gaussic/text-classification-cnn-rnn

Implements character-level CNN and RNN architectures using TensorFlow 1.3+ with Conv1D operations and multi-layer GRU/LSTM cells for sequence modeling. Provides complete preprocessing pipeline including vocabulary building, fixed-length sequence padding (600 characters), and batch iteration with shuffling for the THUCNews dataset (10 categories, 65K training samples). Achieves 96%+ test accuracy on Chinese news classification with detailed evaluation metrics including per-category precision/recall and confusion matrices.

4,294 stars. No commits in the last 6 months.