| Summary: | Sentiment analysis (SA) has garnered significant attention due to its application in understanding and interpreting human emotions from text data. Traditional SA models often face challenges when applied across different domains because of varying vocabulary, context, and language usage. Cross-domain SA aims to bridge this gap by transferring knowledge from one domain to another. Data distributions are different across domains and a possible solution is to learn a different system for each domain. However, it is challenging to design a resilient and cost-effective sentiment classifier. CNN has proven effective for text classification tasks, including SA. This paper explores the development and application of hybrid PSO-CNN models for cross-domain SA. The hybrid PSO-CNN model is designed to learn transferable features from source domains to target domains. Hyperparameter tuning is carefully performed to achieve optimal model performance in different domains. The hybrid PSO-CNN model demonstrates significant improvements in hyperparameter tuning and overall performance in the target domain. The model achieves a high accuracy of about 98.5% for the domain Book to DVD. The findings highlight the effectiveness of the hybrid PSO-CNN model in cross-domain sentiment analysis. Future research may explore advanced domain adaptation techniques and additional bio-inspired algorithms to further enhance model performance. © 2024 IEEE.
|
|---|