Login

Publications  •  Project Statistics

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

Dissertation Information for Paul Thompson

NAME:
- Paul Thompson

DEGREE:
- Ph.D.

DISCIPLINE:
- Library and Information Science

SCHOOL:
- University of California, Berkeley (USA) (1986)

ADVISORS:
- M. E. Maron

COMMITTEE MEMBERS:
- William S. Cooper
- Lotfl A. Zadeh

MPACT Status: Fully Complete

Title: SUBJECTIVE PROBABILITY, COMBINATION OF EXPERT OPINION, AND PROBABILISTIC APPROACHES TO INFORMATION RETRIEVAL

Abstract: This dissertation is a study of probability and its application to the problem of information retrieval. It opens with a survey and examination of current work on probabilistic indexing, probabilistic queries, and rules for how these can be combined and used in order to rank output documents by computed values of probability of relevance. In these current approaches to probabilistic information retrieval (PIR), probability is interpreted as a relative frequency or as a person's estimate of a relative frequency.

There are three major conceptual difficulties with current theories, or models of PIR: (1) Independence/dependence of individual term probabilities; (2) The small sample problem; (3) The validity of probabilities used.

There is also the question of how to interpret the probability of a single case, i.e., the relevance of a specific document to a specific retrieval system user.

This dissertation divides into three main sections. One is a study of the psychological literature on human probability assessment. It is motivated by a concern with how well humans can be expected to perform in assessment of term probabilities. The second is a computer simulation study which examines how errors in estimation of term probabilities propagate into the combined probability for each document and how this affects the ranked output of documents. It is motivated by the question of determining the impact of various levels of input term probability errors on retrieval effectiveness. Finally, the third (and major section) is the development of a new mathematical model for PIR which uses a subjective interpretation of probability, probability distributions (as opposed to point estimates of probabilities), and the technique of combination of expert opinion. It is argued that such an approach shows promise of largely overcoming the obstacles to PIR mentioned above.

MPACT Scores for Paul Thompson

A = 0
C = 0
A+C = 0
T = 0
G = 0
W = 0
TD = 0
TA = 0
calculated 2008-05-29 14:05:24

Advisors and Advisees Graph