The IndRNN capsule approach for persian multi-domain sentiment analysis

Sentiment Analysis (SA) is the computational analysis of ideas, feelings, and opinions using Natural Language Processing (NLP) techniques, computational methods, and text analysis to extract polarity (positive, negative, or neutral) from unstructured documents or textual comments. Multi-domain SA is based on a labeled dataset, which reduces the dependence on large amounts of domain-specific data and addresses data scarcity issues by leveraging existing data from other domains. This paper presents a novel deep learning-based approach for Persian multi-domain SA analysis. The proposed Bidirectional Independent Recurrent Neural Network (Bi-IndRNN) Capsule technique combines bidirectional IndRNN and CapsuleNet, which use Bi-GRU to extract features for CapsuleNet. In IndRNN, recurrent layer neurons operate independently, with simple RNN computing the hidden state h via element-wise vector multiplication u* state, indicating that each neuron has a solitary recurrent weight linking it to the most recent hidden state. We evaluated the proposed approach on the Digikala dataset and found it to provide acceptable accuracy compared to existing techniques.