A Chinese unsupervised word sense disambiguation method based on semantic vector

The supervise machine learning word sense disambiguation method need to annotate the words of the training corpus, in order to overcome the data sparseness problem to achieve the good word sense disambiguation effect we must establish a large-scale marked Corpus, but obtaining the marked corpus requires high artificial price. Against this problem this paper proposes an unsupervised learning method without manual annotation. Firstly we mine the feature words based on PMI (Point-wise Mutual Information) and Z test, defining the v words to describe a certain sense of polysemy, and then calculating the similarity between sense words and the features of polysemy in the context to determine the correct sense of the polysemy. This paper disambiguates ten typical polysemy, and experimental results prove that the method is effective.