The user's context can immensely help in disambiguating user queries as well as returning more relevant results. The goal of this project is to exploit user's context efficiently and effectively to improve search.
Let us start with a couple of examples. One example of user's context is her current location. This context is very easy to obtain due to proliferation of GPS-equipped mobile devices. If the user searches for [starbucks] or [gas station] from a mobile device, we can use the user's exact location to return the nearest Starbucks and nearest gas station respectively. This is much better than returning a Starbucks or a gas station in the same city (or same zipcode) as done by most local search engines today.
Another example of user's context is the set of queries asked by the user in the current search session and/or the web pages clicked by her. If the user searches for [saturn] and the user's last query was [chrysler], we know that the user is referring to Saturn, the car company and not Saturn, the planet. We can return the results accordingly.
We started this project with a specific problem: type-ahead search over spatial databases. The idea is to return results directly while the user is typing in the query -- this helps reduce the amount of typing and clicking the user needs to perform (which is extremely cumbersome from mobile devices). Our goal is to use the user's context, primarily her location, to return relevant results for type-ahead search. This paper will appear in the 2011 SIGMOD conference. Going forward, we plan to generalize beyond location and beyond type-ahead search.
- Senjuti Basu Roy and Kaushik Chakrabarti, Location-Aware Type Ahead Search on Spatial Databases: Semantics and Efficiency, in ACM SIGMOD Conference, June 2011.