Determining the critical thresholds for co-word network based on the theory of percolation transition: A case study in Buddhist studies

One of the chief purposes of bibliometric analysis is to reveal the intellectual structure of a knowledge domain. Yet due to the magnitude and the heterogeneous nature of bibliometric networks, some sorts of filtering procedures are often required to make the resulting network interpretable. A co-word analysis of more than 135,000 scholarly publications on Buddhism was conducted to compare the intellectual structure of Buddhist studies in three language communities, Chinese, English and Japanese, over two periods (1957–1986 and 1987–2016). Six co-word similarity networks were created so social network analysis-based community-detection algorithm can be identified to compare major research themes in different languages and eras. The paper aims to discuss this issue.

A series of filtering procedures was performed to exclude less discriminatory keywords and spurious relationships of a large, cross-language co-word network in Buddhist studies. Chief among the filtering heuristics was a percolation-transition based method to determine the similarity threshold that involves observing the relative decrease of nodes in the giant component with the increasing similarity threshold.

It was found that the topical patterns in the Chinese and Japanese scholarship of Buddhism are alike and observably distinct from that of the English scholarship. Furthermore, a far more drastic changes of research themes were observed in the English literature relative to the Chinese and Japanese literature.

The filtering procedures were shown to greatly enhance the modularity values and limited the number of modularity classes; thus, domain expert interpretation is feasible.