SoulDGXu/Sentiment-Analysis-Chinese-pytorch

Implements a complete NLP pipeline with Chinese text preprocessing using jieba tokenization, stopword filtering, and pre-trained Word2Vec embeddings to build semantic representations before classification. The BiLSTM+Attention architecture uses a bidirectional encoder with learned attention weights to dynamically focus on different input sequence positions, enabling the model to identify salient sentiment-bearing phrases across variable-length reviews. Built in PyTorch with modular components for data processing, model training, and evaluation metrics (accuracy, precision, recall, F1), featuring balanced binary classification (positive/negative) on 20K Chinese film reviews with separate train/validation/test splits.

127 stars. No commits in the last 6 months.