Dissertation Information for Dinesh Rathi NAME: - Dinesh Rathi
DEGREE:
- Ph.D.
DISCIPLINE:
- Library and Information Science
SCHOOL:
- University of Illinois, Urbana-Champaign (USA) (2008)
ADVISORS: - Michael B. Twidale
COMMITTEE MEMBERS: - Linda C. Smith - P. Bryan Heidorn - Chen Xiang Zhai
MPACT Status: Fully Complete
Title: Mining Help Desk Emails for Problem Domain Identification and Email Feature Engineering for Routing Incoming Emails
Abstract: Text mining typically involves a variety of decisions to be made about the use of the techniques for mining, including data type, algorithm selection, pre-processing, intermediate processing and parameter setting. Choices within parameter space are mostly regarded as judgment calls made by developers or expert users of the text mining software. This thesis explores the parameter space and associated processing decisions in the context of text mining of emails for technical help. This is a relatively unexplored area, but one of growing economic importance. In doing so it reveals issues relevant to the particular use context, to other applications of text mining of emails, and implications for investigations of other parameter spaces.
In this research work, emails from the help desk domain were chosen as the dataset on which text mining techniques were applied to investigate the impact of different features of email on problem domain identification and routing emails.
The study investigates the impact of different features of email on problem domain identification by using machine-expert approach. The clustering technique (machine) was used to generate a number of clusters and the knowledge of the experts from the help desk domain was used to label the clusters to identify the problem domains in the help desk. This machine-expert approach was also used to study the impact of different features of an email (such as header, content, reply and signature) on clustering of these emails and identification of problem domains in the help desk area. The Expectation Maximization algorithm was used to generate different numbers of clusters from email data.
This study investigated the impact of different features of an email i.e., header, content, reply and signature on the supervised and unsupervised classification of emails to different problem domains. A classification algorithm based on Naïve Bayes was used for routing of emails into different problem domains.
This work also outlines different pre-processing steps done on email from the help desk domain, such as removal of different types of noise elements and removal of stop words using a customized stop word list, and the outcome of each of these pre-processing steps.
| |
MPACT Scores for Dinesh Rathi A = 0
C = 0
A+C = 0
T = 0
G = 0
W = 0
TD = 0
TA = 0
calculated 2009-07-26 21:00:04
Advisors and Advisees Graph
generating graph, please reload |