Chinese Text Classification

NLP component of AI education platform (professional work): **Project Context**: - Part of larger education platform at ZhiHui BianJie - Needed to classify Chinese educational content **Technical Work**: - Fine-tuned BERT-base-chinese (110M parameters) for classification - Implemented few-shot learning for categories with limited data - Optimized inference with INT8 quantization and ONNX Runtime - Reduced latency from ~450ms to ~280ms - Built data preprocessing pipeline for Chinese text - Handled class imbalance and data quality issues **What I Learned**: - Chinese NLP has unique challenges (tokenization, character vs. word) - Model optimization trade-offs (accuracy vs. latency) - Production ML deployment (batch vs. real-time inference) - Monitoring model performance degradation **Limitations**: Model performance varied across content types. Continuous improvement was needed. Cannot share detailed metrics or code (company proprietary).