Improving polarity identification in sentiment analysis using sarcasm detection and machine learning algorithms in Persian tweets

Sentiment analysis is a branch of computer science and natural language processing that seeks to familiarize machines with human emotions and make them recognizable. Both sentiment analysis and sarcasm which is a sub-field of the former, seek to correctly identify the hidden positive and negative emotions of the text. The use of sarcasm on social media, where criticism can be exercised within the context of humor, is quite common. Detection of sarcasm has a special effect on correctly recognizing the polarization of an opinion, and thus not only it can help the machine to understand the text better, but also makes it possible for the respective author to get his message across more clearly. For this purpose, 8000 Persian tweets that have emotional labels and examined for the presence or absence of sarcasm have been used. The innovation of this research is in extracting keywords from sarcastic sentences. In this research, a separate classifier has been trained to identify irony of the text. The output of this classifier is provided as an added feature to the text recognition classifier. In addition to other keywords extracted from the text, emoticons and hashtags have also been used as features. Naive Bayes, support vector machines, and neural networks were used as baseline classifiers, and finally the combination of classifiers was used to identify the feeling of the text. The results of this study show that identifying the irony in the text and using it to identify emotions increases the accuracy of the results.