Keywords:emerging research field, node embedding
It's important to identify promising research and research areas early to determine which research to invest. In addition, it's necessary to develop a technology for automatically predicting future research trends because of increasing the number of publication and the research fragmentation. There are many metrics for research performances and it depends on the objective which future trends to show. So, the problem is developing the technology to predict research trends for various metrics. Therefore, in this paper, we propose a method to extract distributed representations for automatically predicting various research metrics in the future using various heterogeneous network information with research papers. Experimental results show that prediction of the reference relation between the research papers was 95.6% F-value, and the h-index after three years from publication was 64.4% under certain conditions. The first result shows that the proposed method can sufficiently map the reference relations to a vector space. On the other hand, the prediction accuracy of the future h-index is equivalent to the comparison method, and further research needs. The results suggest that distributed representations of heterogeneous networks for scientific paper may be the basis for the automatic prediction of technology trends.