Natural language processing and machine learning in the categorization of scientific papers: a study around “cultural heritage”

• Autor(a) principal: Jesus, Ananda Fernanda de
• Data de Publicação: 2023
• Outros Autores: Triques, Maria Lígia, Segundo, José Eduardo Santarem, Albuquerque, Ana Cristina de
• Tipo de documento: Artigo
• Idioma: por
• Título da fonte: Revista Ibero-americana de Ciência da Informação
• Texto Completo: https://periodicos.unb.br/index.php/RICI/article/view/47537
• topic: Aprendizado de máquina; Processamento de linguagem natural; Algoritmo de rede neural; Algoritmo de clusterização hierárquica; Patrimônio cultural; Aprendizaje automático; Procesamiento natural del linguaje; Algoritmo de red neuronal; Algoritmo de agrupamiento jerárquico; Patrimonio cultural; Machine learning; Natural language processing; Neural network algorithm; Hierarchical clustering algorithm; Cultural heritage

Aims to verify the potential of applying Natural Language Processing (NLP) and Machine Learning (ML) techniques in the thematic categorization of scientific articles on the theme “cultural heritage” from two situations in which categories are established a priori and later. Applied research is developed, with quantitative and qualitative results, where the first corpus consisting of scientific articles in Portuguese, on a thematic basis of Information Science, manually selected and categorized; and the second corpus, composed of scientific articles in English retrieved from the Web of Science, automatically categorized by search strategies and application of Booleans. Both were submitted to two categorization test procedures (supervised and unsupervised algorithm). The results show that in both, the participation of the researcher is essential in defining the representativeness of the chosen sample, and this has an impact on the precision and accuracy of the applied algorithms. The importance of detailing and rigor in the pre-processing of data and sample size is highlighted, however, it is emphasized that, in the case of this study, only a larger volume of data did not guarantee that the results were representative from the point of view of the domain studied, which warns that there are always multidisciplinary discussions and analyzes that allow verifying and readjusting the sample parameters.