BERT disaster-tweet coursework model

Fine-tuned from bert-base-uncased using the Transformers v4.57.6 official classification example. Dataset: Kaggle NLP Getting Started tweets, from ttxy/nlp mirror; deduplicated by text, stratified split seed 42. One epoch, batch size 32, learning rate 2e-5, max sequence length 512, dynamic padding. On the held-out validation set of 750 rows the recorded accuracy was 0.8213333333333334. Label 1 means a real-disaster-related tweet; label 0 means not disaster related. This model is a course experiment, not a verified emergency decision system.

Downloads last month: 19

Safetensors

Model size

0.1B params

Tensor type

F32

Model tree for Ningzzz/chapter1-bert-disaster-20261030

Base model

google-bert/bert-base-uncased

Finetuned

(7068)

this model