Semi-supervised Bootstrapping of Relation Triples from the Web, Query Languages over these Noisy Triples, their Semantics, and Query Execution Systems

Information Extraction (IE) is the process of retrieving structured information from unstructured text. IE has traditionally relied on extended human interposition to extract small set of predefined relations from the corpus. Now with Web coming in to picture, methods and goals of IE have taken a slight detour, with increasing focus on following challenges 1. Domain independent/Open Information Extraction (minimizeobviate human intervention) 2. Leverage redundant (implicitly structured) information on the ever growing web. 3. Capable of dealing with large level of ambiguity/noise in Natural Language. 4. Scalability of the IE systems to web-size dynamic corpora 5. Ability to Extract/Integrate Relations of arbitrary-arity 6. Ability to efficiently retrieve answers to arbitrary structured queries on the extracted data/schema In the remaining portion of this document, I intend to highlight some of the significant developments 1 made in this area starting from domain specific, hand-tagged IE systems to state of the art selfsupervised Open IE framework and beyond. Instead of describing the algorithms used in the IE Systems, I have tried to enlist some significant characteristics inferred while studying these systems.