University of Twente Student Theses


“You are what you Tweet!” : Analysing Pro-Anorexia and Pro-Recovery Messages On Twitter Using Transformer-Based Text Mining Applications

Heinke, Julius (2023) “You are what you Tweet!” : Analysing Pro-Anorexia and Pro-Recovery Messages On Twitter Using Transformer-Based Text Mining Applications.

[img] PDF
Abstract:Background & Objective. “Pro-anorexia” (PA) communities that encourage disordered eating are a common phenomenon on social media sites like Twitter. In response, “pro-recovery” (PR) communities that postulate a recovery-oriented perspective emerged. Past research has struggled with the vast quantity of social media content, limiting their insights into these communities. To better understand PA/PR community content and dynamics, this study utilized modern transformer-based text mining methods and metadata to analyse all English PA/PR-related tweets posted until February 1, 2023. Method. A total of 288.773 PA and 17.839 PR tweets were scraped using a specified set of hashtags, and analysed on their metadata, sentiment, and discussed topics using novel roBERTa-based-sentiment and BERTopic analyses in Python. Social metrics (likes, retweets etc.), overall findings, and changes over time in popularity, topics, and sentiment were analysed and compared. Results. More than 16 times as many PA than PR tweets were found. PA tweeters had more likes, retweets, quotes and replies. PR users posted more tweets, followed more accounts, and engaged in broader content. PA sentiment was both more often positive and negative. PR sentiment was more neutral and more stable. For both communities, higher tweet sentiment coincided with greater tweet frequency in one community and lower sentiment and fewer tweets in the other. PA topics were focussed on anorexia and over time shifted from inspiration/connection to no current primary topic. PR had a broader focus and over time showed a shift from discussing eating habits to media content. Conclusion. Based on metadata and topics, the PA community appeared to show a stronger focus on social interaction within the community and anorexia-related topics, while also being more secluded from other communities. Sentiment and topics indicate that PA tweets tended to more often use polarising language and weight loss encouraging topics. The PR community appears more neutral, with a larger focus outside AN. Future research is needed to verify these findings and test methods of intervention.
Item Type:Essay (Master)
Faculty:BMS: Behavioural, Management and Social Sciences
Subject:77 psychology
Programme:Psychology MSc (66604)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page