Unsupervised word sense disambiguation method based on word vector

Word Sense Disambiguation (WSD) is an open problem in Natural Language Processing which is particularly challenging and useful in the unsupervised setting where all the words in any given text need to be disambiguated without using any labeled data. A disambiguated method was proposed which based on word vector and typical training method such as CBOW and Skip-gram model under sequence training or Negative sampling. The sequence training process and probability of typical models in word2vec are derived. 7 key parameters for training process are discussed about default and optimization. In order to achieve higher accuracy, different corpora are used and the parameters of the training dictionary are adjusted to optimize the results of word disambiguation and improve accuracy and rationality. The results are analyzed and the training model of different parameters and the results of WSD of the corpus are discussed.