Topic-Modeling

Topic modeling architectures on COVID-19 tweets.

Steps:

• Tweet extraction from pymongo
• Tweet cleaning
• Run topic models

We experiment on different topic modeling architectures:

• LDA
• GSDMM
• BTM
• LDA2Vec
• BERT
• Twitter-LDA (Java implementation)
• Twitter-LDA (Python implementation)

BTM, GSDMM performs quite well for short text corpus like tweets. We intend to experiment on other models and show topic-coherence from the ouptput.

Right now, Twitter-LDA is the winner. We have implemented both JAVA and Python version of the algorithm. We are workng on the performance metrics for evaluating the quality of our topic.