- 213 views
- 195 downloads
Polarity Classification of Reviews by Expanding a Preliminary Lexicon
-
- Author / Creator
- Yadollahi, Ali
-
Polarity classification in text is the problem of automatically detecting the general opinion
of textual data. Analyzing the general opinion toward a topic of interest is important for
different audiences, such as companies, politicians or even regular users. On the other
hand, the availability of opinionated text data has been rapidly increasing in WWW during
the past decade. However, most of the available text data is not provided with a polarity
label and hence one may not be able to learn from it. Addressing this challenge shapes the
main motivation of the research in this field of study.
Previous works in polarity classification suffer from some drawbacks. As the major
drawback, most of the research in the literature naively assumes the availability of the la
belled data and takes the advantage of supervising their method by the given ground truth.
Furthermore, they do not generally perform well in all domains of textual data, since they
are specifically trained over one domain. Although there exists a limited number of works
proposing an adaptable classification method, their performance is not promising on the
unlabelled data.
This thesis tackles the problem of multi-domain unsupervised polarity classification.
We propose a hybrid polarity classification technique, which is the combination of a lexical
and a learning classification. We also propose a lexicon expansion method, which uses
the hybrid classification along with the association rule mining at the sentence level. Our
experiments on various datasets show that our hybrid classification method outperforms the
previous state-of-the-art unsupervised works. Furthermore, our expansion method is shown
to increase the performance of the initial lexicon. -
- Subjects / Keywords
-
- Graduation date
- Fall 2015
-
- Type of Item
- Thesis
-
- Degree
- Master of Science
-
- License
- This thesis is made available by the University of Alberta Libraries with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.