How maal0lan/cyberbully works
The repository combines cyberbullying dataset creation and model development with a single-text detection workflow. The documented user path is text input → CLI prediction → label, probability, and threshold; the training script prepares data, trains, evaluates, and exports a model. Dataset-generation pipelines are represented as separate product capabilities. Some training and CLI internals were not excerpted, so those links rely on README statements rather than claims about unshown implementation.
Main entry points: generate_pipeline.py Explicit generation · pipeline_clean.py Clean generation · predict_cyberbullying_cli.py Prediction CLI · best_model.keras Model checkpoint
Components 20
Dataset generation
- Explicit generation
dataset_generation/helper_files/gen/explicit generator/generate_pipeline.py - Explicit prompts
dataset_generation/helper_files/gen/explicit generator/templates.py - Clean generation
dataset_generation/helper_files/gen/non-explicit-generator/pipeline_clean.py - Clean prompts
dataset_generation/helper_files/gen/non-explicit-generator/templates_clean.py - Sentence generation
dataset_generation/data_words/sentence_maker.py - Language file renaming
dataset_generation/data_words/name_changer.py
Dataset curation
- Explicit augmentation
dataset_generation/helper_files/expander/explicit_dataset_generation/exp_augment_pipeline.py - Clean augmentation
dataset_generation/helper_files/expander/non-explicit_dataset_generation/non-exp_augment_pipeline.py - Dataset merger
dataset_generation/helper_files/expander/merge_datasets.py - Merged dataset
dataset_generation/helper_files/dataset/generated_dataset/cyberbullying_dataset.csv - Dataset analytics
dataset_generation/helper_files/dataset/analytics/dataset_analytics.py
Model workflow
- Training pipeline
cyberbully_final.py - Prepared splits
cyberbully_final.py - Model checkpoint
cyberbully_output/best_model.keras
Prediction and evaluation
- Prediction CLI
predict_cyberbullying_cli.py - Label and score
- Model evaluation
eval_results/evaluate_models.py - Evaluation results
eval_results
Other components
- Text submitter
- Ollama model service
Connections 18
- Text submitter → Prediction CLI: submits text
README.md - Prediction CLI → Model checkpoint: loads model
README.md - Prediction CLI → Label and score: prints prediction
README.md - Training pipeline → Prepared splits: creates splits
cyberbully_final.py - Training pipeline → Model checkpoint: exports model
README.md - Model evaluation → Model checkpoint: evaluates model
README.md - Model evaluation → Evaluation results: writes metrics
eval_results/evaluate_models.py - Explicit generation → Explicit prompts: uses prompts
dataset_generation/helper_files/gen/explicit generator/generate_pipeline.py - Explicit generation → Ollama model service: requests generation
dataset_generation/helper_files/gen/explicit generator/generate_pipeline.py - Explicit generation → Ollama model service: requests judging
dataset_generation/helper_files/gen/explicit generator/generate_pipeline.py - Explicit generation → Explicit generation: filters agreements
dataset_generation/helper_files/gen/explicit generator/generate_pipeline.py - Clean generation → Clean prompts: uses prompts
dataset_generation/helper_files/gen/non-explicit-generator/pipeline_clean.py - Clean generation → Ollama model service: requests generation
dataset_generation/helper_files/gen/non-explicit-generator/pipeline_clean.py - Clean generation → Ollama model service: requests judging
dataset_generation/helper_files/gen/non-explicit-generator/pipeline_clean.py - Explicit augmentation → Explicit augmentation: mutates source words
dataset_generation/helper_files/expander/explicit_dataset_generation/exp_augment_pipeline.py - Clean augmentation → Clean augmentation: mutates source words
dataset_generation/helper_files/expander/non-explicit_dataset_generation/non-exp_augment_pipeline.py - Dataset merger → Merged dataset: writes merged data
dataset_generation/helper_files/expander/merge_datasets.py - Dataset analytics → Merged dataset: analyzes dataset