Relevance feedback + Text classification (Chapter 9+13)
-vo,*..i
Definition 1 (Rocchio relevance feedback)
Rocchio relevance fi-idhach. has the form
ID,
£ i-
'\D„\
Text classification and Naive Bayes (Chapter 13)
Definition 2 (Naive Bayes Classifier)
Naive Bayes IND] Class-lf'tur iisxwutfv that Ihr. ttffccA of tin; cubic of a urztlkior x on a given "la>t f iv cb.i.^ ro!.;iii.;.;iiial iinlrpcinl.clia./c liicvmn ptocidc.i a way of calcr.ialrna lb' posterior probability P(c\x) from cfast prkr probability Pic), predictor prior probability P{x) and probability of ihr. predictor given the class P(x\c)
where qo is the original query vector, Dr is Hit set of relevant documents, Dn, is the sel of non-relevant documents and the ridiir.i ,t, H, ; dcj.rnd or. the ni/atem setting.
Exercise 9/1
What is the main purpose of Rocchio relevance feedback?
Exercise 9/2
A \iters primary query is cheap CD* • heap DVDs • zire.:iifjy thtap CDs. The user ha* n on two rlncimwrls^-fii-.-M a dm-i •tir.rkTi<l,»-~ I})).* ,■.,>■■.>!* Hnp.Wr--. rfoap CDt as relevant and doc2 cheap thrill? DVDs as non-relevant Assume that we use a simple It scheme v.itiioul vector length norma li-iitua]. Wluit wmilil In lln- iratriK 1 uii-c sjimrv !T <miK:rifiri-]f; III? Rix.-hio .■lr^1i,.:i- frwdtiark wil.li win** .1 = 1, /? = (1.75. flJid
	re e mi 11.	mri niUiľniľ	
t en in	docl	doc2	query
CDs -			■1
cheap		/	:i
			0
thrills		i Q	i)
DVD?		i /	i
j.itn uLi'K		0	i
rafpnéirlon r = (*j,... P(«|i')
1.....*.)
P(i,)..i"(««!
ifcr hiqhŕM pn.'ífHr.r pmhúitlií.ij in the. aiitt
t o o / / y