Leveraging Social Media Analytics in Engineering Education Research

The purpose of this method paper is to explore how social media can be leveraged in engineering education research and provide a step-by-step method for social media analytics. Social media platforms serve as a space for people to share their thoughts, feelings, and experiences. These platforms give space to marginalized voices (although censorship of these voices still occur) and are underutilized data sources to understand marginalized groups in engineering education research. For example, in our larger research project focusing on neurodivergent (e.g., ADHD, autism, dyslexia, anxiety) engineering students, published knowledge on what it means to be neurodivergent is limited to deficit framing and language developed by researchers and clinicians. However, a plethora of knowledge and first-hand accounts of neurodivergent experiences exist on social media where emancipatory– as opposed to deficit– language is used to describe their experiences. Researchers should immerse themselves in these online communities as a way to join their conversations and to understand the experience of being neurodivergent and other marginalized experiences in the larger social context rather than just academic contexts. To leverage social media in research, data mining and natural language processing techniques are necessary. In this paper, we detail a natural language processing technique called the latent Dirichlet allocation (LDA) method, which is a probabilistic topic modeling tool that considers various hidden text elements using machine learning. LDA identifies themes and text structures from large amounts of text-based data sets, such as data mined social media content, to categorize the data. We then present an example of how we used LDA in engineering education research. Our work utilized LDA to identify neurodivergent themes discussed on TikTok, a video-based, social media platform. Our analysis leveraged video hashtags (e.g, ADHD, neurodivergent, and neurospicy) to train the LDA model and visualize the topic clusters from the data produced. We then compared our results to our smaller scale qualitative thematic analysis which aligned with our themes generated. These themes included neurodivergent classifications, neurodivergent manifestations, and societal misconceptions.