J-REED: Joint Relation Extraction and Entity Disambiguation

Information extraction (IE) from text sources can either be performed as Model-based IE (i.e, by using a pre-specified domain of target entities and relations) or as Open IE (i.e., with no particular assumptions about the target domain). While Model-based IE has limited coverage, Open IE merely yields triples of surface phrases which are usually not disambiguated into a canonical set of entities and relations. This paper presents J-REED: a joint approach for entity disambiguation and relation extraction that is based on probabilistic graphical models. J-REED merges ideas from both Model-based and Open IE by mapping surface names to a background knowledge base, and by making surface relations as crisp as possible.