Talks Similarity Dashboard is an experimental web application that explores semantic relationships between academic talk titles and abstracts using natural language processing techniques. This project leverages large-scale language models and dimensionality reduction via t-SNE to map the textual content of talks into a two-dimensional interactive space. The closer two points are, the more similar their content is — enabling users to intuitively explore clustering tendencies, thematic overlaps, or outliers among submissions. While still a prototype, the project serves as a testbed for experimenting with embeddings, visualization strategies, and interaction design in the context of academic data exploration.

For further exploration, please feel free to visit the following link: https://ilas2025-talks.streamlit.app

Notice:

1. Due to screen size limitations, we recommend accessing the application on a computer or tablet for an optimal experience.
2. If the application appears to be in sleep mode, please click “Restart” to reactivate it.