Epi2Diff framework: episode-induced process representations combined with semantic embeddings for item difficulty prediction¶
Research Method · All research methods
Evidence · 3 claims (3 for) · 1 study (1 design),q2· 0 of 1 report an effect size · 3 claims rest on one study
Description¶
Epi2Diff (Episode to Difficulty) converts model-generated reasoning traces into sentence-level episode sequences and extracts compact features capturing reasoning effort. Its episode-induced process representation "combines three complementary groups of features: length features capturing the overall scale of the reasoning process, episode-distribution features characterizing its composition across episode types, and transition features encoding its local sequential organization". These 83 process features are concatenated with a Sentence-BERT item embedding, averaged across direct, weak, medium, and strong prompting profiles, and fed to a supervised predictor (XGBoost across all datasets). Code is released at the article's GitHub website.
Accounts¶
- Epi2Diff framework: episode-induced process representations combined with semantic embeddings for item difficulty prediction: Epi2Diff (Episode to Difficulty) converts model-generated reasoning traces into sentence-level episode sequences and extracts compact features capturing reasoning effort. Its episode-induced process representation "combines three complementary groups of features: length features capturing the overall scale of the reasoning process, episode-distribution features characterizing its composition across episode types, and transition features encoding its local sequential organization". These 83 process features are concatenated with a Sentence-BERT item embedding, averaged across direct, weak, medium, and strong prompting profiles, and fed to a supervised predictor (XGBoost across all datasets). Code is released at the article's GitHub website. (Chenguang Wang et al. (2026))
Claims¶
- Epi2Diff achieves the best performance across all four item-difficulty test sets and all reported metrics, outperforming SLM fine-tuning, LLM prompting, and supervised LLM adaptation baselines [+M]
- Combining semantic and process features yields the best difficulty prediction; semantic embeddings alone trail the full model by 7.4–12.1 macro-F1 points on SAT Math [+M]
- Reasoning-process features are particularly useful when difficulty depends on multi-step problem solving, and the representation also generalizes to text-based assessment items [+W]
Related Research Methods¶
-
Key Sources¶
- Chenguang Wang, Ming Li, Xinyue Zeng, Zhuochun Li, Hong Jiao, Tianyi Zhou, Dawei Zhou. (2026). Cognitive Episodes in LLM Reasoning Traces Enable Interpretable Human Item Difficulty Prediction. https://arxiv.org/abs/2606.28186
Page metadata
| Field | Value |
|---|---|
type |
research-method |
id |
epi2diff-framework-episode-induced-process-representations-combined-with-semantic-embeddin |
title |
Epi2Diff framework: episode-induced process representations combined with semantic embeddings for item difficulty prediction |
description |
Epi2Diff (Episode to Difficulty) converts model-generated reasoning traces into sentence-level episode sequences and extracts compact features capturing reasoning effort. |
status |
review |
generated |
by process:settle-candidates on 2026-10-11 |
trust |
unverified — no one has confirmed this page's content |
Frontmatter is the page's machine-readable record — see the Schema & Guide.