Leveraging GANs for citation intent classification and its impact on citation network analysis

• We propose a semi-supervised model for citation intent classification that integrates GANs with domain-specific SciBERT embeddings. • This paper applies intent-based filtering to large-scale citation networks, including the unarXiv dataset with 76k nodes and 171k edges. • cGAN-SciBERT yields competitive F1-scores (0.887 on SciCite) with a minimal number of parameters. • Citation intent filtering changes centrality rankings and enables a more refined assessment of paper influence. Citations pla