Positional analysis in cross-media information diffusion networks

Abstract

This paper describes a network reduction technique to reveal possibly hidden relational patterns in information diffusion networks of interlinked content published across different types of online media. Topic specific content items such as tweets (Twitter), web pages, or versions of Wikipedia articles can reference each other through hyperlinks, revisions, or retweet relationships, and thus, constitute a network that reflects the dissemination of information on the web. Beyond focusing on the structural linking of content items alone, the temporal aspect of information diffusion is explicitly taken into account by modelling the edge weight between two interlinked items according to the difference in their publication times. Non-negative matrix factorisation (NMF) is applied to decompose the resulting networks into groups of nodes occupying similar positions, which means that they have similar abilities to spread or receive information to or from other nodes. This allows for an easier observation of the basic underlying structure of cross-media information diffusion networks and their main information pathways. The utility of the approach and differences to other techniques will be demonstrated along two application scenarios related to two popular news stories and their dissemination in online media in 2016.