Model Enhancement with Data Augmentation for Slang Detection

Computational Semantics Project, ETH Zürich

Environment Build Up and Activate

For Linux system, use the command below to build up the conda environment and then activate the environment.

conda env create -f environment.yml
conda activate csnlp

We can use the pipeline.py file to process different parts of the project.

Data extraction

Use the file data_processing.py

Data Augmentation

method: GPT2 + top_k/top_p

metrics: bleu, perplexity, frequency

to get the final augmentation examples.

python pipeline.py --pipeline final_augmentation

Mask Language Modelling

run rs_mlm.py for random search for parameters for MLM.

run run_mlm.py for final mask language modellig adpater training and storing.

Classification

Baseline Model

run rs_cls_baseline.py for random search baseline.

run cls_baseline_base.py and cls_baseline_mini.py for final baseline.

Enhanced Model

run rs_cls_enhanced.py for random search enhanced model.

run cls_enhanced_base.py and cls_enhanced_mini.py for final enhanced model.