2009 CCPR Keynotes

Minimally supervised machine learning methods based on bootstrapping are an attractive approach to advanced information extraction. Complex patterns signalling relevant semantic relations in free texts can be detected in this way. However, the potential and limitations of such methods are not yet sufficiently understood. We have systematically analyzed a bootstrapping approach. The starting point of the analysis is a pattern-learning graph, which is a subgraph of the bipartite graph representing all connections between linguistic patterns and relation instances exhibited by the data. It is shown that the performance of such general learning framework for actual tasks is dependent on certain properties of the data and on the seed construction. However, the greatest improvements can be obtained through the systematic learning of negative patterns.