ACARIS - preliminary work

ACARIS - preliminary work
Photo by Google DeepMind / Unsplash

We have recently wrapped up the initial phase of work on ACARIS, the Advanced Communication Augmentor and Relational Insights System, a project that aims to revolutionize the way people talk, while also giving machines the understanding they need to comprehend human emotions, intentions, and interest metrics in text-based communication.

Our guiding hypothesis for ACARIS is twofold: firstly, that given enough conversational data per person, a high degree of prediction accuracy can be achieved regarding human behavior in text communication. This is due to the fundamental similarities that humans share in their communication patterns. Secondly, we believe that by individualizing predictions using unique vector representations of a person's emotional state, intent, interest, and personality - which we call "user embeddings" - we can attain person-specific improvements in performance. This belief is rooted in the core concept of neural networks as universal function approximators. Thus, theoretically, a neural network should be capable of associating a user embedding with a person's behavior in text communication, utilizing this information to refine its predictions.

In our initial approach, we leveraged the DistilBERT architecture, adapting it to accommodate the individualized context provided by user embeddings. Despite our efforts, this iteration did not yield any significant improvements over the existing state-of-the-art in sentiment analysis. However, this has only served to reinvigorate our determination. We are convinced that our approach has merit and potential - it's simply a matter of refining our methods and expanding our dataset.

Looking forward, our mission is not solely focused on enhancing ML model performance but also address a growing social concern. The digital age has revolutionized communication, but not without a cost. As face-to-face interactions have been largely replaced by online text communication, we've observed a decline in interpersonal skills, particularly among the younger generation. ACARIS aims to redress this by providing insightful analysis of text-based communication, potentially aiding users in improving their social skills.

Moreover, as conversational models become more prevalent, the need for these systems to comprehend human emotions, intent, and interest becomes increasingly critical. Systems like digital assistants, chatbots, and other interactive systems stand to benefit enormously from the advancements ACARIS aims to provide.