Dataset:

• NTC-SCV: https://github.com/congnghia0609/ntc-scv

Pipeline:

• Text Preprocessing: Cleaning, Stemming, Contraction Removal, Special Char Removal
• EDA: Word2Vec Embedding Enrichment, Misspell Removal, Feature Creation
• Representation: Tokenization, Text To Sequence, Padding Sequences
• Modelling: Bi-LSTM/ GRU, Attention, Capsule
• Deployment: Prediction and Model Evaluation

Result:

• Accuracy (test set): 0.8765
• Compare:
  • BOW: 0.8779
  • TF-IDF: 0.8822