Citation Prediction and Recommendation with Semantic and Graph-Based Learning
DOI: https://doi.org/10.62517/jbdc.202601414
Author(s)
Shenghao Luo
Affiliation(s)
University of Illinois - Urbana-Champaign, USA
Abstract
Citation prediction can be formulated as both a link prediction and recommendation task. Given a source paper A and a candidate reference paper B, the objective is to estimate the probability that A cites B. Unlike undirected link prediction, citation prediction is inherently directed because the relationship from A to B is not equivalent to that from B to A, and it is temporally constrained because a valid reference must be published before the source paper. This study begins with a text-based baseline using semantic document embeddings and logistic regression, then examines probability calibration, Top-K recommendation, embedding robustness, leakage-free graph message passing, directed message passing, candidate-reference graph features, and transfer experiments on the larger OGBL-citation2 benchmark. The results show that semantic text representations provide a strong foundation, while graph information offers complementary signals when incorporated without temporal or edge leakage. End-to-end graph representation learning is not consistently superior to a comparable neural text-only model, especially in cold-start settings where the source paper has limited citation history. In contrast, simple historical graph features derived only from the candidate reference paper consistently improve citation identification and ranking. Overall, combining semantic text representations with carefully controlled candidate-side graph features improved Cit-HepTh F1 by about 0.54% and OGBL-citation2 subset MRR by about 22.6% over the corresponding text-only baselines.
Keywords
Citation Prediction; Citation Recommendation; Graph Neural Networks; Semantic Embedding; Link Prediction; Graph Leakage
References
[1] Bhagavatula, C., Feldman, S., Power, R., & Ammar, W. (2018). Content-based citation recommendation. Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, 238–251. https://doi.org/10.18653/v1/N18-1022
[2] Cohan, A., Feldman, S., Beltagy, I., Downey, D., & Weld, D. S. (2020). SPECTER: Document-level representation learning using citation-informed transformers. Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, 2270–2282. https://doi.org/10.18653/v1/2020.acl-main.207
[3] Ebesu, T., & Fang, Y. (2017). Neural citation network for context-aware citation recommendation. Proceedings of the 40th International ACM SIGIR Conference on Research and Development in Information Retrieval, 1093–1096. https://doi.org/10.1145/3077136.3080730
[4] Gori, M., Monfardini, G., & Scarselli, F. (2005). A new model for learning in graph domains. Proceedings of the IEEE International Joint Conference on Neural Networks, 729–734. https://doi.org/10.1109/IJCNN.2005.1555942
[5] Hamilton, W. L., Ying, R., & Leskovec, J. (2017). Inductive representation learning on large graphs. Advances in Neural Information Processing Systems, 30, 1024–1034. https://papers.nips.cc/paper_files/paper/2017/hash/5dd9db5e033da9c6fb5ba837a7ebea9-Abstract.html
[6] He, Q., Pei, J., Kifer, D., Mitra, P., & Giles, C. L. (2010). Context-aware citation recommendation. Proceedings of the 19th International Conference on World Wide Web, 421–430. https://doi.org/10.1145/1772690.1772734
[7] Hu, W., Fey, M., Zitnik, M., Dong, Y., Ren, H., Liu, B., Catasta, M., & Leskovec, J. (2020). Open Graph Benchmark: Datasets for machine learning on graphs. Advances in Neural Information Processing Systems, 33, 22118–22133.
[8] Jeong, C., Jang, S., Park, E., & Choi, S. (2020). A context-aware citation recommendation model with BERT and graph convolutional networks. Scientometrics, 124(3), 1907–1922. https://doi.org/10.1007/s11192-020-03561-y
[9] Kipf, T. N., & Welling, M. (2017). Semi-supervised classification with graph convolutional networks. International Conference on Learning Representations. https://arxiv.org/abs/1609.02907
[10] Liben-Nowell, D., & Kleinberg, J. (2007). The link-prediction problem for social networks. Journal of the American Society for Information Science and Technology, 58(7), 1019–1031. https://doi.org/10.1002/asi.20591
[11] Open Graph Benchmark. (n.d.). Leaderboards for link property prediction: ogbl-citation2. Retrieved August 12, 2026, from https://ogb.stanford.edu/docs/leader_linkprop/
[12] Open Graph Benchmark. (n.d.). Link property prediction: ogbl-citation2. Retrieved August 12, 2026, from https://ogb.stanford.edu/docs/linkprop/#ogbl-citation2
[13] Page, L., Brin, S., Motwani, R., & Winograd, T. (1999). The PageRank citation ranking: Bringing order to the web. Stanford InfoLab. http://ilpubs.stanford.edu:8090/422/
[14] Scarselli, F., Gori, M., Tsoi, A. C., Hagenbuchner, M., & Monfardini, G. (2009). The graph neural network model. IEEE Transactions on Neural Networks, 20(1), 61–80. https://doi.org/10.1109/TNN.2008.2005605
[15] Stanford Network Analysis Project. (n.d.). High-energy physics theory citation network. SNAP Datasets. Retrieved August 12, 2026, from https://snap.stanford.edu/data/cit-HepTh.html
[16] Tang, X., Wan, X., & Zhang, X. (2014). Cross-language context-aware citation recommendation in scientific articles. Proceedings of the 37th International ACM SIGIR Conference on Research & Development in Information Retrieval, 817–826.
[17] Veličković, P., Cucurull, G., Casanova, A., Romero, A., Liò, P., & Bengio, Y. (2018). Graph attention networks. International Conference on Learning Representations. https://arxiv.org/abs/1710.10903