Date
AI Overview
Why it matters
Automated scoring systems can evaluate open-ended student responses efficiently, but they remain vulnerable to error, bias, and fairness concerns. Improving transparency and interpretability has become an imperative, yet practical interpretable systems are rarely used in real-world assessments.
Study at a glance
- Question: Can an interpretable automated scoring framework score open-ended student responses accurately while aligning with human judgment?
- Setting and data: ASAP-SAS dataset with English constructed responses from US schoolchildren across 10 open-ended exam questions in science, reading informational text, and literature.
- Method: System development and model evaluation comparing scoring accuracy and human-alignment experiments (ordinal logistic regression).
- AI tools: ANALYTICSCORE, GPT-4.1, GPT-4.1-mini, o4-mini, Llama-3.1-8B-Instruct, DeBERTa
Key findings
- The interpretable framework scored within 0.06 quadratic weighted kappa of state-of-the-art uninterpretable models across 10 assessment items and outperformed multiple automated scoring baselines.
- Distilling the featurizer into an open-source model substantially improved its scoring accuracy, increasing its average quadratic weighted kappa by 0.13 to match larger models.
- The model's identification of statements in student work aligned closely with human judgments, reaching kappa scores between 0.72 and 0.90 across subject areas.
Summary drafted with AI from the full paper and may contain errors. See the full paper for details.
AI-driven automated scoring systems offer scalable and efficient means of evaluating complex student-generated responses. Yet, despite increasing demand for transparency and interpretability, the field has yet to develop a widely accepted solution for interpretable automated scoring to be used in large-scale real-world assessments. This work takes a principled approach to address this challenge. We analyze the needs and potential benefits of interpretable automated scoring for various assessment stakeholders and develop four principles of interpretability -- Faithfulness, Groundedness, Traceability, and Interchangeability (FGTI) -- targeted at those needs. To illustrate the feasibility of implementing these principles, we develop the AnalyticScore framework for short answer scoring as a baseline reference framework for future research. AnalyticScore operates by (1) extracting explicitly identifiable elements of the responses, (2) featurizing each response into human-interpretable values using LLMs, and (3) applying an intuitive ordinal logistic regression model for scoring. In terms of scoring accuracy, AnalyticScore outperforms many uninterpretable scoring methods, and is within only 0.06 QWK of the uninterpretable SOTA on average across 10 items from the ASAP-SAS dataset. By comparing against human annotators conducting the same featurization task, we further demonstrate that the featurization behavior of AnalyticScore aligns well with that of humans.
Research Categories
What is the application?
Who is the user?
Why use AI?
Study design
Publication Source
Publisher
arXiv
