Word Sense Disambiguation Research Based On Dependency Parsing

In the era when big data and artificial intelligence are rapidly developed,word sense disambiguation(WSD),as a key step in machine translation,information retrieval and information extraction,plays an increasingly critical role in reducing data volume and improving algorithm coverage and accuracy.WSD is the progress of identifying the context-based true meaning expressed by ambiguous words.Choosing features from the context and establishing decision rules for word sense are pressing issues that need to be addressed in WSD.However,WSD methods which mainly relies on semantic relations while neglecting the syntactic structure results in insufficient knowledge for disambiguation.Available disambiguation features cannot be selected and effective decision rules for word sense cannot be established,leading to low disambiguation accuracy.To solve this problem,dependency parsing manner is introduced in this study to suggest a WSD method based on dependency tree.In this manner,contextual features are selected by the path length between word nodes to identify the true meaning of the words by a similarity computation.In this manner,the failure to consider syntactic structure is remedied.To solve the problem of lack of knowledge in a single sentence,the syntactic correlation information between words is also utilized in this study to propose a WSD method based on dependency relation.This method counts the frequency of dependency relations between words in the corpora and then builds a disambiguation knowledgebase,providing evidence for word sense identification.The feature selection method is then improved,and the meaning of words is determined by calculating the sense-context relatedness by using the established disambiguation knowledgebase.Experiments show that proposed WSD methods based on dependency tree and dependency relation could solve deficiencies in syntactic knowledge and improve the accuracy of WSD.The accuracy of our methods founded on multiple datasets is proven to be better than those of other methods.