Referencing of Document Content Using Similarity Measures

Abstract

One of the biggest challenges with scientific writing automation is still the difficulty of automatically locating and adding relevant references in scholarly papers. This paper addresses this issue by proposing a three-phase automatic referencing system based on semantic similarity measures: reference insertion, semantic similarity computation, and preprocessing (tokenization, stop word removal, morphosyntactic marking, and lemmatization). Based on semantic similarity, our experimental results confirm that the system can automatically identify and insert relevant references. The Resnik measure outperformed the Mihalcea measure (43% accuracy, 50% precision, and 59% F1-score), achieving the best performance with (57% accuracy, 58% precision, and 64% F1-score).