A Comprehensive Review of Multi-Domain Sentiment Analysis: Techniques, Models and Future Directions
Keywords:
multi-domain, sentiment classification, cross-domain sentiment classification, transfer learningAbstract
The increase in user-generated content on online platforms has underscored the significance of advanced sentiment analysis methods that can effectively function across various fields. Sentiment analysis, a component of natural language processing (NLP) focuses on recognising and organisi ng subjective information within text data. While there have been strides in domain-specific sentiment analysis, developing models that can excel across diverse domains remains a notable challenge. This comprehensive review of 19 research papers aims to summarise the current status of research on sentiment analysis spanning multiple domains examining the strategies, models and datasets employed to enhance performance across various areas. Techniques like domain adaptation, including transfer learning and adversarial training have shown promise by refining trained models such as BERT with target domain data thereby improving their capacity to generalise across domains. The study also sheds light on future research directions concerning domain shift, feature variations and lexical and semantic aspects to optimise multi-domain sentiment classification.
https://doi.org/10.59200/ICONIC.2024.012