UNIVERSITY COLLEGE DUBLIN

cobblerbeggarAI and Robotics

Oct 15, 2013 (3 years and 9 months ago)

84 views

UNIVERSITY COLLEGE DUBLIN


NATIONAL UNIVERSITY OF IRELAND, DUBLIN


An Colaiste Ollscoile Baile Atha Cliath

Ollscoil na hEireann, Baile Atha Cliath

___________________________________

SUMMER EXAMINATIONS 2001

___________________________________


SCBDF0005
-

B.Sc. HONOU
RS DEGREE EXAMINATION


COMPUTER SCIENCE


COMP 4016: The Intelligent Internet


Prof. J. Hughes

Prof. M. Keane

Dr. N. Kushmerick*


Answer any 3 of the following 5 questions. All questions carry equal marks.


Time allowed: 1 hour 45 minutes


This examination

is “open book”. You may refer to any course materials during the
examination.



1.

Information extraction

features centrally in Rapier (Califf & Mooney), WWKB
(Craven et al), Wrapper induction (Kushmerick et al), Cora (McCallum et al), and
DIPRE (Brin).

First, use detailed examples to illustrate the particular information
extraction tasks on which each of the five system focuses. Second, characterize
the kinds of documents for which each system is best suited.




2.

Ontologies

play an important role
in SHOE (Luke et al), Ariadne (Barish et al),
and Staab et al’s community portals. First, describe how the ontologies used by
these three systems differ from conventional object
-
oriented class hierarchies.
Second, for each of these three systems, use det
ailed examples to describe one
specific aspect of the system that is enabled or simplified by the use of
ontologies.











-

1
-





p.t.o.



3.

A
machine learn
ing

system transforms its input into some form of knowledge
that enable
s

the

system to perform some particular task. Machine learning
underlies the performance of the ShopBot (Doorenbos et al), AdEater
(Kushmerick), ILA (Perkowitz & Etzioni), and Sahami et al’s junk email
classifier systems. First, for each of these four systems
, use detailed examples to
describe the input to the learning algorithm, the knowledge that is learned, and
the task that is performed. Second, use
any one

of these systems to explain the
idea of a “learning curve”, and specifically explain how learning c
urves are used
to predict

the performance of a machine learning system.





4.

“Classical”
information retrieval

is the task of identifying documents in a large
unstructured corpus that are relevant to some ‘information need’, typically
specified by a
set of keywords. Scatter/Gather (Cutting et al), Pagerank (Brin &
Page), HITS (Gibson et al) and cluster mining (Perkowit
z

& Etzioni) each
construe the task of helping people find relevant documents somewhat
differently. Use detailed examples to describ
e how each of these four systems
differs from the classical model of information retrieval.





5.

Recommendation
, the task of suggesting specific items to a user, lies at the
center of GroupLens (Resnick et al), WebWatcher (Joachims, et al), Letizia
(Li
eberman), and Cohen & Fan’s music recommender. Recall that the two main
approaches to recommendation are collaborative techniques, and content
-
based
techniques. First, compare and contrast these two approaches, and describe their
strengths and weaknesses
. Second, classify each of these four systems according
to which of the two techniques are employed, and explain your answer. Third,
discuss whether it would be possible to develop a hybrid approach that combines
both techniques.


















oOo