AGSTA-NET: adaptive graph spatiotemporal attention network for citation count prediction.

Autor/inn/en	Wang, Bin; Wu, Feng; Shi, Lukui
Titel	AGSTA-NET: adaptive graph spatiotemporal attention network for citation count prediction.
Quelle	In: Scientometrics, (2022) 1, S.511-541 PDF als Volltext Verfügbarkeit
Dokumenttyp	gedruckt; online; Zeitschriftenaufsatz
ISSN	0138-9130
DOI	10.1007/s11192-022-04541-0
Schlagwörter	Citation count prediction; Graph neural network; Attention; Spatiotemporal; Recurrent neural network
Abstract	Abstract With the rapid development of scientific research, a large number of scientific papers are produced every year. It is very important to find influential papers quickly from the massive literature resources, which can not only help researchers identify papers with reference value, but also help scientific research management departments to allocate resources. Among the quantification measures of academic impact, citation count stands out for its frequent use in the research community. Previous studies have either treated papers as independent individuals without considering their citation relationships in the citation network or have not adequately considered the long-time dependence of citation time series. In this paper, we consider the structural features of citation networks and propose a deep learning method AGSTA-NET from the perspective of spatio-temporal fusion, which models heterogeneous citation networks formed early in the publication of a paper and predicts the citation count for an article in the next few years. AGSTA-NET contains capturing module of spatial dependence and capturing module of time dependence. It could fully dig the complex spatio-temporal information from the dynamic heterogeneous citation network by only inputting the heterogeneous citation network to the model. Meanwhile, the sub-networks designed in this paper could adaptively determine the threshold of the loss function according to the samples for better training. Experiments validate that AGSTA-NET outperforms current state-of-the-art methods in citation count prediction.
Erfasst von	OLC
Update	2023/2/05