text-classification

Text Classification

A content moderation system receives user-generated text that needs categorization. Each post requires preprocessing (tokenization, stop-word removal), feature extraction (word frequencies, text length metrics), classification into a category, and confidence scoring so low-confidence predictions get routed to human review.

Pipeline

[txc_preprocess]
     |
     v
[txc_extract_features]
     |
     v
[txc_classify]
     |
     v
[txc_confidence]

Workflow inputs: text, categories

Workers

ClassifyWorker (task: txc_classify)

• Writes prediction, scores

ConfidenceWorker (task: txc_confidence)

• Clamps with math.max(), formats output strings, filters with predicates, maps to primitives for aggregation
• Writes confidence, margin, reliable

ExtractFeaturesWorker (task: txc_extract_features)

• Clamps with math.max(), matches against regex, formats output strings
• Writes features

PreprocessWorker (task: txc_preprocess)

• Lowercases strings, trims whitespace
• Writes cleanedText, tokenCount

---

8 tests | Workflow: txc_text_classification | Timeout: 60s

See RUNNING.md for setup and usage.