[mythtv-users] Feature request: Detect existence of closed captioning

[f-myth-users at media.mit.edu]

    > Date: Sun, 18 Jul 2010 21:38:31 -0700
    > From: Douglas Peale <Douglas_Peale at comcast.net>

    > Brand names are almost never mentioned in US TV shows, but almost always in commercials. That would be a good indicator that a
    > CC was in a commercial if it had a brand name in it. I would not expect that a list of brand names would be difficult to assemble.

That's a good idea.  It would fail on various DIY-type shows (and
-especially- on the rare "2010 Builder's Expo" sorts of shows where
people are wandering around a trade show) but should be okay on most
others.

One problem, though, might be expanding from "detected a brand name"
to detecting the boundaries of the entire block of commercials.  But
it could be used as a weight to increase the detection probability
of a segment that the commflagger is already evaluating.

Another approach to learning might be to only conclude a phrase is a
commercial if it appears on multiple shows with different programids
(in other words, multiple different series, etc).  You'd have to avoid
common memes (including from quotes on news shows).  And many ads have
remarkably little in the way of actual dialog; the words/second density
is low.

But it'd be amusing to take a large corpus and either try to do
unsupervised learning (tagging the boundaries via statistical
methods, say) or supervised learning (taking a pretagged corpus
and attempting to extend it).  I could easily see some undergraduate
AI projects trying this out.

    > Even product placements usually only show the product, not talk about it.

    > The only way for advertisers to defeat this method would be to pay to have their brands mentioned in the shows.

Well, we're way too tiny a market for them to even notice it's
happening, much less attempt to defeat it...

    > Might need to special case "Sponsered by" occurrences since that show up a bit in sports.

And the start of most PBS programs, but that's probably a feature---
commflag exactly that segment and don't bother w/the rest of the
program.