Dissertation Information for Daniel Patterson Dabney NAME: - Daniel Patterson Dabney
DEGREE:
- Ph.D.
DISCIPLINE:
- Library and Information Science
SCHOOL:
- University of California, Berkeley (USA) (1993)
ADVISORS: - William S. Cooper
COMMITTEE MEMBERS: - Ray R. Larson - Robert C. Berring
MPACT Status: Fully Complete
Title: Statistical modeling of relevance judgments for probabilistic retrieval of American case law
Abstract: This study investigates the use of models derived by logistic regression to overcome some of the difficulties encountered in the implementation of probabilistic document retrieval systems. It is often desirable for the ranking function of such a system to consider several different relevance clues. Many of these clues are not initially expressed in probabilistic terms, and cannot plausibly be assumed to satisfy the linked dependence assumptions required by classic models of probabilistic retrieval system design. A model derived by logistic regression can be used to combine such problematic clues into a single estimate of the relevance of candidate documents. The estimate may be used as the system's ranking function, or may be refined by applying likelihood ratios calculated from other relevance clues that are not problematic.
The study illustrates the application of this method by using a logistic regression model to meld a variety of highly-dependent relevance clues into a single probability estimate. The clues considered are several different coefficients of association taken from a conventional vector-space retrieval model. The modeled probability estimate ranked system output more usefully than did any of the individual clues. In addition, the process of building the model shed light on the usefulness of individual clues, and identified a subset of the clues that contain nearly all of the non-redundant relevance information in the larger set.
The study then considers a more complex subject domain: case-finding in American law. The case finding problem is considered in historical context, and the performance of commercial Boolean text-retrieval systems is evaluated. The study then shows how the logistic regression method could be applied to a wide variety of relevance clues available for case-finding, including the intellectual indexing embodied in the West Key Number System, citation vectors, text words, case pedigrees, and relative citation frequency. The study concludes by suggesting how the system might be applied in other subject domains.
| |
MPACT Scores for Daniel Patterson Dabney A = 0
C = 0
A+C = 0
T = 0
G = 0
W = 0
TD = 0
TA = 0
calculated 2008-01-31 06:14:17
Advisors and Advisees Graph
generating graph, please reload |