Georgian homonym classification using Artificial Intelligence techniques

Author: Davit Melikidze
Co-authors: Davit Melikidze
Annotation:

The project discusses homonym classification in Georgian language. The goal is to give the right definition of a homonym given context. In natural language processing it is important to analyze text. Computer programs require numerical data. Therefore, Word2Vec model is used for word vectorization. For homonym classification two models are used: Support Vector Machines and Convolutional Neural Network. In the project only one homonym is considered: “Bar” for which we’ve given 4 definitions: “a shovel,” “a plain or a flatland,” “a café-bar” and everything else. For training We’ve used 4000 sentences, each labeled accordingly. As a result, We’ve achieved 80% accuracy on newly given sentences (test data) for homonym classification.