Opinion mining and sentiment analysis for terrorist reviews on social media

Sentiment analysis has been used to bring out the emotions of people on the things they talk or write about. It is a text mining method, which is made by calculating the frequencies of positive and negative sentiment terms that make up the sentence. In this research, we argue that the frequency of negative and positive values of text units do not say enough about their content in some domains, for example exploring the sentiment of reactions after a terrorist attack. In this thesis, we proposed a model which is a hybrid between Sentiment Analysis, Topic modeling and Fuzzy logic system. This model has been evaluated on reactions about two almost identical terrorist attacks that occurred in London and Barcelona in 2017. A dataset of reactions to these events was collected from Twitter. First, a preliminary analysis of the tweet content was performed to understand how the semantics of tweet content relates to each of the attacks. Then the main topics of the reactions were extracted using the LDA topic modeling algorithm with human judgment. The lexicon approach was used to score the opinion words in the tweets. All opinion words from the tweets were identified by matching them with the words in the opinion lexicon. Then, a fuzzy logic system was used to determine polarity and label the tweet for sentimental classification. Features such as POS and N-grams were extracted. Finally, different machine learning methods were used to classify the sentiment of the tweets. According to the results obtained in this research, it was concluded that the topics from people's reactions are relevant and it can help to significantly predict the real emotions in the text.