Cognitive Episodes in LLM Reasoning Traces Enable Interpretable Human Item Difficulty Prediction

2026-06-26

AI Summary

Existing approaches to item difficulty prediction rely either on manual calibration or solely on item text, lacking interpretable modeling of the cognitive processes involved in problem solving. This work proposes Epi2Diff, a novel framework that, for the first time, parses reasoning traces generated by large reasoning models (LRMs) into sequences of cognitively meaningful “cognitive episodes.” By integrating dynamic features and semantic representations of these episodes, Epi2Diff jointly models human problem-solving cognition. The method significantly outperforms strong baselines across four real-world human difficulty datasets, achieving an average relative improvement of 8.1% over supervised fine-tuned LLMs on SAT classification tasks. Moreover, it reveals that difficult items elicit more effortful, iterative, and goal-directed cognitive dynamics, demonstrating both high performance and strong interpretability.

Abstract

Predicting human item difficulty is central to educational assessment, where reliable estimates support fairness and effective test construction. Existing methods often depend on costly human calibration or item-level textual representations, providing limited evidence about the cognitive processes that make items difficult. We argue that difficulty should be viewed not only as a property of item text, but also as an observable consequence of the problem-solving burden an item induces. Large Reasoning Models (LRMs) offer scalable process evidence through reasoning traces, but such evidence must be structured to support interpretable modeling. To this end, we introduce Epi2Diff (Episode to Difficulty), a framework that maps LRM reasoning traces into cognitively grounded episode sequences. These episodes group trace segments into functional problem-solving states, enabling difficulty to be modeled through reasoning scale, effort allocation, and state transitions. Epi2Diff extracts compact episode-dynamic features and combines them with semantic item representations for human difficulty prediction. Experiments on four real-world human difficulty datasets show that Epi2Diff consistently outperforms strong baselines, including fine-tuned small language models, LLM in-context learning, and supervised LLM adaptation. On SAT-derived classification benchmarks, Epi2Diff achieves an 8.1% average relative gain over supervised LLM fine-tuning baselines. Further analyses show that harder items induce more effortful, iterative, and implementation-centered episode dynamics, rather than merely longer responses. These results demonstrate that cognitive episodes in LRM reasoning traces provide a predictive and interpretable process representation for human item difficulty, offering a new lens for educational measurement with reasoning models.

Problem

• human item difficulty
• reasoning traces
• cognitive episodes
• educational assessment
• Large Reasoning Models

Innovation

• cognitive episodes
• reasoning traces
• item difficulty prediction
• Large Reasoning Models
• interpretable modeling