Abstract
dc:description.abstractIn computational linguistics, most sentiment analysis builds binary classification models on customer reviews data to predict whether a review is positive or negative. In this thesis, we go a step further and build interpretable classification models to predict fine-grained emotions associated with text (such as happy, sad, productive and tired). This analysis is enabled by a unique journaling dataset containing short pieces of text and associated emotional status self-reported by writers. To further study what people feel emotional about (emotion triggers), we perform model interpretation. We make two main contributions. First, we apply state-of-the-art text mining methodologies to extract emotion triggers from text, during which we discover and solve an issue with the attention mechanism in a popular deep learning model (Dynamic Memory Network). Second, we obtain data-driven evidence of emotion triggers based on a group of 67,000 people, which contributes to a better understanding of emotion triggers from the perspective of public mental health.
Degree
thesis:*- Grantor dc:publisher
- University of Waterloo
- Year dc:date.issued
- 2019
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Chen, Liuyan
- Advisor dc:contributor.advisor
-
- Golab, Lukasz
Rights
- Language dc:language.iso
- en
Identifiers
dc:identifier.*- Handle dc:identifier.uri
- http://hdl.handle.net/10012/14636