Automatic Emotion Identification from Text

摘要: Wang, Wenbo. Ph.D., Department of Computer Science and Engineering, Wright State University, 2015. Automatic Emotion Identification from Text. Peoples emotions can be gleaned their text using machine learning techniques to build models that exploit large self-labeled emotion data social media. Further, the effectively adapted train classifiers in different target domains where training are sparse. Emotions both prevalent essential most aspects our lives. They influence decision-making, affect relationships shape daily behavior. With rapid growth emotion-rich textual content, such as microblog posts, blog forum discussions, there is a growing need develop algorithms for identifying people’s expressed text. It has valuable implications studies suicide prevention, employee productivity, well-being people, customer relationship management, etc. However, identification quite challenging partly due following reasons: i) multi-class classification problem usually involves at least six basic emotions. Text describing an event or situation causes devoid explicit emotion-bearing words, thus distinction between very subtle, which makes it difficult glean purely by keywords. ii) Manual annotation human experts labor-intensive errorprone. iii) Existing labeled datasets relatively small, fails provide comprehensive coverage emotion-triggering events situations. This dissertation aims understanding developing general tackle above challenges. First, address challenge fine-grained classification, we investigate variety lexical, syntactic,