Triples and Knowledge-Infused Embeddings for Clustering and Classification of Scientific Documents

TL;DR

The paper tackles the challenge of organizing rapidly expanding scientific literature by integrating structured knowledge in the form of subject-predicate-object triples with unstructured abstract text. It introduces a modular pipeline that creates four representations, embeds them with four transformer models, and evaluates them with three clustering methods and supervised classifiers on a filtered arXiv dataset. Key findings show that full abstracts drive the strongest clustering signals, while hybrid representations that fuse text and triples yield the best classification performance, notably reaching 92.60% accuracy and 0.925 macro-F1. The results also reveal that lightweight encoders like MiniLM and MPNet outperform domain-specific models for clustering, whereas SciBERT shines in structured-input classification. Overall, knowledge-infused, hybrid representations offer practical benefits for semantic organization and retrieval in digital libraries and scholarly search systems, highlighting a scalable path toward knowledge-augmented document understanding.

Abstract

The increasing volume and complexity of scientific literature demand robust methods for organizing and understanding research documents. In this study, we explore how structured knowledge, specifically, subject-predicate-object triples, can enhance the clustering and classification of scientific papers. We propose a modular pipeline that combines unsupervised clustering and supervised classification over multiple document representations: raw abstracts, extracted triples, and hybrid formats that integrate both. Using a filtered arXiv corpus, we extract relational triples from abstracts and construct four text representations, which we embed using four state-of-the-art transformer models: MiniLM, MPNet, SciBERT, and SPECTER. We evaluate the resulting embeddings with KMeans, GMM, and HDBSCAN for unsupervised clustering, and fine-tune classification models for arXiv subject prediction. Our results show that full abstract text yields the most coherent clusters, but that hybrid representations incorporating triples consistently improve classification performance, reaching up to 92.6% accuracy and 0.925 macro-F1. We also find that lightweight sentence encoders (MiniLM, MPNet) outperform domain-specific models (SciBERT, SPECTER) in clustering, while SciBERT excels in structured-input classification. These findings highlight the complementary benefits of combining unstructured text with structured knowledge, offering new insights into knowledge-infused representations for semantic organization of scientific documents.