Artificial Intelligence for Social Issues and Sentiment Analysis

• Main Authors: Sung-Yun Tsai, 蔡松芸
• Other Authors: Jiann-Liang Chen
• Format: Others
• Language: en_US
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/ps5358

碩士 === 國立臺灣科技大學 === 電機工程系 === 107 === In this diversified society, the daily affairs of life are full of everyone's every day. In this pile of information, we need to find related issues to be searched. How can we search precisely? This study uses the method of Artificial Intelligence to classify various social issues and detect sentiments, so that it can quickly detect which issues and sentiments a large amount of data belongs to. Sentiment analysis has gained more and more attention in recent years, and its application is widely used in business and education. This research is also a preliminary study of fake news. By exploring text processing and classification, research on detecting false news can be carried out more quickly. The source of the dataset used in this study was International Workshop on Semantic Evaluation. The issues and sentiments were the same text, and the issues were Atheism, Climate Change is a Real Concern, Donald Trump, Feminist Movement, Hillary Clinton, and Legalization of Abortion. These issues are related to the public's attention, and the sentiment part is divided into positive, neutral and negative. This study uses the Long Short Term Memory (LSTM) method in the Deep Learning Algorithm. All data sets are processed in Natural Language Processing (NLP) before training, remove all unnecessary training elements, such as: blank, punctuation, meaningless words, etc., using Part of speech for Lemmatization. The text data is tokenized and assigned to each word a unique number, giving the word uniqueness for training. Training uses the LSTM algorithm. The sequential architecture allows the previous message to be passed to the back without gradient disappearance or gradient explosion. The trained model outputs six issue models and a sentiment model. Each model is selected to be the most effective of the eight possible selection models. In the study, when the training results are output, the parameters are transformed, and multiple models are trained at the same time. Finally, the optimal output model is automatically selected, which can save the time cost of manual transformation parameters. This study trains six issue models and one sentiment model. These models can help to quickly detect related issues. A text material can have multiple topics at the same time. In this study, such test results can also be achieved, and sentiment detection of text can be obtained. This research can be used in many areas such as product analysis, customer service, and customer experience on the application side.