NeuroClustr: Empowering Biomedical Text Clustering with Neuro-Symbolic Intelligence

Fakhare Alam, Khalid Mahmood, Madan Krishnamurthy

Keywords: Clustering, Neuro-Symbolic, Deep Learning, Biomedical Text, Healthcare, knowledge Infusion

TL;DR: NeuroClustr: Empowering Biomedical Text Clustering with Neuro-Symbolic Intelligence

Abstract: The exponential growth of biomedical unstructured data being generated every day has made it increasingly challenging to accurately cluster them and extract meaningful insights. Traditional clustering algorithms such as K-means are limited in their ability to capture contextual information and semantically explain the reasoning behind the clustering results when applied to a large corpus of unstructured data. As a result, there is a need for more advanced techniques that can integrate deep learning and symbolic reasoning to improve clustering performance. Integrating domain-specific knowledge from external sources through a Neuro-Symbolic approach can facilitate the optimization of clustering algorithms by generating new hypotheses. This research paper proposes a novel framework NeuroClustr to cluster biomedical text corpus using a Neuro-Symbolic approach in conjunction with deep learning. The framework employs Recurrent Neural Network (RNN) architecture to capture important sequence to sequence information in textual data and uses BioBERT based encoded representation and infused knowledge rules from external sources such as domain specific ontology to effectively cluster the biomedical documents. The evaluation results show that the proposed framework outperforms traditional baseline models by 43% and achieves average precision of 88% across all identified clusters for COVID-19 Dataset. This demonstrates the potential of deep neural networks with knowledge infusion in improving clustering accuracy for large and complex biomedical text corpus.