lecture7-vectorspace-handout-6-per

713 at query me only compute scores for docs in the

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

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: e many docs → these (many) docs get eliminated from set A of contenders   Say, at least 3 out of 4   Imposes a som conjunc*on on queries seen on web search engines (early Google)   Easy to implement in pos*ngs traversal 3 Introduc)on to Informa)on Retrieval S ec. 7.1.2 3 of 4 query terms 3 4 8 Brutus 2 4 8 16 32 64 128 3 5   Precompute for each dic*onary term t, the r docs of highest weight in t s pos*ngs 16 32 64 128 Calpurnia 1 2 8 13 21 34 13 16 32   Call this the champion list for t   (aka fancy list or top docs for t)   Note that r has to be chosen at index build *me   Thus, it s possible that r < K Scores only computed for docs 8, 16 and 32. Introduc)on to Informa)on Retrieval Sec. 7.1.3 Champion lists Antony Caesar Introduc)on to Informa)on Retrieval Sec. 7.1.3   At query *me, only compute scores for docs in the champion list of some query term   Pick the K top ­scoring docs from amongst these Introduc)on to Informa)on Retrieval Sec. 7.1.4 Exercises Sta*c quality scores   How do Champion Lists relate to Index Elimina*on? Can they be used together?   How can Champion Lists be implemented in an inverted index?   We want top ­ranking documents to be both relevant and authorita)ve   Relevance is being modeled by cosine scores   Authority is typically a query ­independent property of a document   Examples of authority signals   Note that the champion list has nothing to do with small docIDs           Introduc)on to Informa)on Retrieval Sec. 7.1.4 Wikipedia among websites Ar*cles in certain newspapers A paper with many cita*ons Many bitly s, diggs or del.icio.us marks (Pagerank) Introduc)on to Informa)on Retrieval Quan*ta*ve Sec. 7.1.4 Modeling authority Net score   Assign to each document a query ­independent quality score in [0,1] to each document d   Consider a simple total score combining cosine relevance and authority   net ­score(q,d) = g(d)...
View Full Document

Ask a homework question - tutors are online