Related article that I wrote, which is at 80K+ views, published in Towards Data Science is here.
Training data was pulled from Vicemo and my own personal feed. Additionally, there's support to count Unicode emojis in the calculations, thanks to the emoji library for Python.