Skip to content

Personalised LLM for Emotion-Aware Human-Robot and Human-Computer Interaction

MDPI Sensors 2025 • Ghamati, Dehkordi & Zaraki • University of Hertfordshire

About This Work

Khashayar Ghamati, Maryam Banitalebi Dehkordi, Abolfazl Zaraki
University of Hertfordshire, UK

Published in MDPI Sensors 2025

DOI: 10.3390/s25072024

Towards AI-Powered Applications: The Development of a Personalised LLM for HRI and HCI introduces a groundbreaking Personalised Large Language Model (PLLM) framework that achieves real-time adaptation to user emotions in both Human-Robot Interaction (HRI) and Human-Computer Interaction (HCI). The PLLM integrates EEG-based emotion recognition with advanced dialogue generation, enabling context-aware and emotion-sensitive responses tailored to individual users.

Innovation

This is the first demonstration of real-time EEG-driven personalisation in LLM-based dialogue systems, opening new frontiers in neuroadaptive AI.

PLLM framework architecture showing EEG-driven emotion-adaptive AI dialogue system for human-robot interaction
Figure 1: Overview of the PLLM framework architecture. EEG signals are processed in real-time for emotion classification, which then conditions the large language model to generate personalised, emotion-aware responses through a user-friendly Streamlit interface.

Framework Architecture & Methodology

The PLLM system represents a paradigm shift in adaptive AI, combining neuroscience with cutting-edge language models to create truly personalised interactions. Our framework consists of three integrated components working in seamless harmony:

Core Components

EEG-based Emotion Recognition
Real-time processing of neural signals using the NeuroSense dataset, employing advanced deep learning pipelines to classify emotional states with high accuracy.
Dynamic LLM Conditioning
Emotional state outputs directly influence prompt engineering, allowing the language model to adapt its tone, content, and interaction style based on the user's current affective state.
Interactive Deployment Platform
A sophisticated Streamlit application provides real-world accessibility, enabling researchers and end-users to experience emotion-adaptive AI in practice.

Technical Achievement

The system processes EEG signals and generates contextually appropriate responses within milliseconds, demonstrating the feasibility of real-time neuroadaptive AI.

Real-time integration pipeline showing EEG emotion recognition feeding into personalised LLM dialogue generation
Figure 2: Real-time integration pipeline demonstrating the closed-loop system from EEG signal acquisition through emotion classification to adaptive dialogue generation, showcasing the seamless flow of neuroadaptive personalisation.

Experimental Results & Validation

Our comprehensive evaluation using the NeuroSense dataset demonstrates the effectiveness of emotion-driven personalisation in AI dialogue systems. The results establish new benchmarks for neuroadaptive human-computer interaction:

Key Findings

Impact

This work establishes the foundational framework for next-generation neuroadaptive AI systems, with applications spanning healthcare, education, entertainment, and assistive technologies.

Scientific Contribution & Novelty

Our research addresses a critical gap in adaptive AI systems by introducing the first practical framework for EEG-driven LLM personalisation. This work contributes to multiple domains:

Research Contributions

Neuroadaptive AI Architecture
Novel integration of brain-computer interfaces with large language models for real-time personalisation.
Emotion-Aware Dialogue
Demonstration of how affective states can enhance AI communication quality and user experience.
Open-Source Implementation
Reproducible research framework enabling widespread adoption and extension by the research community.
Cross-Domain Applications
Validated approach applicable to both HRI and HCI contexts, broadening the impact across multiple fields.

Future Implications

This work paves the way for emotionally intelligent AI assistants, therapeutic robots, and adaptive learning systems that respond to users' mental states in real-time.

Open Science & Collaboration

We are committed to advancing open, reproducible science in neuroadaptive AI. This research provides a foundation for collaborative development of more sophisticated emotion-aware systems and welcomes partnerships to extend these capabilities.

Full Paper & Code Access

Download Full Paper (Open Access) Access Source Code

The article is published under the CC BY 4.0 license, ensuring open access for the research community.