Collaborations on YouTube: From Unsupervised Detection to the Impact on Video and Channel Popularity

Detection and Analysis of Content Creator Collaborations in YouTube Videos using Face- and Speaker-Recognition

“…For the detection of collaborations in YouTube, the existing work CATANA [13] leverages occurring faces in the video through face recognition to identify the collaborators. While CATANA yields good results on video content with a high number of appearing faces, in cases with less or no appearing faces it would not perform well resulting in no detections.…”

“…In the following, we introduce existing work on the subject of collaboration detection, specifically the CATANA [13] framework. Further related work on the applied methods of speaker recognition as well as face tracking and active speaker detection are discussed in this section.…”

“…Concerning the analysis and detection of collaborations in YouTube, to the best of our knowledge, not much existing work was conducted until now. Koch et al [13] proposed an unsupervised framework for collaboration detection in YouTube videos based on face recognition named CATANA. In addition to the detection, the framework is also capable of collecting and analyzing YouTube statistics concerning collaborations and their impact on YouTube popularity.…”

“…Detected faces are then further processed through a CNN-based face recognition method extracting 1,792-dimensional face embeddings which are stored for clustering in a later step. Global clustering on all found face embeddings using the HDBSCAN clustering (a) CATANA system architecture [13].…”

“…(b) Video views growth factor for collaborations [13]. algorithm [11] then associates the appearing individuals across videos.…”