Enhanced Sentiment Classification in Game Reviews with Sarcasm Detection Using NLP and XAI

Abstract

Growth in online gaming, the number of reviews for every game has grown exponentially. Because of this, it has actually become tough for developers and analysts to find out the accurate sentiment of the users. Sentiment analysis is normally applied to textual data; however, state-of-the-art models have failed in understanding subtle cues in text-sarcasm and context in gaming reviews. It detects these lacunas and thus proposes a hybrid model for the detection of sentiment and sarcasm, leveraging BERT with an added layer for capturing subtle language cues related to sarcasm. Based on a dataset of gaming reviews from the Steam gaming platform, this model leverages high-level NLP software techniques for transparency, enhancing interpretability via Explainable AI. The experimental results prove the effectiveness of the proposed approach to yield high accuracy and robust performance in detecting both sentiments and sarcasms, thus giving valuable insights to developers and stakeholders.