How Trend Detection Can Flag Bursty Advertising Campaigns

One of the metrics we track in theĀ Visual Global Entity Graph (VGEG) 2.0 using Google's Cloud Video API to analyze CNN's television news programming in collaboration with the Internet Archive's Television News Archive is the list of distinct visual entities (objects and activities) depicted onscreen throughout the day. Using simple trend detection metrics like TFIDF it is straightforward to surface distinct bursts of entities that are more common on a given day than other days. At present we don't draw a distinction between advertisements and news programming for the VGEG, meaning such trend detection can sometimes be led astray by bursty advertising campaigns that run a flurry of ads on a given day or set of days and then halt.

One simple trend analysis identified "chicken nuggets" as being an unusual trend for a week, with few depictions before or after. Unsurprisingly, it turns out this trend resulted from a set of advertising campaigns those days resulting in a burst of ads from Long John Silver's and McDonalds (the french fie thumbnail images are cases where the same second of airtime showed both nuggets and fries, with the 1fps thumbnail coming down on the fries).

Live Version.