<?xml version="1.0" encoding="UTF-8"?><?xml-stylesheet type="text/xsl" href="static/style.xsl"?><OAI-PMH xmlns="http://www.openarchives.org/OAI/2.0/" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xsi:schemaLocation="http://www.openarchives.org/OAI/2.0/ http://www.openarchives.org/OAI/2.0/OAI-PMH.xsd"><responseDate>2026-11-22T08:22:04Z</responseDate><request verb="GetRecord" identifier="oai:uwspace.uwaterloo.ca:10012/3250" metadataPrefix="dim">https://dspacemainprd01.lib.uwaterloo.ca/server/oai/request</request><GetRecord><record><header><identifier>oai:uwspace.uwaterloo.ca:10012/3250</identifier><datestamp>2017-03-21T16:01:48Z</datestamp><setSpec>com_10012_9900</setSpec><setSpec>com_10012_1</setSpec><setSpec>col_10012_9908</setSpec><setSpec>col_10012_6</setSpec></header><metadata><dim:dim xmlns:dim="http://www.dspace.org/xmlns/dspace/dim" xmlns:doc="http://www.lyncode.com/xoai" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xsi:schemaLocation="http://www.dspace.org/xmlns/dspace/dim http://www.dspace.org/schema/dim.xsd">
   <dim:field mdschema="dc" element="contributor" qualifier="author" authority="d7df29a0-8506-4b14-a24d-9a9a0d7e163d" confidence="600">Makrehchi, Masoud</dim:field>
   <dim:field mdschema="dc" element="date" qualifier="accessioned">2007-09-18T17:42:45Z</dim:field>
   <dim:field mdschema="dc" element="date" qualifier="available">2007-09-18T17:42:45Z</dim:field>
   <dim:field mdschema="dc" element="date" qualifier="issued">2007-09-18T17:42:45Z</dim:field>
   <dim:field mdschema="dc" element="date" qualifier="submitted">2007</dim:field>
   <dim:field mdschema="dc" element="identifier" qualifier="uri">http://hdl.handle.net/10012/3250</dim:field>
   <dim:field mdschema="dc" element="description" qualifier="abstract" lang="en">Feature selection based on feature ranking has received much&#xd;
attention by researchers in the field of text classification. The&#xd;
major reasons are their scalability, ease of use, and fast computation. %,&#xd;
However, compared to the search-based feature selection methods such&#xd;
as wrappers and filters, they suffer from poor performance. This is&#xd;
linked to their major deficiencies, including: (i) feature ranking&#xd;
is problem-dependent; (ii) they ignore term dependencies, including&#xd;
redundancies and correlation; and (iii) they usually fail in&#xd;
unbalanced data.&#xd;
&#xd;
&#xd;
While using feature ranking methods for dimensionality reduction, we&#xd;
should be aware of these drawbacks, which arise from the function of&#xd;
feature ranking methods.  In this thesis, a set of solutions is&#xd;
proposed to handle the drawbacks of feature ranking and boost their&#xd;
performance. First, an evaluation framework called feature&#xd;
meta-ranking is proposed to evaluate ranking measures. The framework&#xd;
is based on a newly proposed Differential Filter Level Performance&#xd;
(DFLP) measure. It was proved that, in ideal cases, the performance&#xd;
of text classifier is a monotonic, non-decreasing function of the&#xd;
number of features. Then we theoretically and empirically validate&#xd;
the effectiveness of DFLP as a meta-ranking measure to evaluate and&#xd;
compare feature ranking methods. The meta-ranking framework is also&#xd;
examined by a stopword extraction problem. We use the framework to&#xd;
select appropriate feature ranking measure for building&#xd;
domain-specific stoplists. The proposed framework is evaluated by&#xd;
SVM and Rocchio text classifiers on six benchmark data. The&#xd;
meta-ranking method suggests that in searching for a proper feature&#xd;
ranking measure, the backward feature ranking is as important as the&#xd;
forward one.&#xd;
&#xd;
Second, we show that the destructive effect of term redundancy  gets&#xd;
worse as we decrease the feature ranking threshold. It implies that&#xd;
for aggressive feature selection, an effective redundancy reduction&#xd;
should be performed as well as  feature ranking. An algorithm based&#xd;
on extracting term dependency links using an information theoretic&#xd;
inclusion index is proposed to detect and handle term dependencies.&#xd;
The dependency links are visualized by a tree structure called a&#xd;
term dependency tree. By grouping the nodes of the tree into two&#xd;
categories, including hub and link nodes, a heuristic algorithm is&#xd;
proposed to handle the term dependencies by merging or removing the&#xd;
link nodes. The proposed method of redundancy reduction is evaluated&#xd;
by SVM and Rocchio classifiers for four benchmark data sets.&#xd;
According to the results, redundancy reduction is more effective on&#xd;
weak classifiers since they are more sensitive to term redundancies.&#xd;
It also suggests that in those feature ranking methods which compact&#xd;
the information in a small number of features, aggressive feature&#xd;
selection is not recommended.&#xd;
&#xd;
Finally, to deal with class imbalance in feature level using ranking&#xd;
methods, a local feature ranking scheme called reverse&#xd;
discrimination approach is proposed. The proposed method is applied&#xd;
to a highly unbalanced social network discovery problem.  In this&#xd;
case study, the problem of learning a social network is translated&#xd;
into a text classification problem using newly proposed actor and&#xd;
relationship modeling. Since social networks are usually sparse&#xd;
structures, the corresponding text classifiers  become highly&#xd;
unbalanced. Experimental assessment of the reverse discrimination&#xd;
approach validates the effectiveness of the local feature ranking&#xd;
method to improve the classifier performance when dealing with&#xd;
unbalanced data.  The application itself suggests a new approach to&#xd;
learn social structures from textual data.</dim:field>
   <dim:field mdschema="dc" element="language" qualifier="iso" lang="en">en</dim:field>
   <dim:field mdschema="dc" element="publisher" lang="en">University of Waterloo</dim:field>
   <dim:field mdschema="dc" element="subject" lang="en">Feature Ranking</dim:field>
   <dim:field mdschema="dc" element="subject" lang="en">Feature Selection</dim:field>
   <dim:field mdschema="dc" element="subject" lang="en">Information Theory</dim:field>
   <dim:field mdschema="dc" element="subject" lang="en">Text Classifier</dim:field>
   <dim:field mdschema="dc" element="subject" lang="en">Social Network</dim:field>
   <dim:field mdschema="dc" element="subject" lang="en">Link Mining</dim:field>
   <dim:field mdschema="dc" element="subject" qualifier="program" lang="en">Electrical and Computer Engineering</dim:field>
   <dim:field mdschema="dc" element="title" lang="en">Feature Ranking for Text Classifiers</dim:field>
   <dim:field mdschema="dc" element="type" lang="en">Doctoral Thesis</dim:field>
   <dim:field mdschema="dc" element="pending" lang="en">false</dim:field>
   <dim:field mdschema="uws-etd" element="degree" qualifier="department" lang="en">Electrical and Computer Engineering</dim:field>
   <dim:field mdschema="uws-etd" element="degree" lang="en">Doctor of Philosophy</dim:field>
   <dim:field mdschema="uws" element="typeOfResource" lang="en">Text</dim:field>
   <dim:field mdschema="uws" element="peerReviewStatus" lang="en">Unreviewed</dim:field>
   <dim:field mdschema="uws" element="scholarLevel" lang="en">Graduate</dim:field>
   <dim:field mdschema="others" element="access-status">open.access</dim:field>
</dim:dim>
</metadata></record></GetRecord></OAI-PMH>