Automated Clustering and Extraction of Distinctive Words in Legal Documents

Abstract

An agent is described to automatically organize and annotate a collection of text documents data set. Thisagent clusters the data collection and generates a setof distinctive word labels for each cluster of documents,all entirely autonomously. The agent is based on the useof Principal Direction Divisive Partitioning and k-means,applied to both the documents and the words, using a bag ofwords model. The agent is capable of extracting the wordsthat are most useful in distinguishing among the documents.All this processing by the agent occurs without input froma human user, except to specify the original document set.The agent is illustrated with a collection of alcohol lawsenacted in the 50 states of the U.S.