Datasets:

Modalities:
Text
Formats:
json
Languages:
English
Size:
< 1K
ArXiv:
Tags:
math
Libraries:
Datasets
License:
bridge / README.md
rose-e-wang's picture
Update README.md
659a684 verified
metadata
license: cc-by-nc-4.0
task_categories:
  - text-generation
language:
  - en
tags:
  - math
pretty_name: bridge
size_categories:
  - n<1K

TLDR: This dataset is a real-world math tutoring dataset from the NAACL 2024 paper ``Bridging the Novice-Expert Gap via Models of Decision-Making: A Case Study on Remediating Math Mistakes''. The dataset targets scenarios where the student makes a math mistake.

  • c_h is the conversation history
  • c_r is the original tutor's response
  • c_r_ is the experienced teacher's response

Optionally, there is other interesting metadata from our Bridge method:

  • e is the student error type that the experienced teacher identified
  • z_what is the strategy that the experienced teacher wants to use in their response
  • z_why is the intention that the experienced teacher wants to achieve in their response

🌁 Bridging the Novice-Expert Gap via Models of Decision-Making

Paper Link, Code Link

NAACL 2024

Title: Bridging the Novice-Expert Gap via Models of Decision-Making: A Case Study on Remediating Math Mistakes

Authors: Rose E. Wang, Qingyang Zhang, Carly Robinson, Susanna Loeb, Dorottya Demszky

Main Idea: We contribute Bridge 🌁, a method that uses cognitive task analysis to translate an expert's implicit thought process into an explicit decision-making model.

Scaling high-quality tutoring remains a major challenge in education. Due to growing demand, many platforms employ novice tutors who, unlike experienced educators, struggle to address student mistakes and thus fail to seize prime learning opportunities. Our work explores the potential of large language models (LLMs) to close the novice-expert knowledge gap in remediating math mistakes. Bridge 🌁 leverages cognitive task analysis to model an expert's internal decision-making in remediation: Experts internally identify (A) the student's error, (B) a remediation strategy, and (C) their intention before generating a response. We construct a dataset of 700 real tutoring conversations, annotated by experts with their decisions. We evaluate state-of-the-art LLMs on our dataset and find that the expert's decision-making model is critical for LLMs to close the gap: responses from GPT4 with expert decisions (e.g., ``simplify the problem'') are +76% more preferred than without. Additionally, context-sensitive decisions are critical to closing pedagogical gaps: random decisions decrease GPT4's response quality by -97% than expert decisions. Our work shows the potential of embedding expert thought processes in LLM generations to enhance their capability to bridge novice-expert knowledge gaps.

For more information about how the dataset is curated, please check out our codebase: https://github.com/rosewang2008/bridge/, and paper: https://arxiv.org/pdf/2310.10648