Clustering techniques for open relation extraction

This work investigates clustering techniques for Relation Extraction (RE). Relation Extraction is the task of extracting relationships among named entities (e.g., people, organizations and geo-political entities) from natural language text. We are particularly interested in the open RE scenario, where the number of target relations is too large or even unknown. Our contributions are in two aspects of the clustering process: (1) extraction and weighting of features and (2) scalability. In order to evaluate our techniques in large scale, we propose an automatic evaluation method based on pointwise mutual information. Our preliminary results show that our clustering techniques as well as our evaluation method are promising.