chinese_text_cnn

Why we included this project

This is a compact PyTorch implementation of the TextCNN architecture from Kim's 2014 sentence classification paper, built for Chinese sentiment analysis. It is a good fit if you want to see how the model behaves on Chinese text without wading through a larger framework. The repo covers four embedding setups, random initialization, static pretrained vectors, fine-tuned vectors, and a multichannel mix, and the README reports test accuracy for each so you can compare them directly. The code also handles the surrounding plumbing: jieba tokenization, loading pretrained Chinese word vectors, and early stopping during training. For a simple Chinese classifier, it works as a readable baseline before moving to transformer-based models, and it is a clear teaching example of how convolutional filters operate over embedded text.