Application of natural language processing for the recognition of obesity topics in the discourses of Argentinean Twitter users Languages in cultural perspectives: practices, discourses, cognition

HALUSZKA, EUGENIA; NICLIS, CAMILA; PAREJA LORA ANTONIO; ABALLAY, LAURA ROSANA

Conferencia; Languages in cultural perspectives: practices, discourses, cognition; 2023

University of Applied Sciences in Konin

Obesity is a chronic, heterogeneous and multifactorial disease (De Girolami, 1999). Its prevalence has increased globally and locally in recent years due to the influence of different factors, adding to the lack of a comprehensive health policy approach that considers the sociocultural aspects of the problem. The determinants and social representations (SR) of obesity can help us deepen our knowledge about this problem (Moscovici, 1979). Social networks activate new processes of social interaction and enable the construction of SRs. A diversity of discourses is shared on social networks and may allow us to identify ways of thinking as cultural reflections of this time.Our aim is to recognise the topics of discussion on obesity widely shared on Twitter-Argentina, by means of Natural Language Processing techniques.First, 134,766 tweets discussing obesity were collected between August 2021 and July 2022. Then, a geolocation filter was applied to remove non-Argentinian messages, obtaining a final sample of 48,149 tweets, whose body was then processed. Finally, two different models (Latent Dirichlet Allocation and K-means) were developed to identify topics and clusters, respectively, aggregations in general henceforth. The first 15 most relevant words of each aggregation obtained with both techniques were reviewed. Eventually, the application of K-means provided crispier aggregations. In this case, 21 clusters were identified after a theoretical saturation (Glaser & Strauss, 1967), and then a cluster was assigned to each tweet depending on the words that it contained. Subsequently, 10 tweets of each cluster were randomly sampled to analyze the use of the cluster words in context and identify the conceptual link among them (the cluster theme). This allowed naming each cluster by means of the cluster theme in relation to obesity. Thus, the main themes were hate speech, gender-related discourse, self-perception and body image, demonisation of carbohydrate-rich foods, consequences of obesity, and habits and lifestyle. These aggregations, together with their corresponding themes, have allowed us to understand obesity as a sociocultural phenomenon, the Argentinian culture-specific discourses around it, and their links with health. We firmly believe that this valuable information will enlighten the planning of future social-political actions to address this health-disease process.