Now we just need a model trained against the full content of these 6454 videos to use as a recommendation engine for high-quality new videos.