融合学术水平相似性的合作者推荐模型

合作者推荐工作对科学研究的发展和科技成果的转化很有帮助，然而学者间水平的差距严重影响了合作的建立。模型从学者间学术水平差距，合作网络的拓扑距离以及研究兴趣三个角度进行合作者推荐。首先，定义了学者—学者、学者—主题、学者—水平标签三种网络，并融合成主题—学者—水平标签图；之后对该图中的边赋权重，从而将合作者推荐任务转换为链路预测问题；最后使用偏向重启随机游走算法计算学者间的访问概率，并筛选访问概率大的学者作为推荐建议。在三个数据集上的实验表明，模型在推荐的准确率、召回率、F1指数上平均提高了5.4%、2.7%、3.8%，同时目标学者与推荐学者的学术水平匹配度更高。

Collaborator recommendation model fused academic level similarity

Collaboration recommendation is helpful to the development of scientific research and the transformation of technological achievements. However, the gap between scholars'' academic levels seriously affects the establishment of cooperative relations. This paper made recommendations from three perspectives: the academic level gap between scholars, the topological distance in the collaborative network, and research interests. Firstly, this paper defined three networks, namely scholar-scholar network, scholar-topic network and scholar-level label network, and merged them into a graph of topic-scholar-level label, and set weight to the edges in the graph. Then it turned collaboration recommendation task into a link prediction task. Finally, it employed the biased restart random walk algorithm to calculate the probability of visits among scholars, and recommended the candidate scholars with high visit probability to target scholars. Experiments on three datasets show that the proposed model can improve the precision rate, recall rate and F1 index by 5.4%, 2.7%, and 3.8%. In addition, the academic levels of target scholars and recommended scholars are more closely matched.