Xiaoyi (Mimi) Chen

A study of categorical sexism in tweets about Kamala Harris from the 2024 U.S. Presidential Election using a domain-adapted a DeBERTa model with continued pretraining and fine-tuning. Detected and classified hierarchical sexism in 1.9 million tweets to understand how backlash theory manifests in the public arena.

An analysis of ideological polarisation in the political podcast ecosystem through a network of hosts and their guests with community detection via both projected and bipartite modularity maximization. Demonstrated how elite actors contribute to affective polarisation in the digital public sphere.

An empirical analysis of the regional and temporal variations in drivers of higher education attainment in 21st century China using fixed effects panel regression

Multi-label genre classification of anime poster images. Evaluated performance of a custom Convolutional Neural Network (CNN), transfer learning (ResNet50), semi-supervised learning, and AutoML methods

A Natural Language Processing (NLP) project that classifies ghost kitchens based on menu item descriptions scraped from food delivery platforms through text embeddings and logistic regression

A multi-class genre classification pipeline for Spotify tracks using engineered audio features with XGBoost. Utilised K-means clustering to explore underlying structure in the data, revealing genre overlaps and latent groupings.