What I learn from this competition- my first model is a basic model where I used POLITICS as a target for every row of the test dataset and that gives me accuracy around 19.01 ish and so far this is my best one

2. I try to build to simple Machine learning model Using countVectorizer and Tf-Idf embeddings and train a bunch of model Decision tree, Radome forest, Xgboost, these models give a score like 12 13 ish something but with my validation dataset they give 30 40 45 ish so I am not clear why there are different results with validation and test dataset hope so anyone helps me on this

3.Naive Bayes which gives my best score of 19.34 ish I also try to train LSTM but failed to do so I use data cleaning like removing multiple spaces, punctuations, and removing stop words but nothing improve my accuracy, not a single percentage I want to know how you guys approach this I know only top 3 get the prize but at least we get learning cause sharing is learning

Thanks for any help

Anish Jain

@itsanishjain - Twitter