Pushing Forward Hybrid AI: Two Journal Papers Accepted

We’re excited to announce that two journal papers from our group have been accepted for publication, reflecting our ongoing work at the intersection of Hybrid AI. One paper explores neuro-symbolic data augmentation techniques, while the other delves into human-AI collaboration in knowledge validation processes.

The first paper, authored by Majlinda Llugiqi, Fajar J. Ekaputra, and Marta Sabou, is titled “Semantic-based Data Augmentation for Machine Learning Prediction Enhancement” [link] and has been accepted in the Neurosymbolic Artificial Intelligence journal.

This work explores how integrating knowledge graph (KG) information into tabular datasets can boost machine learning performance, particularly in data-scarce environments. By leveraging neuro-symbolic techniques and embedding-based distance features, the authors demonstrate that semantic augmentation using neuro-symbolic techniques can lead to more accurate and robust predictions. The approach was evaluated in the medical domain, highlighting its potential for improving classification tasks where high-quality labeled data is limited.

The second paper, authored by Stefani Tsaneva, Danilo Dessì, Francesco Osborne and Marta Sabou and titled “Knowledge Graph Validation by Integrating LLMs and Human-in-the-Loop’‘ [link]  has been accepted in the journal of Information Processing & Management.

The paper deals with the problem of scalable KG validation, as part of the automated construction of large-scale knowledge graphs: We propose and subsequently evaluate various workflows relying on human-in-the-loop methods, large language models (LLMs) or a combination thereof, applied to a large scale, industry strength resource – the Computer Science Knowledge Graph. We believe that our findings in terms of human-LLM workflows and their trade-offs would likely be of interest to researchers focusing on other knowledge intensive tasks, beyond knowledge graph validation.