Abstract

The shared interest among existing research topics matures over time until it emerges as a topic of its own. This paper detects emerging topics as well as general predictor models spanning multiple research domains through the network-based topic evolution approach, which offers additional topic evolution capabilities such as extrapolation of data and separation of topic transition and correlation. Topics are represented as their neighbors in the past, or ancestors, and their structural properties are used to train binary classification models in capturing the materialization of such topics. The entirety of 197 million publications within the Microsoft Academic Graph was used to build multiple datasets, where machine learning algorithms were trained with structural features resulting in over 0.98 area under the precision–recall curve. General topic emergence predictor equations are then proposed based on the models trained specifically for each domain, which were able to capture a common pattern shared by emerging topics in general.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call