Cleaning English Abstracts of Scientific Publications

TL;DR

The paper tackles the problem of clutter in English scientific abstracts, which can distort downstream analyses and embeddings. It introduces an open-source Abstract Cleaner that uses spaCy-based named-entity recognition to identify and remove clutter as REM spans, trained on 9,000 Scopus abstracts with manual labeling. The approach is conservative and precise, achieving high token-level performance (precision 0.973, recall 0.919, F1 0.945) and improving the information content of fixed-length embeddings, with observed shifts in similarity rankings after cleaning. The tool is publicly available on Huggingface, runs quickly, and provides practical guidance and code for decluttering abstracts in real-world pipelines.

Abstract

Scientific abstracts are often used as proxies for the content and thematic focus of research publications. However, a significant share of published abstracts contains extraneous information-such as publisher copyright statements, section headings, author notes, registrations, and bibliometric or bibliographic metadata-that can distort downstream analyses, particularly those involving document similarity or textual embeddings. We introduce an open-source, easy-to-integrate language model designed to clean English-language scientific abstracts by automatically identifying and removing such clutter. We demonstrate that our model is both conservative and precise, alters similarity rankings of cleaned abstracts and improves information content of standard-length embeddings.

Figures (2)

• Figure 1: Loss and performance metrics per epoch
• Figure 2: Relation abstract length and misses/excess