[Objective]This paper proposed a literature citation prediction model, which was used to discover potential research hotspots and promote the optimization of journal editing.

[Methods]Considering the relevant factors such as keywords, authors, institutions, countries, and citations of the literature, we used graph convolution to extract features, then used recurrent neural network and attention model to mine the time-series information of citations and important literature features.

[Results]Using transportation related literature published in WoS core journals, compared with the benchmark model, the model has a maximum improvement of 15.23% and 16.91% on RMSE, MAE and other indicators.

[Limitations]In the pre-training step of the proposed model, multiple graph convolutions were performed, which made the time complexity high.

[Conclusions] The model proposed in this paper fully integrated the features of the literature, then greatly improved the performance of the citation prediction model.