11.3.1 Text Classification Roadmap: Text In, Label Out

Text classification takes one piece of text and predicts one label, such as sentiment, topic, intent, or risk type.

See the Classification Pipeline First

Always build a baseline before a complex model. Most classification problems fail because labels are vague or examples are skewed.

Run a Keyword Baseline

texts = ["great course and clear examples", "confusing setup error"]
positive_words = {"great", "clear", "good", "useful"}

for text in texts:
    score = sum(word in positive_words for word in text.split())
    label = "positive" if score > 0 else "needs_review"
    print(label, "-", text)

Expected output:

positive - great course and clear examples
needs_review - confusing setup error

Simple baselines are not the final model, but they expose label rules and failure cases quickly.

Learn in This Order

Step	Read	Practice Output
1	Traditional methods	Build TF-IDF or keyword baseline
2	Deep learning methods	Compare embeddings, pooling, CNN/RNN/Transformer features
3	Project practice	Track split, metrics, label ambiguity, and error samples

Pass Check

You pass this chapter when you can train or simulate a classifier, report accuracy/F1, and explain at least one ambiguous label case.