- Keywords: Information extraction, recall, completeness
- TL;DR: Information extraction usually estimates precision but does not know about recall; we propose to fix this.
- Abstract: Information extraction from text, IE for short, is the backbone of automated knowledge base construction. IE usually comes with precision estimates; however, it lacks awareness of recall. This paper introduces and discusses the issue of IE recall estimation and its practical importance. We present RecallIE, a methodology for estimating the possible recall from a given text segment. RecallIE uses distant supervision to estimate from language features whether a passage contains all objects for a given subject-predicate pair. We evaluate RecallIE across various granularities of text, and across various predicates. Our preliminary results indicate that estimating recall is a promising direction and technically feasible.
- Archival status: Archival
- Subject areas: Natural Language Processing, Information Extraction, Knowledge Representation