Document structure-driven investigative information retrieval

Abstract

Data-driven investigations are increasingly dealing with non-moderated, non-standard and even manipulated information Whether the field in question is journalism, law enforcement, or insurance fraud it is becoming more and more difficult for investigators to verify the outcomes of various black-box systems To contribute to this need of discovery methods that can be used for verification, we introduce a methodology for document structure-driven investigative information retrieval (InvIR) InvIR is defined as a subtask of exploratory IR, where transparency and reasoning take centre stage The aim of InvIR is to facilitate the verification and discovery of facts from data and the communication of those facts to others From a technical perspective, the methodology applies recent work from structured document retrieval (SDR) concerned with formal retrieval constraints and information content-based field weighting (ICFW) Using ICFW, the paper establishes the concept of relevance structures to describe the document structure-based relevance of documents These contexts are then used to help the user navigate during their discovery process and to rank entities of interest The proposed methodology is evaluated using a prototype search system called Relevance Structure-based Entity Ranker (RSER) in order to demonstrate its the feasibility This methodology represents an interesting and important research direction in a world where transparency is becoming more vital than ever.