多任务学习的论文被引量预测

Predicting paper citations via multi-task learning

提出一种基于多任务的学习模型,进行论文被引量预测.首先,该模型对论文相关数据进行预处理,获取论文的网络拓扑特征和文本特征.然后,提出一个带有注意力机制的图卷积神经网络和Transformer模型分别处理这两种特征,并对论文被引量进行预测.最后,在交通领域论文数据集上的实验表明,该方法取得了较好的性能,这为未来文献分析相关工作提供了新的思路.

In this paper, a multi-task learning model is proposes to predict the paper citations. Firstly, we preprocess the relevant data of the paper to obtain the network topology and text features of the paper, and propose a graph convolution network with an attention mechanism and Transformer to process these two features respectively. Based on this, we can predict the citation amount of the paper. After that, experiments on the data set of papers in the field of transportation show that this method achieves the better performance, which provides a new idea for future literature analysis.