ACARIS: Attempting to Improve Conversational AI and Human Social Skills using User Embeddings
Tejas Patel, Sandeep Shivam, Amit Kumar Padhy, Bharadwaj Vulugunda, Chaitanya Kulkarni, Chandrashekhar Medicherla
Abstract
ACARIS (Advanced Communication Augmentor and Relational Insights System) proposes a novel method to analyze emotional state, intent, and interest in text communication, aiming to improve human social skills and conversational AI performance. By leveraging user embeddings and DistilBERT architecture modifications, ACARIS seeks to predict human behavior with high confidence, offering personalized insights for better interpersonal relationships and AI interactions. Despite initial challenges, ongoing research aims to refine ACARIS for broader applications in understanding human dynamics. This work provides valuable insights into when static user embeddings do and do not improve sentiment classification performance.