This experiment was developed to discover queries that have a bursty or spiky behavior. A bursty query is one that has a sudden and unusual change in volume with respect to its expected volume during a short timeframe. The experiment developed is based on Jon Kleinberg’s algorithm to detect temporal bursts in streams [KDD 2003].
The main idea is to model the query activity level as a (high & low) two-state Markov chain and find the best state sequence that favors both a small number of state transitions and states that conform with observed data. The experiment takes into account daily and weekly periodicity of queries and gives the time interval of the burst, if any, for a query.
It has a data visualization component with reverse IP lookup on the addresses in the data, plotting the patterns of activities such as a particle system. In a time-lapse animation, particles show queries from each location worldwide. Visualized are searches for "miss teen usa", "cricket" (world playoffs) "mattel" (recalls of toys with lead paint), and a close up on the Gulf of Mexico for "hurricane felix."
Bursty Queries was created by the Search Technologies group.
Hello, http://sandbox.yahoo.com/Bursty_Queries Great concept falls in line with something I am conceptualizing along with some partners. I can't hear the video. Steve Brooks
Comment by Steve Brooks (February 23, 2009, 3:42 pm)Tell us what you think, leave a comment: