Date of Award

4-2013

Document Type

Thesis

Degree Name

Doctor of Philosophy (PhD)

Department

Computer Science

First Advisor

Qiuming Zhu

Second Advisor

Ann Fruhling

Third Advisor

Dora Matache

Fourth Advisor

Kenneth Kriz

Abstract

Predicting the outcome of uncertain events that will happen in the future is a frequently indulged task by humans while making critical decisions. The process underlying this prediction and decision making is called information aggregation, which deals with collating the opinions of different people, over time, about the future event’s possible outcome. The information aggregation problem is non-trivial as the information related to future events is distributed spatially and temporally, the information gets changed dynamically as related events happen, and, finally, people’s opinions about events’ outcomes depends on the information they have access to and the mechanism they use to form opinions from that information. This thesis addresses the problem of distributed information aggregation by building computational models and algorithms for different aspects of information aggregation so that the most likely outcome of future events can be predicted with utmost accuracy. We have employed a commonly used market-based framework called a prediction market to formally analyze the process of information aggregation. The behavior of humans performing information aggregation within a prediction market is implemented using software agents which employ sophisticated algorithms to perform complex calculations on behalf of the humans, to aggregate information efficiently. We have considered five different yet crucial problems related to information aggregation, which include: (i) the effect of variations in the parameters of the information being aggregated, such as its reliability, availability, accessibility, etc., on the predicted outcome of the event, (ii) improving the prediction accuracy by having each human (software-agent) build a more accurate model of other humans’ behavior in the prediction market, (iii) identifying how various market parameters effect its dynamics and accuracy, (iv) applying information aggregation to the domain of distributed sensor information fusion, and, (v) aggregating information on an event while considering dissimilar, but closely-related events in different prediction markets. We have verified all of our proposed techniques through analytical results and experiments while using commercially available data from real prediction markets within a simulated, multi-agent based prediction market. Our results show that our proposed techniques for information aggregation perform more efficiently or comparably with existing techniques for information aggregation using prediction markets.

Comments

A DISSERTATION Presented to the Faculty of The Graduate College at the University of Nebraska In partial Fulfillment of Requirements for the Degree of Doctor of Philosophy. Copyright 2013 Janyl Jumadinova.

Files over 3MB may be slow to open. For best results, right-click and select "save as..."

COinS