From Social Network Graphs to Causal Bayes Nets

This paper proposes a new method for morphing an existing social network graph into a Causal Bayes Net. We assume only that an undirected graph of a social network exists with large amounts of text data associated with each distinct node (person or organization). It is desired to convert such a graph into a causal probabilistic representation for predictive analysis. The probabilistic representation can also be used as part of a system for managing a set of soft sensors. The proposed method relies on lexical rather than semantic analysis by extracting the corpus of words used by a node, removing common connective words, and computing the set intersections among the vocabularies of several nodes. These set intersections can be used to identify distinct domains of knowledge as well as those nodes which have common interests. We further propose that temporal analysis of the movement of words from one node's domain of knowledge to another node can differentiate influencers from disciples in order to convert the undirected graph to a directed graph with conditional probabilities, e.g., a Bayes Net. This research is differentiated from other lexical analysis based on document streams in that it seeks to manage the data acquisition rather than process streams for topics of interest.