An Efficient Approach for Event Prediction Using Collaborative Distance Score of Communities

Abstract

An effective technique for prediction of events can capture the evolution of communities and help understand the collaborative trends in massive dataset applications. One major challenge is to find the derived features that can improve the accuracy of ML models in efficient prediction of events in such evolutionary patterns.

It is often observed that a group of researchers associate with another set of researchers having similar interests to pursue some common research goals. A study of such associations forms an essential basis to assess collaboration trends and to predict evolving topics of research. A hallmarked co-authorship dataset such as DBLP plays a vital role in identifying collaborative relationships among the researchers based on their academic interests.

The association between researchers can be calculated by computing their collaborative distance. Refined Classical Collaborative Distance (RCCD) proposed in this paper is an extension of existing Classical Collaborative distance (CCD).

This computed RCCD score is then considered as a derived feature along with other community features for effective prediction of events in DBLP dataset.

The experimental results show that the existing CCD method predicts events with an accuracy of 81.47%, whereas the accuracy of our proposed RCCD method improved to 84.27%. Thus, RCCD has been instrumental in enhancing the accuracy of ML models in the effective prediction of events such as trending research topics.