Please use this identifier to cite or link to this item:
Title: Public opinion trend analysis on nuclear energy via aspect-based sentiment analysis methods
Authors: Huang, Fan
Keywords: Library and information science
Issue Date: 2022
Publisher: Nanyang Technological University
Source: Huang, F. (2022). Public opinion trend analysis on nuclear energy via aspect-based sentiment analysis methods. Master's thesis, Nanyang Technological University, Singapore.
Abstract: As online social media platforms like Twitter prevail in recent years, many public ideas and opinions are generating and spreading around the virtual world every minute. When people are more willing to share and discuss their thoughts on online platforms, monitoring and analyzing the trends of public opinions towards heated topics are valuable for both platforms themselves and related government departments. The subjective emotions and sentiment information within textual data present people's thoughts and attitudes towards specific themes. With the help of the fine-grained Aspect-Based Sentiment Analysis (ABSA) method using Deep Learning models, the public opinions towards specific aspects within massive data could be understood and predicted more accurately. The Bidirectional Encoder Representation Transformers (BERT) model was applied to detect the sentiment polarities of collected tweets. We then used the trained BERT Model to predict the sentiment polarities of the tweets collected in three different periods with massive data. We applied the Latent Dirichlet Allocation (LDA) based Topic Modelling method to confirm the reliability aspect categories with detailed keyword sets in the data preprocessing process. After that, with the help of the aspect-based sentiment polarity prediction results supported by ABSA Deep Learning models, we analyze public opinion trends on different aspects through time changes. We also introduced the statistical parameter NuclearOpinionIndex (NOI) to indicate the general sentiment polarity in massive data, providing a clearer image of public opinion trends. The collection of datasets is about the public discussion on nuclear energy usage tweets like Fukushima nuclear leakage related News events. The three datasets were collected separately in 2011, 2018 and 2021. The first part of this paper was to build one solid aspect-based sentiment analysis model with high accuracy. The second part was preprocessing the three collected datasets and applying the category auto labelling algorithm to generate tweets with related aspect categories. At last, we analyzed the sentiment polarity prediction results of three datasets to present aspect-based public opinion trends on the nuclear energy-related tweets from 2011 to 2021. As the contribution of this study, firstly, we introduced the Aspect-Based Sentiment Analysis (ABSA) BERT model trained on the well revised nuclear energy dataset. Secondly, we built one automatic aspect category labelling algorithm based on well-revised keywords specifically for nuclear energy related tweets. Thirdly, we applied the ABSA BERT model to analyze the sentiment polarities on massive datasets. The results were used to conclude public opinion trend changes through time. In general, we built a system based on the ABSA Deep Learning model to make fine-grained and topic-specific sentiment analyses on public opinion trends. The study's limitations are the imperfect understanding of neutral sentiments, the possible wrong predictions of the ABSA BERT model, and possible automatic aspect labelling algorithm errors. Those above would lead to bias, indicating that the final trend analysis result may not perfectly present the actual public opinion trends. However, the massive amount of collected tweets in all three datasets could then largely avoid the biases caused by the limitations.
Fulltext Permission: restricted
Fulltext Availability: With Fulltext
Appears in Collections:WKWSCI Theses

Files in This Item:
File Description SizeFormat 
Public Opinion Trend Analysis on Nuclear Energy via Aspect-Based Sentiment Analysis Methods.pdf
  Restricted Access
Thesis with Declaration Statements716.81 kBAdobe PDFView/Open

Page view(s)

Updated on Jul 2, 2022


Updated on Jul 2, 2022

Google ScholarTM


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