TACCTI: Deep Learning-Based Cultural Capital Theme Identification in STEM Education using Reflective Journals
Author: Gian Carlo Baldonado
Faculty Supervisor: Anagha Kulkarni
Department: Computer Science
The Alma Project addresses challenges faced by historically underrepresented students in STEM education by promoting affirmation of their intersectional identities and cultural capitals through reflective journaling. The project employs a method where students respond to prompts like "Why am I here?" to reconnect with their values and purpose in STEM. These responses are analyzed for cultural capital themes (CCT), such as Attainment, to identify tangible goals. Manual CCT identification is time-consuming, leading to the introduction of TACCTI (Text Analysis and Machine Learning for Cultural Capital Theme Identification) program, utilizing natural language processing and machine learning. However, the baseline model faces challenges due to data scarcity and lack of hyperparameter tuning. To address these, two approaches are proposed: expanding the dataset using pre-labeled essays and fine-tuning more complex pre-trained models like BERT and RoBERTa. Early evaluations show improved performance with larger datasets and complex models, with hyperparameter analysis revealing the significant influence of learning rate and scheduler. This study contributes to enhancing the TACCTI program and supporting underrepresented students in STEM through reflective journaling.