Aprendizagem contínua aplicada ao desenvolvimento de um chatbot baseado na arquitetura Sequence-to-Sequence
Loading...
Date
2023-12
Authors
relationships.isAdvisorOf
Franzen, Evandro
relationships.isRefereeOf
Pretto, Fabrício
Dertzbacher, Juliano
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
A evolução das ferramentas de chatbot oferecem oportunidades significativas, permitindo interações automatizadas benéficas para empresas e clientes. Pesquisas voltadas ao Processamento de Linguagem Natural ajudam a melhorar a precisão dessas ferramentas, mas a volatilidade da linguagem apresenta grandes desafios, exigindo treinamento contínuo para evitar obsolescência. O aprendizado contínuo é uma abordagem promissora para aprimorar a adaptação de modelos a ambientes dinâmicos e capacitar a sua evolução com as demandas em constante mudança. Este trabalho possui o objetivo de investigar os benefícios e desafios das técnicas de aprendizado contínuo no desenvolvimento de um chatbot com base na arquitetura Sequence-to-Sequence. Assim possibilitando identificar a possibilidade da implantação de um treinamento adaptativo e progressivo por meio da retroalimentação do modelo com base nas interações dos usuários. Como resultado, o aprendizado contínuo demonstrou resiliência, permitindo a assimilação de conhecimento sem comprometer as informações obtidas anteriormente. Embora a arquitetura Sequence-to-Sequence mostre promissores resultados na abstração de dados sequenciais, a necessidade de treinamentos mais intensivos é crucial para garantir a alta qualidade do modelo. Isso implica que modelos versáteis e flexíveis demandam hardware robusto, dados de alta qualidade e períodos de treinamento prolongados.
The evolution of chatbot tools offers significant opportunities, enabling interactive interactions beneficial to businesses and customers. Research outside of Natural Language Processing helps improve the accuracy of these tools, but the volatility of the language presents major challenges, requiring ongoing training to avoid obsolescence. Continuous learning is a promising approach to improve the adaptation of models to dynamic environments and enable them to evolve with changing demands. This work aims to investigate the benefits and challenges of continuous learning techniques in the development of a chatbot based on the Sequence-to-Sequence architecture. This makes it possible to identify the possibility of implementing adaptive and progressive training through feedback from the model based on user interactions. As a result, continuous learning provides resilience, allowing the assimilation of knowledge without compromising previously obtained information. Although the Sequence-to-Sequence architecture shows promising results in abstracting sequential data, the need for more intensive training is crucial to ensure high model quality. This implies that versatile and flexible models require robust hardware, high-quality data, and extended training periods.
The evolution of chatbot tools offers significant opportunities, enabling interactive interactions beneficial to businesses and customers. Research outside of Natural Language Processing helps improve the accuracy of these tools, but the volatility of the language presents major challenges, requiring ongoing training to avoid obsolescence. Continuous learning is a promising approach to improve the adaptation of models to dynamic environments and enable them to evolve with changing demands. This work aims to investigate the benefits and challenges of continuous learning techniques in the development of a chatbot based on the Sequence-to-Sequence architecture. This makes it possible to identify the possibility of implementing adaptive and progressive training through feedback from the model based on user interactions. As a result, continuous learning provides resilience, allowing the assimilation of knowledge without compromising previously obtained information. Although the Sequence-to-Sequence architecture shows promising results in abstracting sequential data, the need for more intensive training is crucial to ensure high model quality. This implies that versatile and flexible models require robust hardware, high-quality data, and extended training periods.
Description
Keywords
processamento de linguagem natural; sequence-to-sequence; aprendizagem contínua; chatbot; natural language processing; continuous learning
Citation
SCHEIBLER, Cristine. APRENDIZAGEM CONTÍNUA APLICADA AO DESENVOLVIMENTO DE UM CHATBOT BASEADO NA ARQUITETURA SEQUENCE-TO-SEQUENCE. 2023. Monografia (Graduação em Engenharia de Software) – Universidade do Vale do Taquari - Univates, Lajeado, 05 dez. 2023. Disponível em: http://hdl.handle.net/10737/4085.