Scholarly networks have attracted great attentions, such as scholarly impact evaluation, scholarly impact prediction, scholarly recommendation, co-author relationships analysis, and team identification. Ranking research institutions as an important aspect of scholarly impact research is of great significance for decision makers, such as funding allocation, promotion, and transfer. There has been much debate about the scientific correctness behind those rankings. Predicting the number of accepted conference papers of research institutions next year is proposed by KDD Cup 2016, which aims to measure the impact of research institutions. To accurately predict the impact of different institutions in the eight top conferences: FSE, ICML, KDD, MM, MobiCom, SIGCOMM, SIGIR, and SIGMOD, a novel model was proposed in which the number of accepted papers of each institution, country, and time factors driving the impact of institution change are used as training features. Correspondingly, a hybrid model of support vector machine and neural network is constructed for resolving the predictive task. The experimental results show that the proposed method is better than the Markov model and the neural network model in terms of normalized discounted cumulative gain.