AlphaLyrae at SemEval-2026 Task 9: Metric Learning and Asymmetric Loss for Chinese Polarization Analysis

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

Summary

AlphaLyrae's system for SemEval-2026 Task 9, "Detecting Online Polarization" in Chinese, tackles implicit language and severe class imbalance. The team proposes a metric learning approach, framing polarization detection as semantic similarity matching to better capture implicit patterns. This involves fine-tuning an ERNIE-3.0 encoder with SoftTriple loss and using ik/iNN retrieval for binary detection (Subtask 1). For multi-label categorization (Subtasks 2 and 3), learned representations are transferred and fine-tuned with Asymmetric Loss. A priority-based stratified cross-validation strategy addresses extreme label skew. Evaluated on the official 1,927-sample test set, the system achieved Macro-F1 scores of 0.9190 (Rank 6) for Polarization Detection, 0.8244 (Rank 5) for Type Classification, and 0.6670 (Rank 4) for Manifestation Identification.

Key takeaway

For NLP Engineers developing content moderation systems for nuanced or implicit language, especially in Chinese, consider integrating metric learning with models like ERNIE-3.0. Your approach should incorporate SoftTriple loss for semantic similarity and Asymmetric Loss for multi-label classification to handle severe class imbalance. This strategy can significantly improve detection accuracy for polarized content that evades traditional methods, as demonstrated by achieving top ranks in SemEval-2026 Task 9.

Key insights

Metric learning and asymmetric loss effectively address implicit language and class imbalance in online polarization detection.

Principles

• Implicit language patterns benefit from semantic similarity matching.
• Asymmetric loss improves multi-label categorization with skewed data.
• Stratified cross-validation is crucial for extreme label skew.

Method

Fine-tune ERNIE-3.0 with SoftTriple loss for metric learning and ik/iNN retrieval for binary detection. Transfer representations for multi-label tasks, fine-tuning with Asymmetric Loss.

In practice

• Use ERNIE-3.0 for Chinese NLP tasks.
• Apply SoftTriple loss for semantic similarity.
• Implement Asymmetric Loss for imbalanced multi-label data.

Topics

• Chinese Polarization Analysis
• Metric Learning
• Asymmetric Loss
• ERNIE-3.0
• SemEval-2026 Task 9
• Class Imbalance

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

Related on AIssential

• mdok-style at SemEval-2026 Task 9: Finetuning LLMs for Multilingual Polarization Detection — Finetuning mid-sized LLMs with QLoRA and augmented data improves multilingual online polarization detection.
• Symmetrization of Loss Functions for Robust Training of Neural Networks in the Presence of Noisy Labels — Symmetrizing multi-class loss functions, especially cross-entropy, yields a unique convex unhinged loss robust to noisy labels.
• Bridging the Semantic-Collaborative Gap: An Asymmetric Graph Architecture for Cold-Start Item Recommendation — An asymmetric graph architecture can bridge the semantic-collaborative gap for cold-start recommendations by aligning intrinsic features with behavioral signals.
• Continuous Language Diffusion as a Decoder-Interface Problem — Continuous language diffusion models succeed by guiding latent states into "decoder-basin" regions where decoders can reliably interpret tokens.
• Aligning Implied Statements for Implicit Hate Speech Generalizability with Context-Bounded Semi-hard Negative Mining — Aligning implied statements with context-bounded semi-hard negative mining improves implicit hate speech detection generalizability.
• JSPG: Dynamic Dictionary Filtering via Joint Semantic-Pinyin-Glyph Retrieval for Chinese Contextual ASR — Jointly integrating semantic, pinyin, and glyph features significantly improves Chinese contextual ASR keyword retrieval accuracy.

Open in AIssential →

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