Augmenting Word Space Models for Word Sense Discrimination Using an Automatic Thesaurus

摘要

This paper presents an algorithm for Word Sense Discrimination that divides the global representation of a word into a number of classes by determining for any two occurrences whether they belong to the same sense or not. We rely on the notion that words that are used in similar contexts will have the same or a closely related meaning, thus, given a target word, we group its dependency co-occurrences in a Word Space Model. Each cluster represents a distinct meaning or sense of that word. We experiment with augmenting the bag of words of each cluster of co-occurrences, the dictionary of sense definition, and augmenting both. Then we count the number of intersections of each word of the bag of clustered senses and the bag of the dictionary of senses following the Lesk method. We find an increase in recall and a decrease in precision when augmenting. However, the best resulting F-measure is for the option of augmenting the both dictionary of senses and the bag of words from the clusters.