DeepSentiParsBERT: A Deep Learning Model for Persian Sentiment Analysis Using ParsBERT

Abstract

Social media has provided a platform for sharing opinions and feelings on a variety of topics. Automated analysis of these opinions is of particular importance to business organizations for improving their products and services. In recent years, deep learning techniques have become very popular due to their high efficiency. Several DNN models have been proposed for the task of sentiment analysis and their performance is promising. In this paper, a new deep architecture consisting of ParsBERT and Bidirectional LSTM models (DeepSentiParsBERT) is proposed for the sentiment analysis of Persian texts. Results from comparison with the most recent state-of-the-art models show the superiority of DeepSentiParsBERT on the Digikala corpus (91.57% F1-Score).