QuakeSense-LLM

Overview

QuakeSense-LLM is a four-class text classification model for identifying perceived shaking expressed in Chinese social media comments following earthquakes.

The model is designed to classify short Weibo comments into four perceived-shaking categories.

Model Architecture

The model is based on DeepSeek-7B-Base with task-specific LoRA fine-tuning and a four-class classification head.

The released model files include model weights, configuration files, and tokenizer files.

Classification Labels

• Class 0: No obvious shaking / irrelevant
• Class 1: Slight shaking
• Class 2: Moderate shaking
• Class 3: Strong shaking / affected

Intended Use

The model is intended for academic research on post-earthquake social media analysis and preliminary perceived-shaking classification.

Its predictions represent perceived shaking expressed in social media text, not official seismic intensity measurements.

Limitations

The model predictions should not be treated as substitutes for official seismic intensity assessments or instrumental earthquake observations.

Repository

Source code and associated dataset files are available at:

https://github.com/mumu9899/QuakeSense-LLM

Citation

Please cite the associated research paper when using this model in academic work.