Week 14 · Machine-Learning Text Classification 机器学习分类

Dec 14, 2026 · ← Course home

Planned session. Can a machine tell a news report from a short story? A translation from an original? We train classifiers on the feature tables built last week — authorship, genre, translated vs. original — using zero-code tools. More important than the training is the evaluation: train/test splits, accuracy, precision and recall, and the honest question of what the model actually learned (and what it merely memorised).

We revisit the stylometry classics from Week 1 — the Federalist Papers, 红楼梦’s disputed chapters — as worked cases of exactly this pipeline.

Key concepts: supervised classification; training and test data; evaluation metrics; overfitting; feature importance as stylistic evidence.

Slides and materials will be posted here after class.