This strikes me as a classic example where its a lot harder to positively identify than to negatively exclude.<p>What I mean is assign the class/tag of "political commercial" to all commercials. Then subtract any commercial seen more than two days after and less than three years before the next election. If the pool is big enough, I think it would work pretty well.<p>Or what they're really asking for is a commercial detector not so much a political commercial detector. Once you have a commercial detector, the political part of the task isn't going to be all that difficult.<p>Redoing the specs might make it easier to accomplish. I think they're not trying to feed an arbitrary video file to a program and have it pass political commercials, they're trying to select out a pile of political commercials given a fixed pile of footage, which is a totally different task than "given a single arbitrary video file in isolation from all others"