Emotion Detection from Text in Social Networks

Emotion detection from text in social networks is an important tool for monitoring and analyzing discussions on social networks. However, the difficulty of the problem related to emotional analysis from texts is caused by the lack of additional features such as facial expressions or tone of voice, and the use of colloquial language and various signs and symbols in the texts contributes to misinterpretation of the content of the statement. For this reason, this article proposes an approach to automatically detect emotions from real text data from social networking sites and classify them into appropriate emotion categories. The aim of the article is to develop a model for detecting and classifying emotions through the use of natural language processing techniques and machine learning algorithms. In addition, it was decided to analyze the impact of the number of types of emotions on the classification results. Our research was tested for six classifiers (CART, SVM, AdaBoost, Bagging, Random Forest and K-NN) and three word weighting measures (TF, TF-IDF and Binary). Four measures of classification quality were used to evaluate the classifiers, i.e. accuracy, precision, recall and F1-score. Two datasets from different social networking sites were used for the research - one is a collection of comments from a social networking site and the other is tweets from Twitter. The analysis of the results confirmed that the proposed solution allows detecting emotions from the actual content of tweets, and the results obtained are better for a smaller number of emotion categories.