Temporal Text Classification with Large Language Models

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

Summary

Nishat Raihan and Marcos Zampieri conducted the first systematic evaluation of Large Language Models (LLMs) for Temporal Text Classification (TTC), which involves estimating text publication dates based on linguistic changes. Their study, presented at the 6th International Conference on Natural Language Processing for the Digital Humanities in July 2026, assessed leading proprietary models like Claude 3.5, GPT-4o, and Gemini 1.5, alongside open-source alternatives including LLaMA 3.2, Gemma 2, Mistral, and Nemotron 4. They utilized three historical corpora, two in English and one in Portuguese, testing zero-shot, few-shot prompting, and fine-tuning settings. Results indicate proprietary LLMs perform well, particularly with few-shot prompting. While fine-tuning significantly improves open-source models, they still do not match the performance of proprietary LLMs for TTC tasks.

Key takeaway

For NLP Engineers developing historical text analysis systems, consider proprietary LLMs like Claude 3.5 or GPT-4o for Temporal Text Classification, especially when few-shot prompting is feasible. If open-source models are necessary, prioritize fine-tuning LLaMA 3.2 or Gemma 2 to improve accuracy, though anticipate they may not fully match proprietary model performance. Evaluate the trade-offs between cost, data privacy, and performance for your specific application.

Key insights

Proprietary LLMs excel at Temporal Text Classification, especially with few-shot prompting, outperforming fine-tuned open-source models.

Principles

• Proprietary LLMs show strong TTC performance.
• Few-shot prompting enhances proprietary LLM efficacy.
• Fine-tuning improves open-source LLMs.

Method

The study systematically evaluated proprietary and open-source LLMs on Temporal Text Classification using three historical corpora. It tested zero-shot, few-shot prompting, and fine-tuning settings to assess performance.

In practice

• Employ few-shot prompting for proprietary LLMs.
• Consider fine-tuning open-source models for TTC.

Topics

• Temporal Text Classification
• Large Language Models
• LLM Evaluation
• Few-shot Prompting
• Fine-tuning
• Historical Corpora

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

Related on AIssential

• Scikit-LLM vs. Traditional Text Classifiers: When Should You Use an LLM? — LLMs offer superior zero-shot text classification accuracy and competitive latency, especially with limited data.
• System Report for CCL25-Eval Task 5: New Dataset and LoRA-Fine-Tuned Qwen2.5 — Domain-specific datasets and LoRA fine-tuning significantly enhance LLM performance in specialized tasks like classical poetry analysis.
• Supervision versus Demonstration-Based In-Context Learning for Multiword Expression Classification — In-context learning for Turkish LVCs is highly prompt-sensitive, with few-shot demonstrations improving LLM calibration to match supervised baselines.
• Soft-Prompt Tuning for Fair and Efficient LLM Benchmark Evaluation — Soft-prompt tuning efficiently reveals true LLM knowledge by overcoming format-following limitations in benchmarks.
• Understanding LLM Behavior in Multi-Target Cross-Lingual Summarization — LLMs perform cross-lingual summarization and translation jointly in later layers, not sequentially.
• Prompt Repetition: Google's Surprisingly Strong Baseline for Non-Reasoning LLMs — Repeating prompts can dramatically improve LLM accuracy by providing a "clean second read" with full context.

Open in AIssential →

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