LAUSR

With the development of citation analysis, citation mention times are drawing more attention. Aiming to extract mention times more conveniently and quickly, this study focused on developing a high-accuracy citation recognition algorithm based on neural networks, thereby providing automatic extraction of the number of citation mentions in citing papers, and on assessing its performance in PDF papers with different citation styles. We also used this algorithm to study the distribution rule and contribution of citations to citing papers. The results showed that the proposed algorithm is feasible for use in citation-mention-related research and further verified that the statistical distribution of the number of citation mentions conforms to the generalised Pareto distribution. Meanwhile, references mentioned more than twice accounted for about 20–40% of the total and contributed more than other references.