Abstract
dc:description.abstractIn this work, we explore different techniques to extract opinions, sentiments and beliefs from text. In particular, we look at neural network based approaches since deep learning has gained wide popularity nowadays and is known to perform effectively for the kind of problems we are looking at. The first goal was to solve the BeSt (Belief and Sentiment) Evaluation task from the 2016 Text Analysis Conference. Here we used bidirectional Long Short Term Memory networks (LSTMs) for sentiment detection in a given document. We looked with particular interest at the sentiment polarity to find out whether it is positive, negative or neutral sentiment. Our neural network based model consists of a multi-layered bidirectional LSTM which takes as input embeddings for the sequence of words in a given sentence along with annotated source and target, and outputs probabilities for the different sentiment polarities. We evaluated the model using standard precision, recall and F1 scores and then compared our results to scores in the existing literature. The second goal was to identify opinion expressions along with their sources and targets in a given text. We compared a simple baseline approach, support vector machine and conditional random field to a feedforward neural network and a bidirectional LSTM approach. All approaches were evaluated on a standard dataset from the sentiment analysis literature - the MPQA dataset.
Degree
thesis:*- Name thesis:degree_name
- M.S., Computer Science
- Level thesis:degree_level
- Master of Science
- Discipline thesis:degree_discipline
- Computer Science
- Grantor
- Cornell University
- Year dc:date.issued
- 2018
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- CHOWDHURY, ANUSHA
- Chair dc:contributor.chair
-
- Cardie, Claire T.
- Committee member dc:contributor.committeemember
-
- Mimno, David
Subjects
dc:subject × 8Rights
- Language dc:language.iso
- en_US
Identifiers
dc:identifier.*- Dc Identifier Bibid
- 10489407
- Dc Identifier Other
-
ProQuest Submission ID: 10280
ProQuest Publication ID: 10815180