Identifying the Development and Application of Artificial Intelligence in Scientific Text

TL;DR

This paper presents a machine learning approach to identify AI-related research publications across large scientific corpora by leveraging arXiv metadata, enabling dynamic and automated classification without expert labeling.

Contribution

The authors develop a supervised classification method that generalizes to identify AI-relevant publications in multiple large datasets, achieving high accuracy without manual labeling.

Findings

Predictive F1 scores between 0.75 and 0.86 for key AI subfields.

High precision (0.83) and recall (0.85) for a combined AI relevance model.

Supervised classifiers can effectively identify AI research across diverse datasets.

Abstract

We describe a strategy for identifying the universe of research publications relevant to the application and development of artificial intelligence. The approach leverages the arXiv corpus of scientific preprints, in which authors choose subject tags for their papers from a set defined by editors. We compose a functional definition of AI relevance by learning these subjects from paper metadata, and then inferring the arXiv-subject labels of papers in larger corpora: Clarivate Web of Science, Digital Science Dimensions, and Microsoft Academic Graph. This yields predictive classification $F_1$ scores between .75 and .86 for Natural Language Processing (cs.CL), Computer Vision (cs.CV), and Robotics (cs.RO). For a single model that learns these and four other AI-relevant subjects (cs.AI, cs.LG, stat.ML, and cs.MA), we see precision of .83 and recall of .85. We evaluate the out-of-domain performance of our classifiers against other sources of topic information and predictions from alternative methods. We find that a supervised solution can generalize to identify publications that belong to the high-level fields of study represented on arXiv. This offers a method for identifying AI-relevant publications that updates at the pace of research output, without reliance on subject-matter experts for query development or labeling.