AI-JAM Korea 2020: NLP-Based YouTube Comment Analysis
Gold Medal for an NLP model that clusters real YouTube comments
Overview
At AI-JAM Korea 2020, our team built a system that clusters YouTube comments with natural language processing. We won the Gold Medal.
What we did
- Scraped real comments from YouTube with Python.
- Tokenized the Korean text with KoNLPy, embedded it with FastText, and trained clustering models in TensorFlow.
I worked on the overall algorithm, wrote the scraper and the tokenization, and gave the final presentation.
Code
The code is on GitHub.
Results
Gold Medal at AI-JAM Korea 2020.
Nouns from each comment embedded as 300-dimensional vectors
Final clusters, reduced with t-SNE and plotted with Bokeh