Negative Seeds and Negative Rules for Boosting the Precision of Relation Extraction within the DARE Framework

The work presented here is an extension of an existing machine learning framework called DARE, a system for learning rules that can be used for extracting instances of relations with different complexity from natural language texts ([XuUsLi07] and [Xu07]). In this paper the term “relation”, in contrast to the intuitive sense of the word, refers to a set of tuples with a certain arity. These tuples represent facts or events about real-world objects and concepts. An example is the 4-ary relation whose tuples express the award winning event of nobel prize laureates. The following tuple is an instance of this relation: