Semantic Similarity Analysis of Scientific Papers in Scholarly Knowledge Graphs

Authors

Nguyen T.H., Pruski C., Silveira M.D.

Reference

Ceur Workshop Proceedings, vol. 3977, 2025

Description

Scholarly Knowledge Graphs (SKGs) structure academic information, enabling research discovery and knowledge synthesis. However, detecting their structural and semantic evolution remains challenging due to contextual variations and implicit knowledge shifts. In this paper, we introduce MOKA, a framework that integrates bibliometric data, advanced Natural Language Processing, and Agentic Retrieval-Augmented Generation to characterize SKG evolution. As part of this process, we propose a method to measure semantic similarity between papers by refining the representation of documents and applying four pretrained language models. Our approach assesses semantic similarity across different textual granularity levels, including full papers, abstracts, and the ORKG metadata. Our experimental results demonstrates that the similarity scores highly depend on the combination of language model and the evoked dimensions of the papers to evaluate.