Detection of Sentiment Provoking Events in Social Media

Abstract

Social media has become one of the main sources of news and events due to its ability to disseminate and propagate information very fast and to a large population. Social media platforms are widely accessible to the population making it difficult to extract relevant information from the huge amount of posted data. In our study, we propose an algorithm that automatically detects events using strong sentiment classification and appropriate clustering techniques. We focus our study on a specific type of events that triggers strong sentiment among the public. Results show that the suggested methodology is able to identify important events, such as a mass shooting and plane crash, using a generalized and simple approach.

Sentiment of the public: the role of social media in revealing important events

“…The first step of this study involves choosing a proper machine learning algorithm for the classification of strong sentiment. For this purpose, three different algorithms are tested: (1) supervised ML algorithm (Daou, 2019) (2) VADER (Hutto and Gilbert, 2014) and (3) SentiStrength (Thelwall et al , 2010). The supervised model uses dataset 1 described in Section 3:1 for training.…”

“…The supervised model uses dataset 1 described in Section 3:1 for training. It first handles imbalance using random repetition of samples of the minority class, then builds the sentiment classification model using support vector machine (SVM) (Daou, 2019).…”

“…In order to build the ML supervised model, the labeled dataset, SentiStrength, is used. This dataset has high imbalance of 1,102 samples for class 1 (strong sentiment) and 8,641 samples for class 0 (neutral or weak sentiment) (Daou, 2019). The issue of imbalance is resolved in order to improve the learning of the supervised algorithm.…”

“…Several classification algorithms, such as logistic regression, support vector machine (linear and radial basis function – SVM (RBF)) and Naive Bayes, are tested on the labeled dataset. Testing shows that random oversampling with SVM (RBF) gives the best classification results and therefore is chosen as the ML algorithm (Daou, 2019).…”

Understanding the Mood of Social Media Messages

“…Rather than looking for a burst of activity in terms of tweet volume, the strength of sentiment in the messages can be used. This approach has been used to automatically identify emergency events from tweets based on sentiment classification [7].…”