基于预训练语言模型特征扩展的科研论文推荐

Research paper recommendation based on feature extension of pre-training language models

针对科研学术论文推荐数据稀疏的问题,提出一种基于预训练语言模型特征扩展的科研论文推荐方法.通过预训练语言模型学习论文摘要的特征表示,将其作为辅助信息构建推荐模型,再将辅助特征和用户-论文标签矩阵共同输入半自编码机模型进行训练,最终实现推荐任务.实验结果表明,相比自编码机等神经网络方法,该方法推荐的科研论文更为准确,可提高科研工作效率.

To solve the problem of sparse recommendation data of scientific papers,this paper proposes a research paper recommendation method based on feature expansion of pre-training language model.The feature representations of the abstracts of the papers are learned through a pre-training language model,which is used as auxiliary information to build the recommendation model?and the auxiliary features and the user-paper tags matrix are fed into the semi-autoencoder model for training and finally the recommendation task is realized.Experimental results show that compared with neural network method such as autoencoder,the scientific research papers recommended by this method are more accurate and the research efficiency is improved.