TY - GEN
T1 - Ranking database queries with user feedback
T2 - 13th International Conference on Database Systems for Advanced Applications, DASFAA 2008
AU - Agarwal, Ganesh
AU - Mallick, Nevedita
AU - Turuvekere, Srinivasan
AU - Zhai, Chengxiang
PY - 2008/7/21
Y1 - 2008/7/21
N2 - Currently, websites on the Internet serving structured data allow users to perform search based on simple equality or range constraints on data attributes. However, to begin with, users may not know what is desirable to them precisely, to be able to express it accurately in terms of primitive equality or range constraints. Additionally, in most websites, the results provided to users can be sorted with respect to values of any one particular attribute at a time. For the user, this is like searching for a needle in a haystack because the user's notion of interesting objects is generally a function of multiple attributes. In this paper, we develop an approach to (i) support a family of functions involving multiple attributes to rank the tuples, and (ii) improve the ranking of results returned to the user by incorporating user feedback (to learn user's notion of interestingness) with the help of a neural network. The user feedback driven approach is effective in modeling a user's intuitive sense of desirability of a tuple, a notion that is otherwise near impossible to quantify mathematically. To prove the effectiveness of our approach, we have built a middleware for an application domain that implements and evaluates these ideas.
AB - Currently, websites on the Internet serving structured data allow users to perform search based on simple equality or range constraints on data attributes. However, to begin with, users may not know what is desirable to them precisely, to be able to express it accurately in terms of primitive equality or range constraints. Additionally, in most websites, the results provided to users can be sorted with respect to values of any one particular attribute at a time. For the user, this is like searching for a needle in a haystack because the user's notion of interesting objects is generally a function of multiple attributes. In this paper, we develop an approach to (i) support a family of functions involving multiple attributes to rank the tuples, and (ii) improve the ranking of results returned to the user by incorporating user feedback (to learn user's notion of interestingness) with the help of a neural network. The user feedback driven approach is effective in modeling a user's intuitive sense of desirability of a tuple, a notion that is otherwise near impossible to quantify mathematically. To prove the effectiveness of our approach, we have built a middleware for an application domain that implements and evaluates these ideas.
UR - http://www.scopus.com/inward/record.url?scp=47249118332&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=47249118332&partnerID=8YFLogxK
U2 - 10.1007/978-3-540-78568-2_31
DO - 10.1007/978-3-540-78568-2_31
M3 - Conference contribution
AN - SCOPUS:47249118332
SN - 3540785671
SN - 9783540785675
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 424
EP - 431
BT - Database Systems for Advanced Applications - 13th International Conference, DASFAA 2008, Proceedings
Y2 - 19 March 2008 through 21 March 2008
ER -