Node embedding based community detection and other techniques to analyze implicit social graph Of a literary text series

摘要

Substantial research has been done on how to perform social network analysis on literary text. However, these works have mostly dealt with only a single text or have considered the entire series as a single text leading to one social network. In doing so, we miss out on information and interpretation of how the implicit social graph has evolved over the series instead of a single volume. Our work addresses this gap in the present body of work. For example, protagonists can be identified over the scope of the full series. Our work also has an innovative approach to community detection. Clustering and community analysis are typically two sides of the same coin. Clustering is performed when enough attributes are known about each node. In the absence of this information, community detection can be done based on network characteristics in a network data set. Network or node embedding is a method to capture network characteristics. Our work takes an inspiration from both and combines node embedding with classical clustering algorithm and successfully compares with the ground truth. Various algorithms can be used to predict the links in the implicit social graph and then compare the results with the ground truth. We have also evaluated various existing algorithms on the detected social graph.