In academia, researchers collaborate with their peers to improve the quality of research and thereby enhance academic profiles. However, information overload in big scholarly data poses a challenge in identifying potential researchers for fruitful collaboration. In this article, we introduce a multi-level fusion-based model for collaborator recommendation, DRACoR (Deep learning and Random walk based Academic Collaborator Recommender). DRACoR fuses deep learning and biased random walk model to provide the recommendation for potential collaborators that share similar research interests at the peer level. We run a topic model on abstracts and Doc2Vec on titles on year-wise publications to capture the dynamic research interests of researchers. Author-author cosine similarity is computed from the feature vectors extracted from abstracts and titles and is then used to weigh edges in the author-author graph (AAG). We also aggregate various meta-path features with profile-aware features to bias the random walk behavior. Finally, we employ a random walk with restart(RWR) to recommend top $N$ collaborators where the edge weights are used to bias the random walker’s behavior. Extensive experiments on DBLP and hep-th datasets demonstrate the effectiveness of our proposed DRACoR model against various state-of-the-art methods in terms of precision, recall, F1-score, MRR, and nDCG.

在学术界，研究人员与同伴合作以提高研究质量，从而提高学术水平。但是，大量学术数据中的信息过载给寻找潜在的研究人员进行富有成效的合作带来了挑战。在本文中，我们为协作者推荐引入了基于多层次融合的模型DRACoR（基于深度学习和基于随机游走的学术协作者推荐）。DRACoR融合了深度学习和有偏向的随机游走模型，为在同行中具有相似研究兴趣的潜在合作者提供了建议。我们在年度出版物上对摘要运行主题模型，对标题运行Doc2Vec，以捕捉研究人员动态的研究兴趣。从摘要和标题中提取的特征向量计算出作者-作者的余弦相似度，然后将其用于权衡作者-作者图（AAG）中的边缘。我们还将各种元路径功能与配置文件感知功能聚合在一起，以偏向随机行走行为。最后，我们采用带有重启（RWR）的随机游动来推荐顶部$ñ$协作者，其中边缘权重用于偏向随机步行者的行为。在DBLP和hep数据集上进行的大量实验证明了我们提出的DRACoR模型相对于各种最新方法在有效性，召回率，F1得分，MRR和nDCG方面的有效性。