Application of Graph Entropy for Knowledge Discovery and Data Mining

Entropy, originating from statistical physics, is an interesting and challenging concept with many diverse definitions and various applications. Considering all the diverse meanings, entropy can be used as a measure of disorder in the range between total order (structured) and total disorder (unstructured) as long as by “order” we understand that objects are segregated by their properties or parameter values. States of lower entropy occur when objects become organized, and ideally when everything is in complete order, the entropy value is 0. These observations generated a colloquial meaning of entropy. In this chapter we investigate the state of the art in graph-theoretical approaches and how they are connected to text mining. This prepares us to understand how graph entropy could be used in data-mining processes
Next, we show how different graphs can be constructed from bibliometric data and what research problems can be addressed by each of those. We then focus on coauthorship graphs to identify collaboration styles using graph entropy. For this purpose, we selected a subgroup of the DBLP database and prepared it for our analysis. The results show how two entropy measures
describe our data set. From these results, we conclude our discussion of the
results and consider different extensions on how to improve our approach.