HybridAIR (Hybrid Approach to Intelligent Recommenders for cyber-physical systems of systems) is a three-year national research project on designing, implementing and evaluating domain-specific recommender systems for enhanced decision support in complex industrial environments, funded by the Austrian research funding association (FFG) within the funding programme KDT (call 2022), project #FO999902654. The project started in the fourth quarter of 2023.

Overview

Recommender systems powered by generative AI techniques such as large language models (LLMs), have gained increasing prevalence across diverse industries and daily experiences, drawing considerable attention in media coverage, particularly with notable advancements such as those showcased by ChatGPT. Despite the impressive performance demonstrated by LLMs, their integration encounters several limitations including contextual awareness limitations, stakeholder alignment issues, and the need for explainable, trustworthy recommendations. These limitations hinder their utility within specific industry sectors, where even minor misinterpretations can lead to significant consequences. The main challenges for the integration of such systems in these sectors include:

• The lack of data quality requirements to build meaningful explanations for the respective stakeholder.
• The lack of design and evaluation techniques for multi-stakeholder recommender systems to provide context-aware recommendation services.
• The lack of design and evaluation techniques for responsible recommender systems to increase both understandability and trust in the recommendations.

Objectives

In HybridAIR, we will set data quality standards and use LLMs to improve data quality and drive recommendations. Our focus includes developing advanced multi-stakeholder recommender toolkits to enhance recommendation accuracy, stakeholder satisfaction, and context-aware services through innovative algorithms. Additionally, we will work on techniques to boost trust in recommendations by exploring explanation models based on LLMs and implementing adaptive feedback mechanisms for stakeholders. These techniques will be empirically evaluated for efficiency, effectiveness, and acceptance in industrial applications, with a specific emphasis on the automotive and critical infrastructure sectors.