Graph Neural Networks for Natural Language Processing in Human-Robot Interaction

Human-robot interaction (HRI) has become a promising field that focuses on developing intelligent systems that are capable of understanding human language. Natural Language Processing (NLP) plays a huge role in enabling robots to interpret and generate natural language, making it easier to effectively communicate. However, traditional NLP approaches sometimes struggle to capture the right structural dependencies and contextual information.To overcome these limitations, Graph Neural Networks (GNNs) have emerged as a powerful model for NLP tasks in the context of human-robot interaction. GNNs extend traditional neural network architectures to effectively model and reason about structured data, such as graphs. In the context of NLP, these graphs can represent semantic relationships between words or sentences.In this study, we model human-robot conversations as graphs. The aim of the research is to conduct a test using GNNs to predict personality traits based on conversations between human and robot. Presented results, despite a very small training set, show that GNNs can be effective in capturing the dynamics and context of conversations. GNNs lead then to improved performance in tasks such as predicting personality and estimating engagement levels. These findings suggest that GNNs have the potential to enhance the quality of HRI and improve the overall user experience.