Parallelized Power Iteration Clustering of Nouns and Verbs using Subject-Verb and Verb-Object Pairs

We explore the use of Power Iteration Clustering for large-scale clustering of nouns and verbs, using subject-verb-object triple relations. We have implemented a parallelized version of PIC, which can efficiently handle clustering of hundreds of thousands of sparse “documents.” We tested our implementation on clustering of over 1 million noun phrases, and over 650K verb phrases. We also evaluate our verb clustering quantitatively using verb categories from VerbNet as our “gold standard” clusters.