Sentiment Analysis of Persian Political Tweets Using ParsBERT Embedding Model with Convolutional Neural Network

Recently, social networks have experienced an exponential growth, providing online users with a venue for expressing and sharing their views in a variety of areas. A very popular social media platform for getting user feedback and collecting data is Twitter. Sentiment analysis consists of extracting and analyzing opinions of people and can be used to predict tweet polarities. In this paper, we present a sentiment analysis methodology for Persian political tweets for the first time in order to assist Iranian politicians. We used two datasets of Persian political tweets with three and seven classes which are labeled according their content. This is the first study of this particular subfield of Persian tweets, so we are using a variety of encoding methods, including Bag-of-words, Word Embeddings, as well as neural methods such as Word2Vec, FastText, and ParsBERT Embeddings. We intend to use these techniques to implement Sentiment Analysis in closed domain political tweets using Machine Learning techniques such as Random Forests, Support Vector Machines, and Neural Networks. As a result of our comparisons, we found that CNN+BiLSTM using ParsBERT embeddings had higher robustness than other networks, scoring 0.89 on Dataset 1 and 0.71 on Dataset 2.