Word Sense Disambiguation for Large Documents Using Neural Network Model

Lexical ambiguity in natural language processing is a live problem and needs to be resolved. The lexical ambiguity mainly occurred because of polysemous words. Because of polysemous words, Machine fails to understand the user queries and generates ambiguous results. This research aims to rectify the ambiguous queries with corrected queries by using supervised machine learning approach. The input query is preprocessed first with sentence splitting, tokenization, lemmatization, and stemming. From the filtered output the ambiguous words are detected and context information is stored into local generated data structures. The adaptive word vector is constructed with respect to the ambiguous word and stored context information. While constructing adaptive word vector the mean-similarity and max-similarity of the ambiguous words is considered. The adaptive word vector is provided as an input to the supervised neural network model for the classification. The corrected sense values are available at the out layer. These sense values are mapped with freely available lexical dataset WordNet for getting the most correct meaning of ambiguous words. The ultimate goal of this research is to provide unambiguous communication in between human and machine.