A bibliometric network analysis of Deep Learning publications applied into legal documents

Deep Learning has been gradually adopted as the main methodology to perform Natural Language Processing tasks on legal documents. In this work we provide a bibliometric network analysis of Deep Learning publications (formerly Neural Networks) applied to the analysis of legal documents. Our study includes a total sample of 138 works published between 1987 and 2020 that used DL as primary methodology. We focused on three specific objectives: identification of the journals with more publications on the subject, a co-authorship network analysis, and an examination of the most cited works. Our results show that the publications are concentrated in a small number of specialized journals on the topic, and consequently, the number of works that use DL methodologies in the legal context is smaller compared to other areas. The co-authorship network analysis reveals four broad clusters of researchers that are time-dependent concentrated into Connectionism (1), Neural Networks (1) and Deep Learning (2). Our analysis of highly cited publications delivered 10 articles with two particular authors that centralize the network. Finally, we have found that collaboration between groups of researchers from different areas is minimal, showing a window of opportunity to increase interdisciplinary research, particularly between computer and legal research groups.