| Abstract [eng] |
This master thesis focuses on the application of transformers in natural language processing (NLP), specifically in the task of dialog state tracking. The work begins with the overview of the literature of transformers (and their derivatives, such as BERT and GPT) and dialog state tracking problem. The thesis also presents novel approaches, including improvements in state identification from text and a hybrid architecture, aiming to achieve better results in dialogue state tracking. Through experiments and analysis, the study evaluates the proposed solutions, explores the impact of various parameters, and discusses future directions in this field. |
