Zero-Shot Text Classification

Follow this topic Ask about this topic

Zero-shot text classification studies how to assign texts to predefined categories when no labeled examples are available for the task. Common approaches compare texts with label descriptions using embeddings or natural language inference models, or ask instruction-following language models to classify from prompts; retrieval and clearer label definitions can help distinguish similar categories. These methods support classification in new domains and languages without building a task-specific training set, but results depend strongly on label wording, task difficulty, and evaluation setting. General-purpose models can perform well, yet fine-tuned classifiers often retain an advantage on nuanced categories, and accuracy must be weighed against calibration, latency, and cost.

Latest papers

35 papers7 papers in the last four weeks, up 40% on the four weeks before. 0.0% of all new papers.

Evaluation of Embedding-Based and Generative Methods for LLM-Driven Document Classification: Opportunities and Challenges
Apr 5, 2026Rong Lu, Hao Liu, Song HouEarth Science Text Classification
Learning to Aggregate Zero-Shot LLM Agents for Corporate Disclosure Classification
Mar 21, 2026Kemal KirtacLanguage Model Ensembles Zero-Shot Text Classification
University College London London, United Kingdom
BTZSC: A Benchmark for Zero-Shot Text Classification Across Cross-Encoders, Embedding Models, Rerankers and LLMs
Mar 12, 2026Ilias AarabZero-Shot Text Classification LLM Evaluation
European Central Bank, Frankfurt am Main, Germany
Large Language Models for Biomedical Article Classification
Mar 12, 2026Jakub Proboszcz, Paweł CichoszFew-Shot Text Classification Large Language Model-Based Text Classification
Institute of Computer Science, Warsaw University of Technology, Nowowiejska 15/19, 00-665 Warsaw, Poland
KCLarity at SemEval-2026 Task 6: Encoder and Zero-Shot Approaches to Political Evasion Detection
Mar 6, 2026Archie Sage, Salvatore GrecoPolitical Discourse Analysis Zero-Shot Text Classification
King’s College London

Apr 5, 2026J/K move · Enter open · S save

Evaluation of Embedding-Based and Generative Methods for LLM-Driven Document Classification: Opportunities and Challenges

Rong Lu, Hao Liu, Song Hou

Earth Science Text Classification VLM Evaluation Text Embedding Models Zero-Shot Text Classification LLM Fine-Tuning

Open paper AskPDF arXiv

This work presents a comparative analysis of embedding-based and generative models for classifying geoscience technical documents. Using a multi-disciplinary benchmark dataset, we evaluated the trade-offs between model accuracy, stability, and computational cost. We find that generative Vision-Language Models (VLMs) like Qwen2.5-VL, enhanced with Chain-of-Thought (CoT) prompting, achieve superior zero-shot accuracy (82%) compared to state-of-the-art multimodal embedding models like QQMM (63%). We also demonstrate that while supervised fine-tuning (SFT) can improve VLM performance, it is sensitive to training data imbalance.

Latest papers

Evaluation of Embedding-Based and Generative Methods for LLM-Driven Document Classification: Opportunities and Challenges

Learning to Aggregate Zero-Shot LLM Agents for Corporate Disclosure Classification

BTZSC: A Benchmark for Zero-Shot Text Classification Across Cross-Encoders, Embedding Models, Rerankers and LLMs

Large Language Models for Biomedical Article Classification

KCLarity at SemEval-2026 Task 6: Encoder and Zero-Shot Approaches to Political Evasion Detection