Improved text classification through label clustering

Honors College Thesis

Public Deposited

Citeable URL: https://ir.library.oregonstate.edu/concern/honors_college_theses/zw12z723q

Descriptions

Attribute Name	Values
Creator	Vanderschuere, Christopher
Abstract	This paper introduces an approach to text classification for semi-structured label systems that have poor performance with standard methods. With the perspective that perfect classification for such a system is unattainable, we demonstrate an automated procedure to isolate the learnable elements of the problem. Through analysis of an example dataset, we identify attributes of the label system that hinder performance and demonstrate through manual methods that minimizing these attributes will lead to improved performance. Further we present that label clustering effectively minimizes these attributes. We then show that with a combination of frequency, co-occurrence, and document similarity we are able to construct label clusters in an automated fashion. Finally, we demonstrate that by using label clusters we are able to improve classification performance without excessively limiting the label space.
License	All rights reserved
Resource Type	Honors College Thesis
Date Available	2015-06-10T13:31:08+00:00
Date Issued	2015-06-02
Degree Level	Bachelor's
Degree Name	Honors Bachelor of Science (H.B.S.)
Degree Field	Electrical and Computer Engineering
Degree Grantor	Oregon State University
Commencement Year	2015
Advisor	Fern, Xiaoli
Non-Academic Affiliation	Oregon State University. Honors College
Rights Statement	In Copyright
Publisher	Oregon State University
Peer Reviewed	No
Language	English [eng]
Replaces	http://hdl.handle.net/1957/56076

Parents:

This work has no parents.

In Collection:

Thumbnail	Title	Date Uploaded	Visibility	Actions
	final-draft.pdf	2017-07-29	Public	Download