ParsBERT Post-Training for Sentiment Analysis of Tweets Concerning Stock Market

Social media has become a playground for users to share their ideas freely. Analyzing these data has become of special interest to authorities and consulting firms. They seek to choose right policies based on the insight acquired. Hence, sentiment analysis of data spread in social media has gained significant importance. There are two major approaches for sentiment analysis including lexicon-based and supervised methods. Among supervised methods, deep models have proven to be a better fit for the sentiment analysis task. Since, they are domain free and able to handle large volumes of data effectively. In particular, BERT's state of the art performance on various natural language processing tasks has encouraged us to use this network architecture for sentiment analysis. In this research, over 12000 Persian tweets including the stock market keyword have been crawled from twitter. They are labeled manually in three different categories of positive, neutral and negative. Then a pre-trained ParsBERT model has been fine-tuned on these data. Our model is evaluated on the test dataset and compared to its counterpart, lexicon-based method using Polyglot as its lexicon. Accuracy of 82 percent has been achieved by our proposed model surpassing its lexicon-based contender.