YNU-HPCC at SemEval-2026 Task 4: Narrative Similarity via Multi-Perspective E5-Mistral and Embedding Routing

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, quick

Summary

The YNU-HPCC team's system for SemEval-2026 Task 4 addresses narrative story similarity and representation learning. This task requires distinguishing underlying structural fabula from surface lexical overlaps in long narratives across abstract theme, course of action, and outcomes. Their hybrid architecture decouples retrieval and ranking. For Track A, an instruction-tuned E5-Mistral-7B model handles clear cases, while ambiguous samples are routed to a Gemini-3-Flash reasoner. Track B leverages Gemini-Embedding-001 with structure-preserving chunking and All-But-The-Top (ABTT) inference. The system ranked 5th in Track A and 2nd in Track B, demonstrating effective balance between local instruction following and global generalization.

Key takeaway

For NLP engineers developing systems for nuanced narrative similarity or long-context text analysis, consider a hybrid architecture that dynamically routes samples to specialized models based on complexity. This approach, demonstrated by YNU-HPCC's competitive SemEval-2026 results, can improve accuracy by balancing specific instruction following with broader semantic generalization.

Key insights

Dynamic routing and hybrid architectures enhance narrative similarity detection.

Principles

• Distinguish fabula from surface lexical overlaps.
• Decouple retrieval and ranking tasks.
• Balance local instruction following with global generalization.

Method

A hybrid architecture dynamically routes clear narrative similarity cases to E5-Mistral-7B and ambiguous ones to Gemini-3-Flash, leveraging Gemini-Embedding-001 for Track B.

In practice

• Use instruction-tuned models for clear cases.
• Employ powerful reasoners for hard samples.
• Apply structure-preserving chunking for long contexts.

Topics

• Narrative Similarity
• SemEval-2026
• E5-Mistral
• Gemini-3-Flash
• Embedding Routing
• Hybrid Architecture

Best for: AI Engineer, Research Scientist, AI Scientist, Machine Learning Engineer, NLP Engineer

Related on AIssential

• ttda704 at SemEval-2026 Task 4: Modeling Narrative Structures via Pseudonymization and Multi-View Sentence Alignment — Using contrastive learning with fine-tuned sentence transformers and multi-view alignment effectively models narrative structures and similarity.
• IdioLink: Retrieving Meaning Beyond Words Across Idiomatic and Literal Expressions — Idiom understanding in retrieval requires semantic abstraction beyond lexical overlap.
• SenFlow: Inter-Sentence Flow Modeling for AI-Generated Text Detection in Hybrid Documents — Modeling inter-sentence dependencies via structured prediction significantly enhances AI-generated text detection in hybrid documents.
• SemEval-2026 Task 4: Narrative Story Similarity and Narrative Representation Learning — The NSNRL task operationalizes narrative similarity for evaluation and identifies areas for automated system improvement.
• A Self Consistency Based Reranking for Narrative Question Answering — Semantic agreement among multiple generated answers improves Narrative Question Answering robustness.
• What They Saw, Not Just Where They Looked: Semantic Scanpath Similarity via VLMs and NLP metric — VLMs and NLP metrics can quantify semantic similarity in eye-tracking scanpaths, complementing traditional spatial analysis.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.