Abstract
Deep neural networks have shown good data modelling capabilities when dealingwith challenging and large datasets from a wide range of application areas.Convolutional Neural Networks (CNNs) offer advantages in selecting goodfeatures and Long Short-Term Memory (LSTM) networks have proven good abilitiesof learning sequential data. Both approaches have been reported to provideimproved results in areas such image processing, voice recognition, languagetranslation and other Natural Language Processing (NLP) tasks. Sentimentclassification for short text messages from Twitter is a challenging task, andthe complexity increases for Arabic language sentiment classification tasksbecause Arabic is a rich language in morphology. In addition, the availabilityof accurate pre-processing tools for Arabic is another current limitation,along with limited research available in this area. In this paper, weinvestigate the benefits of integrating CNNs and LSTMs and report obtainedimproved accuracy for Arabic sentiment analysis on different datasets.Additionally, we seek to consider the morphological diversity of particularArabic words by using different sentiment classification levels.