Vertical selection is the task of selecting the relevant verticals, if any, in response to a user's query. We focus on single vertical selection, defined as the task of predicting a single relevant vertical.
They looked at 18 different verticals and compared how their vertical selection methods compared with prior work on federated search. They used:
- Query string features which used key phrases in the query ("weather", "movies", etc...)
- Query log features which used the likelihood of the query given past queries on the vertical
- Corpus features which uses documents retrieved from the different verticals.
My congratulations to Jaime, Fernando, Jamie, and Jean-Francois!
(Note: Since the best paper was from a student there was no separate best student paper award)