Vahid Mottaghi - Department of Computer Engineering, Technical and Vocational University,(TVU), Tehran, Iran
Hamed AfsharFarnia - Department of Computer Engineering, Technical and Vocational University,(TVU), Tehran, Iran

Sentiment analysis is a computational analysis of ideas, feelings and opinions, and uses natural language processing techniques, computational techniques and text analyses to extract polarity from non-structured documents or textual comments. the purpose of the multi-domain SA is that the classifier training is based on a set of tagged data in a way that reduces the need for large amounts of data on specific domains and to address the challenges of data scarcity in them with the help of existing data on other domains. The purpose of this paper is to present a new method for analyzing the multi-domain SA using deep learning approaches. Bi-GRUCapsule approach uses the combination of two networks Bi-directional GRU and CapsuleNet to solve the multi-domain SA problem that Bi-GRU has the role of extracting features for CapsuleNet. the proposed approach was evaluated using the Dranziera protocol and has received acceptable accuracy compared to the existing approaches

Multi-domain sentiment analysis, Sentiment analysis, Bidirectional GRU, Capsule network, Deep learning, Natural Language Processing