distilbert-base-fire-class

Model description

The distilbert-base-fire-class model is a distilled version of BERT, specifically fine-tuned for classification tasks related to the dataset on tweets about fire incidents. This model can be used for analyzing and classifying tweets that mention fire-related events or discussions.

Intended uses & limitations

Intended Uses

The distilbert-base-fire-class model is specifically designed for distinguishing whether a text is discussing fire-related incidents or not. It can be applied in various contexts such as:

• Monitoring social media: Detecting and analyzing tweets or posts about fire incidents to help in early detection and response.
• News filtering: Identifying news articles or reports that focus on fire events for more efficient information processing.
• Emergency response systems: Assisting in categorizing incoming reports or alerts based on their relevance to fire-related topics.
• Research and analysis: Supporting studies on public reactions and discussions about fire incidents by classifying relevant texts.

Limitations

While the model is effective in its intended use, there are several limitations to consider:

• Dataset Specificity: The model is trained on a specific dataset of tweets about fire, which may limit its accuracy when applied to other types of texts or platforms.
• Language and Context: The model may not perform as well on texts that include slang, regional dialects, or highly context-specific language that was not well-represented in the training data.
• Generalization: This model is focused on fire-related text classification and may not generalize well to other topics without further fine-tuning or additional training data.

Training and evaluation data

The model was trained using the tweets_about_fire dataset. This dataset consists of tweets specifically mentioning fire incidents, providing a focused context for training the classification model.

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss
0.4456	1.0	368	0.4997
0.2904	2.0	736	0.6664
0.2408	3.0	1104	0.5989

Framework versions

• Transformers 4.38.2
• Pytorch 2.2.1+cu121
• Datasets 2.18.0
• Tokenizers 0.15.2

---