Belief Mining in Persian Texts Based on Deep Learning and Users' Opinions (revised December 2022)

Abstract

Belief mining and the study of public opinion provide valuable information. Analyzing the feelings and belief mining of social media data leads to understanding users' opinions and has wide applications in decision making and policymaking. This article applies a new method based on deep learning to solve the problems of belief mining for Persian comments on the Twitter. In this method, first, the data is preprocessed with a deep neural network and then classified into political, cultural, economic, and sports classes, and the sentimental polarity is obtained. SentiPers is applied on four different datasets from Persian Twitter, Digikala store, Google translator, and synonym for evaluation. Then the results are compared with other machine learning and deep learning methods such as neural network, support vector machine, DNN, CNN, and LSTM. Python software has been used to implement this method. The accuracy of the proposed word embedding method for LSTM, CNN, DNN on the SentiPres dataset is 0.931, 0.923, 0.916 respectively. For the TF-IDF method, it is 0.837, 0.863, 0.883 respectively. that the accuracy of LSTM-WSD, CNN-WSD model has increased by 8% and 6% compared to TF-IDF. The results show that the LSTM and Word embedding methods work best.