Domain-Adapted Retrieval for In-Context Annotation of Pedagogical Dialogue Acts

1. 1Fine-tune a lightweight embedding model on tutoring corpora
2. 2Index dialogues at the utterance level to retrieve labeled few-shot demonstrations
3. 3Integrate the domain-adapted retrieval pipeline with a generative model for in-context annotation
4. 4Evaluate the pipeline across multiple real-world tutoring datasets

NLP researchers and AI engineers working on LLMs and educational technology can benefit from this approach to improve annotation accuracy and efficiency

💡 Domain-adapted retrieval can improve the accuracy of in-context annotation of pedagogical dialogue acts without requiring fine-tuning of the generative model

📚 Improve pedagogical dialogue annotation with domain-adapted retrieval! 👉 Fine-tune embeddings, index utterances, and integrate with generative models