Abstract:
"This study investigates enhancing sentiment analysis in product reviews by integrating insights
from mental health discussions. While Aspect-Based Sentiment Analysis (ABSA) aims to
associate sentiments with specific product features, it encounters limitations in adapting to
diverse domains. Leveraging transfer learning, this research bridges the gap between mental
health insights and product sentiment analysis. Analysing mental health-related text offers rich
insights into emotional expressions and linguistic nuances, potentially enriching sentiment
analysis in product reviews.
The research aims to design an adaptable sentiment analysis model and contribute to transfer
learning in Natural Language Processing (NLP) by exploring innovative approaches.
Addressing gaps in existing literature, including limited exploration of mental health aspects,
transferability of insights, evaluation metrics, model generalization, and fine-tuning strategies,
the study seeks to refine sentiment analysis methodologies and establish a framework for
broader transfer learning applications.
The XLNet model displayed strong performance in sentiment analysis but yielded mixed
results in aspect analysis. In terms of sentiment analysis, the model demonstrated impressive
metrics, achieving an accuracy, F1 score, precision, and recall of 96.8%. The confusion matrix
revealed effective classification of sentiments with minimal misclassifications. In contrast, the
aspect analysis resulted in an accuracy of 76%, an F1 score of 75.3%, precision of 77.2%, and
recall of 76%. The confusion matrix underscored challenges in accurately classifying the
different aspects, indicating the need for further refinement to improve performance across all
aspect categories. While the sentiment analysis proved robust, additional optimization and data
enhancement are necessary for better aspect classification."
