Difference between revisions of "HeadEx"

Revision as of 16:15, 29 April 2016

HeadEx: Triple Extraction from Stream of News Headlines on Twitter using n-ary Relations

Abstract Description: The ever-growing datasets published on Linked Data mainly contain encyclopedic information. However, there is a lack of datasets extracted from unstructured real-time sources. News Headlines published on Twitter provide a real-time stream of events. In this paper, we propose an approach for extracting triples, leveraging n-ary relations, from News Headlines on Twitter in real-time. First, we introduce a mechanism for representing n-ary relations and their arguments as a background data model. This representation leverages Levin's classification of English Verbs in \cite{levin_english_1993} to support the use of unstructured text for constructing the background data model and capturing mentions of n-ary relations. Then, we use a learning approach, employing proposed syntactic features derived from parsing, to extract information respecting the data model. As a proof-of-concept, we follow a case study containing three distinct n-ary relations. The results of our experiments are promising and can be used to create timely and structured news headlines dataset.

HeadEx Architecture