NCL HKU-NarrSim at SemEval-2026 Task 4: Aspect-Based Agents and Supervised Contrastive Embeddings for Narrative Similarity

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

Summary

The NCL HKU-NarrSim team presented two distinct approaches for SemEval-2026 Task 4 on Narrative Similarity, which focuses on assessing narrative alignment rather than surface lexical similarity. For Track A, they introduced the Aspect-Based Narrative Similarity Agents (ABNS-Agents), a two-stage, agent-based framework. This system extracts three core narrative aspects using a schema constraint and then employs an LLM decision model for aspect-aligned similarity adjudication. ABNS-Agents achieved a 70.25% accuracy on the test set. For Track B, the team developed Narrative Supervised Contrastive Embeddings (NSConE), which utilizes supervised contrastive learning to model narrative similarity, reaching 68.5% test accuracy. These results highlight the efficacy of both aspect-aligned structured modeling and task-specific supervised contrastive learning in capturing deeper narrative relationships.

Key takeaway

For NLP Engineers developing systems for complex semantic understanding, consider integrating aspect-aligned structured modeling or supervised contrastive learning. If your task requires moving beyond surface lexical matching, like narrative similarity, these methods offer superior performance. Explore agent-based frameworks with LLM decision models for structured reasoning. Alternatively, use task-specific contrastive embeddings to capture deeper semantic relationships effectively.

Key insights

Aspect-based agent frameworks and supervised contrastive learning effectively model narrative similarity beyond surface lexical cues.

Principles

• Narrative similarity requires aspect-aligned modeling.
• Supervised contrastive learning enhances narrative embeddings.
• Schema constraints guide aspect extraction.

Method

ABNS-Agents uses a two-stage process: extract three narrative aspects under schema constraints, then apply an LLM decision model for aspect-aligned similarity adjudication.

In practice

• Implement agent-based systems for complex text analysis.
• Apply supervised contrastive learning for semantic tasks.
• Use LLMs for structured decision-making in NLP.

Topics

• Narrative Similarity
• SemEval-2026
• Agent-Based Models
• Supervised Contrastive Learning
• Large Language Models
• Semantic Understanding

Best for: Research Scientist, AI Scientist, NLP Engineer, Machine Learning 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.
• TF-IDF vs. Embeddings: From Keywords to Semantic Search — Embeddings transform text into numerical vectors, enabling semantic search by mapping meaning to geometric proximity.
• ASPECT:Analogical Semantic Policy Execution via Language Conditioned Transfer — LLMs can dynamically remap task descriptions for zero-shot transfer in RL agents.
• Narrative Theory-Driven LLM Methods for Automatic Story Generation and Understanding: A Survey — LLMs bridge abstract narrative theories with NLP, enabling scalable story generation and understanding despite definitional challenges.
• 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.
• AgREE: Agentic Reasoning for Knowledge Graph Completion on Emerging Entities — AgREE uses agent-based reasoning and iterative retrieval to complete knowledge graphs for emerging entities without training.

Open in AIssential →

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