Multi-Class-Text-Classification-Wines-NLP

Multiclass text classification using two models : Keras Custom Embeddings and Google BERT

Change the file paths according to your directory. Code can be run directly in google colab.

The file test_var.csv stores the final predicted results.

The files country_extract.csv and vintage_extract.csv give some insights and data distributions.

The images also reflect some information ragarding wines.

Two models have been tested:

1. Using Keras' Custom Embedding Layer:

The model was trained from scratch for word embeddings; further propogated to RNN architecture resulting in 77% train accuracy and 67% validation accuracy.

2. Using Tansfer Learning with Google BERT weights:

The model was trained using pre-trained Google BERT word embeddings giving better results with 83% train accuracy and 74% validation accuracy.

Results can probably be futher improved by removing stopwords and then feeding it to Google BERT architecture.