Dynamic Selection of Information Retrieval Algorithms.pdf

Dynamic Selection of Information Retrieval Algorithms Claudia Rolker Ralf Kramer Forschungszentrum Informatik FZI HaidundNeuStr Karlsruhe Germany email rolker kramerfzide Abstract Users often search in information retrieval systems iteratively We claim that changing the underlying retrieval algorithm and query expansion algorithm during these iterations yields improved results In the rst part of this pap er results of preliminary exp eriments are presented which indicate the viability of our claim In the second part we fo cus on our approach for the dynamic selection of the prop er retrieval algorithms and query expansion algorithms Our approach is based on casebased planning A retrieval algorithm sequence which holds certain quality criteria is set up due to the consideration of previous exp eriences and cases Intro duction Usually information retrieval systems IRS are used iteratively to nd the information lo oked for Dep end ing on the query result the query is mo died several times Our hyp othesis is that a sequential execution of dierent retrieval algorithms and query expansion algorithms can have synergy eects Hence our in formation retrieval system is based on a dynamic selection of a retrieval algorithm and a query expansion algorithm within each iteration Furthermore we claim that the user is much more satised with the search result when he can negotiate the achieved quality of the result at the b eginning of the search Hence our system lets the user enter the numb er of exp ected relevant do