Please use this identifier to cite or link to this item:
Title: Machine learning technicques for aspect based sentiment analysis
Authors: Chen, Haonan
Keywords: DRNTU::Engineering::Electrical and electronic engineering
Issue Date: 2017
Abstract: Aspect Based Sentiment Analysis (ABSA) is a field of study where sentiments on certain aspects or characteristics of entities are obtained, analyzed, and aggregated from text. Since ABSA facilitates analyzing sentiments at a fine-grained level, it has gained significant attention over the past few years. In particular, ABSA caters real-world applications such analyzing sentiments from product reviews, tweets and emails focusing solely on user intended aspects. In recent research, ABSA has predominantly leveraged on Machine Learning (ML) techniques such as Representation Learning, Kernel Methods and Deep Learning. This dissertation focuses on empirically evaluating two recently proposed ML techniques for ABSA – Support Vector Machine (SVM)-based and Recurrent Neural Networks (RNNs)-based approaches on several datasets. To this end, we re-implemented two state-of-the-art ABSA frameworks which use these classifiers and compared them on the aforementioned ABSA datasets in terms of both accuracy and efficiency. Through these large-scale evaluations, we infer the following: (i) SVMs produce accuracies which are comparable to that of RNNs but they are much computationally lighter, (ii) When there is a significant imbalance among the classes in a multi-class ABSA setting, RNNs perform much better than SVMs. Furthermore, we observe that RNN, as it leverages on word embeddings, are particularly more suited for semantics based ABSA. However, due to its huge computational demands (e.g., large number of GPU cores and high GPU memory), we could not explore the full realm of its performance on our experimental setup which had limited computing resources.
Schools: School of Electrical and Electronic Engineering 
Fulltext Permission: restricted
Fulltext Availability: With Fulltext
Appears in Collections:EEE Theses

Files in This Item:
File Description SizeFormat 
  Restricted Access
2.28 MBAdobe PDFView/Open

Page view(s) 50

Updated on Jun 13, 2024


Updated on Jun 13, 2024

Google ScholarTM


Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.