Login

Publications  •  Project Statistics

Glossary  •  Schools  •  Disciplines
People Search: 
   
Title/Abstract Search: 

Dissertation Information for Robert M. Losee

NAME:
- Robert M. Losee
- (Alias) Robert Losee

DEGREE:
- Ph.D.

DISCIPLINE:
- Library and Information Science

SCHOOL:
- University of Chicago (USA) (1986)

ADVISORS:
- Abraham Bookstein

COMMITTEE MEMBERS:
- Don R. Swanson
- Clement Tak Yu

MPACT Status: Fully Complete

Title: THE PERFORMANCE OF PROBABILISTIC MODELS OF DOCUMENT RETRIEVAL SYSTEMS

Abstract: The sequential learning model makes use of knowledge available
about probabilistic distributions describing term occurrences. Methods
for estimating initial values for parameters of the models consistent
with binary independence and two Poisson independence assumptions are
given. Use of the two Poisson independence assumptions leads to
inferior results with the MEDI033 database when compared to results of
models making binary independence assumptions. Searches using
synthesized random databases meeting the two Poisson independence and
binary independence assumptions suggest that models consistent with two
Poisson independence assumption will outperform models consistent with
binary independence assumptions, assuming all assumptions are met.
Boolean queries are converted into conjunctive normal form and
the disjunctions treated as single facets, or hyperfeatures, with
frequencies computed as the sum or maximum of the frequency of the
disjoined features. Experimental evidence is presented suggesting that
the use of this model may be inferior to using the earlier non-Boolean
model, although the increased query size may be responsible for this
poor performance.
Based on experimental evidence, it is suggested that, for
increased precision, searches use (1) the binary independence model; (2)

Croft-Harper estimates of the initial parameters for non-relevant
documents, with a high degree of confidence; (3) either ratio or equal
term discrimination value estimates of the initial parameters for
relevant documents, with a low degree of confidence; and (4) a proposed
tie breaking rule to rank documents with identical binary profiles.

MPACT Scores for Robert M. Losee

A = 2
C = 15
A+C = 17
T = 2
G = 1
W = 2
TD = 2
TA = 0
calculated 2017-05-10 22:38:48

Advisors and Advisees Graph