Word Sense Induction & Disambiguation Using Hierarchical Random Graphs

摘要: Graph-based methods have gained attention in many areas of Natural Language Processing (NLP) including Word Sense Disambiguation (WSD), text summarization, keyword extraction and others. Most the work these formulate their problem a graph-based setting apply unsupervised graph clustering to obtain set clusters. Recent studies suggest that graphs often exhibit hierarchical structure goes beyond simple flat clustering. This paper presents an method for inferring grouping senses polysemous word. The inferred structures are applied word sense disambiguation, where we show our performs significantly better than traditional agglomerative yielding improvements over state-of-the-art WSD systems based on induction.