Expanded Semantic Graph Representation for Matching Related Information of Interest across Free Text Documents

This research proposes an expanded semantic graph definition that serves as a basis for an expanded semantic graph representation and graph matching approach. This representation separates the content and context and adds a number of semantic structures that encapsulate inferred information. The expanded semantic graph approach facilitates finding additional matches, identifying and eliminating poor matches, and prioritizing matches based on how much new knowledge is provided. By focusing on information of interest, doing pre-processing, and reducing processing requirements, the approach is applicable to problems where related information of interest is sought across a massive body of free text documents. Key aspects of the approach include (1) expanding the nodes and edges through inference using DL-Safe rules, abductive hypotheses, and syntactic patterns, (2) separating semantic content into nodes and semantic context into edges, and (3) applying relatedness measures on a node, edge, and sub graph basis. Results from tests using a ground-truthed subset of a large dataset of law enforcement investigator emails illustrate the benefits of these approaches.