Revolutionizing Content Identification: A Comprehensive Study on YouTube Live and Video-on-Demand Detection

With the growing popularity of over-the-top (OTT) streaming services, distinguishing between live and video-on-demand (VoD) streams is becoming increasingly important for network operators and OEMs [1]. However, traditional Deep Packet Inspection (DPI) techniques struggle to differentiate between the two as live and VoD streams share similar network markers. To address this, two machine learning-based strategies are proposed. The first involves training a convolutional neural network (CNN) model using IO graphs of live and VoD video streams and making predictions for new streams using their IO graphs. The second involves using the nfstream library to extract and identify YouTube live and VoD streams from packet capture (pcap) files, followed by building a machine learning model that can identify live and VoD streams with an accuracy of over 96% [2]. Our proposed solutions have important implications for monitoring and improving customer experience, as well as better traffic management.