Forum Understanding Using NLP Techniques

Hardware

• CPU: AMD 7950X
• GPU: RTX 2080ti

Enviroment

It's recommand to create a virtual enviroment, e.g. conda.

pip install -r requirements.txt

Download

To download the fine-tuned models for reproducing.

bash download.sh

Reproduce

Train

To preporcess the data and training.

• Running shell script.

bash train.sh

or

• And here is how you would use it on your own files, after adjusting the values for the arguments --train_path, --dev_path to match your setup

• You could also setting different hyperameters by adjusting the values for the arguments --learning_rate , --num_epochs

python3 src/span-aste/train.py \
    --batch_size 1 \
    --learning_rate 5e-5 \
    --weight_decay 1e-2 \
    --warmup_proportion 0.1 \
    --train_path processed_data \
    --dev_path processed_data \
    --ckpt_dir ckpt/span-aste \
    --output_dir output \
    --max_seq_len 256 \
    --num_epochs 70 \
    --seed 2022 \
    --logging_steps 480 \
    --valid_steps 480 \
    # --init_from_ckpt \

Test

1. Sentiment analysis

• Running shell script.

bash test.sh

or

• You could also setting different parameter in --test_path to save the output in other location.

python3 src/span-aste/test.py \
    --test_path processed_data \
    --ckpt ckpt/span-aste \
    --output_dir output \

2. Topic model

• Executing .ipynb files in /src/Bertopic
• Name of the file represent the test data crawl from which topic on Reddit.