Manually analyzing large collections of research articles is a time- and resource-intensive activity, making it difficult to stay on top of the latest research findings. Limitations of automated solutions lie in limited domain knowledge and not being able to attribute extracted key terms to a focal article, related work, or background information. We aim to address this challenge by (1) developing a framework for classifying sentences in scientific publications, (2) performing several experiments comparing state-of-the-art sentence transformer algorithms with a novel few-shot learning technique and (3) automatically analyzing a corpus of articles and evaluating automated knowledge extraction capabilities. We tested our approach for combining sentence classification with ontological annotations on a manually created dataset of 1,000 sentences from Information Systems (IS) articles. The results indicate a high degree of accuracy underlining the potential for novel approaches in analyzing scientific publications

Paper Number



Track 5: Data Science & Business Analytics