An Extended Evaluation of Single-Label Multi-modal Field of Research Classification Using a Taxonomy-Based Metric

Florian Ruosch, Rosni Vasu, Ruijie Wang, Luca Rossetto, Abraham Bernstein

Abstract: This paper presents a multi-modal approach to the single-label field of research classification shared task. Our method, SLAMFORC, incorporates metadata, full text, and image data from scholarly articles to generate comprehensive document embeddings. We built a voting ensemble of pre-trained BERT models (SciBERT and SciNCL) and traditional classifiers and achieved competitive performance in the Field of Research Classification of Scholarly Publications shared task. SLAMFORC scored highest in F1 score and precision and second best in recall and accuracy. We extend our original analysis by examining misclassified samples to improve future iterations. Additionally, we apply a taxonomy-based evaluation metric to better assess our results.