Úbeda Castañeda, María de las Nieves2024-05-202024-05-202023https://hdl.handle.net/20.500.14468/14279Conversational assistants and chatbots have become popular tools that can quickly answer users' requests to help them solve simple tasks. One of the most important goals of a conversational assistant is to understand a user's intent, that is, NLU. To carry out this aim, intent classification is one of the most important tasks. Thus, this thesis aims to design and implement a benchmark to test different models that can be included for a intent recognition system in order to improve a stage that already exists in Aura (Telef´onica's virtual assistant). The resulting module can classify user utterances regarding watch actions. Recent advances in natural language by using labels from a predefined set of intents. The best model achieves an improvement of accuracy concerning the solution that is currently used by the company.eninfo:eu-repo/semantics/openAccesstesis de maestríaIntent classificationMachine LearningNatural Language Processingtransformersclasses