KristiyanVachev/Question-Generation

Decomposes question generation into modular steps: keyword identification via Gaussian Naive Bayes classification with spaCy part-of-speech tagging, cloze-style question formation, and distractor generation using word embeddings with cosine similarity filtering. Trains on SQuAD dataset with engineered features (POS tags, named entities, shape patterns) to classify candidate answer words, then leverages semantic similarity to produce plausible incorrect options matched by linguistic properties.

493 stars. No commits in the last 6 months.