=========================preview======================
(COMP336)midterm96F.pdf
Back to COMP336 Login to download
======================================================
COMP336/533InformationRetrieval
MID-TERMEXAMINATION
October17,1996
Iherebydeclarethatnoillegalaidswereusedincompletingthisexamination.AndIunderstand thatanyactofcheatingwillresultinseverepenalty,includingexpulsionfromtheuniversity.
Sign: Date:
Name: StudentID:Circle:336/533
1.[15]Plotonthegraphsbelowthetypicalbehaviorofthefollowingparameterswhenaterm appearsinmoreandmoredocuments(i.e.,documentfrequency,df,increases): (a)theinversedocumentfrequency(idf)oftheterm, (b)thetotaldocumentsimilaritybetweenthesetofdocuments,and (c)thetermdiscriminationvalue(dv)oftheterm. Noticethatweareonlyconcernedwiththe\average"or\typical"behaviour.
1 df1 df1 df
2.[20]Usingthe(improved)KMPmethodasdescribedinthetextbook(i.e.,NOTthe\im-proved"method),.llinthefollowingshiftarray,#istheend-of-stringcharacter. patternno.of charshifts
b
a
b
d
b
d
#
Whatarethevaluesforthenext[]array.
3.[10]Giventhefollowingprecision/recallgraphfortwosystems,AandB,whichhaveexactly thesameaverageprecision: (a)underwhatsituationwillauserpreferAoverB.why. (b)underwhatsituationwillauserpreferBoverA.why. Noticethatanswerslike\Aisbetterwhenrecallislow"and\Bisbetterwhenrecallishigh" areNOTwhatIamlookingfor.
precision
A
B
recall
4.[20]Fillintheprecisionandrecallvaluesinthefollowingtable.`x'meansthedocumentis relevant.Thereareatotalof3relevantdocumentsamongatotalof100documents.
Recall-precisionafterretrievalofndocuments nDocIDRecallPrecisionFallOut
xxx 123456 a b c d e f
... ... .... ... ...
...100 ...... .... ... ...
5.[15]Giventhefollowinginverted.le(orindex):
x
;!
1,2
3,1
y
;!
2,3
z
;!
2,3
1,1
Apostingsisoftheformhdocidtfi.Supposethereareonly3documents,
(a)Whataretheidfvaluesforthethreeterms,xyz.theindex.le.
x:
y:
z:
(b)Giventhequery\xy"withquerytermweightsof1forbothqueryterms,calculatethe scoresofthedocumentsusingtfxidfweightsandthecosinesimilaritymeasures.
(c)Ifthequerytermweightsarechangedto0.5,doesitchangethedocumentscores.does itchangethedocumentranking.Why.Youdon'tneedtorecalculatethescores.A qualitativeexplanationissu.cient.