A systematic assessment of sentiment analysis models on iraqi dialect-based texts

Social media allows individuals, groups, and companies to openly express their opinions, creating a rich resource for trend assessments through sentiment analysis. Sentiment Analysis (SA) uses natural language processing (NLP) to interpret these opinions from text. However, Arabic sentiment analysis faces challenges due to dialect variations, limited resources, and hidden sentiment words. This study proposes hybrid models combining Convolutional Neural Networks with Long Short-Term Memory called as CNN-LSTM, CNN with Gated Recurrent Unit called as CNN-GRU. and AraBERT, a deep transformer model, to enhance Iraqi sentiment analysis. These models were evaluated against various machine learning and deep learning models. For feature extraction, we utilized Continuous Bag of Words (CBOW) for deep learning models and BERT for the AraBERT model, while TF-IDF was used for machine learning models. According to the experimental results, the AraBERT model has been able to achieve superior performance and significantly improve the accuracy of sentiment analysis in case of Iraqi dialect-based texts.