USING TRANSFER LEARNING FOR VIDEO POPULARITY PREDICTION

Akshatha M R, Prathima V R

Abstract: Knowledge gained from social streams can be used to address many multimedia problems which cannot be solved by using traditional multimedia techniques alone. Some portion of videos in video portals exhibit sudden (bursty) rise in popularity, an effect which is not captured by video domain features alone Cross domain real-time transfer learning framework is used which utilizes knowledge from social streams (e.g., Twitter) and improve popularity prediction in the video domain. OSLDA model is used to detect topics from social streams [3]. Social Transfer algorithm is used for classifying videos with topics which is then used to calculate the social prominence and finally leading to the improved popularity prediction in the video domain. The framework has the ability to scale with incoming tweets in real time

Keywords: Cross-domain media retrieval, Topic model, Twitter, Video popularity, YouTube

DOI: https://doi.org/10.15623/ijret.2014.0315069