Knowledge Graphs for Community Detection in Textual Data

Federica Rollo, Laura Po

Abstract: Online sources produce a huge amount of textual data, i.e., freeform text. To derive insightful information from them and facilitate the application of Machine Learning algorithms textual data need to be processed and structured. Knowledge Graphs (KGs) are intelligent systems for the analysis of documents. In recent years, they have been adopted in multiple contexts, including text mining for the development of data-driven solutions to different problems. The scope of this paper is to provide a methodology to build KGs from textual data and apply algorithms to group similar documents in communities. The methodology exploits semantic and statistical approaches to extract relevant insights from each document; these data are then organized in a KG that allows for their interconnection. The methodology has been successfully tested on news articles related to crime events occurred in the city of Modena, in Italy. The promising results demonstrate how KG-based analysis can improve the management of information coming from online sources.