of the set of selected features is calculated using mean square error of the deviation between predicted and actual popularity. The exhaustive examinations on standard benchmark parameters is performed on the datasets, i.e., Facebook 2015, Top and random datasets of YouTube. The performance of the proposed algorithms, namely particle swarm optimization with support vector regression (PSO-SVR), bat algorithm with support vector regression (BA-SVR), dragonfly algorithm with support vector regression (DA-SVR) and existing prediction method support vector regression (SVR), is compared. DA-SVR outperforms SVR, PSO-SVR and BA-SVR in terms of coefficient of regularization (R2) score and number of features selected.