Participação, linguagem e impacto dos grupos de interesse: análise quantitativa das contribuições recebidas pela Anatel

• Ano de defesa: 2022
• Autor(a) principal: Gomes, Lucas Thevenard
• Orientador(a): Salinas, Natasha Schmitt Caccia
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Palavras-chave em Português: Regulação; Processamento de linguagem natural; Participação Social; Políticas Públicas; Aprendizagem de Máquina; Grupos de Interesse
• Link de acesso: https://hdl.handle.net/10438/33068

The present study seeks to answer the following question: for a participant to propose a contribution in a public consultation of the Brazilian National Telecommunications Agency – Anatel and have a chance to see his request accepted, does he have to know the language used by the agency? That is, to effectively participate in the consultations, to impact the regulatory decisions of Anatel, is it necessary, first of all, to “speak the language” of the agency? The question relates to empirical research that studies social participation in regulatory agencies using theories about the influence of interest groups over public policy formation. A crucial factor for understanding these theories relates to the role of ideas and language in the way interest groups act. To approach this problem, the presentstudy proposes using natural language processing techniques to analyze if the impact of a contribution – that is, if the agency has accepted or not what was proposed – can be explained by the textual content of the contribution. This analysis is done here in two stages. First, machine learning classification models are used to predict Anatel’s answers, using a TF-IDF representation of the contributions. This task is implemented with 3 different algorithms (K-Nearest Neighbors, Random Forest e Support Vector Machines), and all of them achieve high levels of accuracy, correctly predicting the impact of contributions in more than 85% of cases. This result is taken as evidence that the representation is adequate and that the answers given by the agency are sufficiently systematic to be understood in light of the textual content of contributions. Next, textual metrics are used in logistic regression models to explain the impact of the contributions. The results of these regressions show that at Anatel: (i) contributions with a text more similar to that of the agency have more chances to be accepted, (ii) textual metrics based on the vocabulary of the contributions explain a large part of the variance present in the impact data, and (iii) the textual content of the contributions explain most of the differences in impact between the interest group categories. Finally, all findings of this research are evaluated, and the conclusion discusses the limitations of this study, suggesting opportunities for further development of the topic.