TamilEcho_Political@DravidianLangTech 2026: Hybrid XLM-RoBERTa with Sarcasm-Aware Feature Fusion for Political Multiclass Sentiment Analysis in Tamil X

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

Summary

The TamilEcho system, submitted to the Shared Task on Political Multiclass Sentiment Analysis of Tamil X (Twitter) Comments at DravidianLangTech@ACL 2026, addresses challenges in Tamil social media sentiment analysis, including informal language, sarcasm, emoji-driven sentiment inversion, and class imbalance. This system employs a hybrid architecture that combines contextual representations from XLM-RoBERTa with lexical TF-IDF features and explicit sarcasm-aware emoji features. It also incorporates domain-specific hashtag expansion to enrich political context. To mitigate class imbalance, TamilEcho utilizes inverse-frequency class weighting and label smoothing during training. Experimental results show that this hybrid feature fusion significantly outperforms transformer-only baselines. The system achieved a Macro-F1 score of 0.3559 on the official test set, securing Rank 10 among participating teams, demonstrating the effectiveness of integrating semantic, lexical, and pragmatic cues for fine-grained political sentiment classification in Tamil.

Key takeaway

For NLP Engineers developing sentiment analysis models for low-resource languages or social media, consider adopting a hybrid architecture. You should integrate contextual embeddings like XLM-RoBERTa with lexical features (TF-IDF) and explicit pragmatic cues such as sarcasm-aware emoji features. This approach, demonstrated by TamilEcho's Macro-F1 score of 0.3559, effectively addresses challenges like informal language and sentiment inversion. Additionally, implement class imbalance techniques like inverse-frequency weighting and label smoothing to improve model robustness.

Key insights

Hybrid feature fusion significantly improves political multiclass sentiment analysis in Tamil social media by combining diverse linguistic cues.

Principles

• Combine contextual, lexical, and pragmatic features.
• Address class imbalance with weighting and smoothing.
• Expand hashtags for domain context.

Method

Integrates XLM-RoBERTa contextual embeddings with TF-IDF and sarcasm-aware emoji features. Applies inverse-frequency class weighting and label smoothing, plus domain-specific hashtag expansion.

In practice

• Implement hybrid models for low-resource languages.
• Use emoji features for sentiment inversion.
• Apply class weighting for imbalanced datasets.

Topics

• Political Sentiment Analysis
• Tamil Language Processing
• XLM-RoBERTa
• Hybrid Models
• Feature Fusion
• Class Imbalance

Best for: AI Scientist, NLP Engineer, Research Scientist

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.
• TeamHerald@CHIPSAL 2026: Hate Speech Detection and Sentiment Analysis of Nepali Memes using Transformer-based Architectures and Ensemble Learning — Ensemble strategies for meme analysis perform differently for binary vs. multi-class tasks, requiring task-specific aggregation.
• An End-to-End Hybrid Framework for Rumour Detection in Low-Resources Algerian Dialect — A hybrid framework combining transformer embeddings with classical classifiers enables effective rumour detection in low-resource Algerian dialect.
• Graph-Augmented LLMs for Swiss MP Ideology Prediction — Graph-augmented LLMs can model complex political ideologies from text and relational data.
• Enhancing SDG-Text Classification with Combinatorial Fusion Analysis and Generative AI — Combinatorial Fusion Analysis (CFA) significantly enhances SDG text classification by combining diverse models and synthetic data.
• Learning Word Vectors for Sentiment Analysis: A Python Reproduction — Word vectors can effectively capture both semantic similarity and sentiment polarity by combining unsupervised and supervised learning objectives.

Open in AIssential →

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