lecture9-queryexpansion-handout-6-per

# if we have a lot of judged documents we want a higher

This preview shows page 1. Sign up to view the full content.

This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: e Theore*cally Best Query x x o x x o o Δ oo o x Optimal query x x x x x x x S ec. 9.1.1 Rocchio 1971 Algorithm (SMART)   Used in prac*ce: x xxx Introduc)on to Informa)on Retrieval x x x non-relevant documents o relevant documents 1 qm = αq0 + β Dr ∑d d j ∈Dr j −γ 1 Dnr ∑d j d j ∈Dnr   Dr = set of known relevant doc vectors   Dnr = set of known irrelevant doc vectors   Diﬀerent from Cr and Cnr !   qm = modiﬁed query vector; q0 = original query vector; α,β,γ: weights (hand ­chosen or set empirically)   New query moves toward relevant documents and away from irrelevant documents 4 Introduc)on to Informa)on Retrieval Sec. 9.1.1 Subtle*es to note Introduc)on to Informa)on Retrieval Relevance feedback on ini*al query   Tradeoﬀ α vs. β/γ : If we have a lot of judged documents, we want a higher β/γ.   Some weights in query vector can go nega*ve Initial query o x Δ o x x x oΔ oo x x Sec. 9.1.1 Introduc)on to Informa)on Retrieval x x x x x x o x x x x known non-relevant documents o known relevant documents Revised query x x x   Nega*ve term weights are ignored (set to 0) Introduc)on to Informa)on Retrieval S ec. 9.1.1 S ec. 9.1.1 Posi*ve vs Nega*ve Feedback Relevance Feedback in vector spaces   We can modify the query based on relevance feedback and apply standard vector space model.   Use only the docs that were marked.   Relevance feedback can improve recall and precision   Relevance feedback is most useful for increasing recall in situa*ons where recall is...
View Full Document

{[ snackBarMessage ]}

Ask a homework question - tutors are online