Sentiment Analysis in Persian Language Using Long Short-Term Memory and Deep Learning

With the growth of social networks and advancements in internet technologies, data management challenges have emerged. Sentiment analysis, as one of the important areas in Natural Language Processing (NLP), aims to extract information from user opinions. In this regard, recent studies have focused on improving the accuracy of sentiment detection using deep learning techniques and Long Short-Term Memory (LSTM). In this research, two datasets consisting of user reviews from the Digikala website and Snapfood application have been utilized. These datasets were processed and analyzed using FastText-BiLSTM-CNN and Word2Vec-BiLSTM-CNN algorithms, and the results were evaluated. Additionally, hybrid models were compared with individual models of BiLSTM and CNN. The results showed that the FastText-BiLSTM-CNN model generally performed better on both datasets compared to the Word2Vec-BiLSTM-CNN model. For instance, the accuracy of the best model, FastText-BiLSTM-CNN, was 0.9754 and 0.8765 for datasets D1 and D2, respectively. Furthermore, hybrid models outperformed individual models in terms of accuracy.