Costa Filho, Bento Alves da2024-03-112024-03-112021FIALHO, André de Oliveira. Análise da eficiência de dois modelos de processamento de linguagem natural (PLN) que classificam o sentimento do consumidor de uma instituição financeira. Goiania, 2021. 73 fl. Dissertação (mestrado em Administração) - Centro Universitário Alves Faria, Goiânia, 2021https://repositorio.unialfa.com.br/handle/123456789/84The opinions collected on social media platforms have captivated the attention of academics and professionals from different areas, given the richness that the analysis of these conversations provides in insights obtained from consumers' feelings and they have a strong influence on the brands they consume. On the other hand, the corporate world faces the challenge of adapting to this new dynamic of consumption and getting closer to its target audience. The literature review demonstrated the great challenges of the evolution of these analytical tools when measuring the accuracy of the activity of monitoring feelings and classifying the bias of posts that can contain countless emotions within a single sentence. But there are also foundations in the literature that demonstrate tools that are in constant evolution of learning capable of obtaining predictive assessments and their performance is directly related to their selection, the creation of codes (programming) and the classification of data. This work aimed to investigate the efficiency level of two NLP frameworks, Word2Vec and BERT used by a financial institution to carry out sentiment analysis on six social media platforms. This research carried out a comparison of agreement between two evaluators and the NLP models used in it, using the statistical indicators of agreement Cohen Kappa, Fleiss kappa - k and weighted Kappa coefficient - kw. The results demonstrated a low level of agreement between the analytical models and the evaluators when using a long scale of characteristics of evaluable feelings, which may imply failures in obtaining insights from the managers of this institution.ptEmoçõesConsumidoresAnálise de sentimentos