Paper acceptance prediction at the institutional level based on the combination of individual and network features

Papers published in top conferences or journals is an important measure of the innovation ability of institutions, and ranking paper acceptance rate can be helpful for evaluating affiliation potential in academic research. Most studies only focus on the paper quality itself, and apply simple statistical data to estimate the contribution of institutions. In this work, a novel method is proposed by combining different types of features of affiliation and author to predict the paper acceptance at the institutional level. Based on the history of the paper published, this work firstly calculates the affiliation scores, constructs an institutional collaboration network and analyzes the importance of the institutions using network centrality measures. Four measures about the authors’ influence and capability are then extracted to take the contributions of authors into consideration. Finally, a random forest algorithm is adopted to solve the prediction problem of paper acceptance. As a result, this paper improves the ranking of the paper acceptance rate NDCG@20 to 0.865, which is superior to other state-of-the-art approaches. The experimental results show the effectiveness of proposed method, and the information between different types of features can be complementary for predicting paper acceptance rate.