SenticNet 8: Fusing Emotion AI and Commonsense AI for Interpretable, Trustworthy, and Explainable Affective Computing

ChatGPT has stunned the world with its ability to generate detailed, original, and accurate responses to prompts. While it unlocked solutions to problems that were previously considered unsolvable, however, it also introduced new ones. One of such problems is the phenomenon known as hallucination, the generation of content that is nonsensical or unfaithful to the provided source content. In this work, we propose SenticNet 8, a neurosymbolic AI framework leveraging an ensemble of commonsense knowledge representation and hierarchical attention networks, which aims to mitigate some of these issues in the context of affective computing. In particular, we focus on the tasks of sentiment analysis, personality prediction, and suicidal ideation detection. Results show that SenticNet 8 presents superior accuracy with respect to all four baselines, namely: bag-of-words, word2vec, RoBERTa, and ChatGPT. Unlike these baselines, moreover, SenticNet 8 is also fully interpretable, trustworthy, and explainable.