Apply pre-trained models to help quickly grasp and analyze the information from investment news, including three tasks, 1. summarizationm 2. sentiment analysis 3. domain classification
Here is the detail: https://www.dropbox.com/scl/fi/8ubber3p4m3etutcaj7vr/Financial-News.pdf?rlkey=72f4olz8mk55ypr44l3o98yv6&dl=1
torch==1.12.1
transformers==4.22.2
datasets
accelerate
sentencepiece
rouge
spacy
nltk
ckiptagger
tqdm
pandas
numpy
jsonlines
evaluate
rouge_score
opencc
bash scripts/download.sh
you could see the code details in train_summary.py. Evaluation is included.
bash scripts/train.sh
if you need only evaluation without training, use this script.
bash scripts/eval.sh
## e.g. bash strategies/beam/beam6/eval.sh
bash strategies/{strategy}/eval.sh
only to inference(predict)
bash scripts/predict.sh ./path/to/test_file ./path/to/output_file
Need more correct data to enhance the result, this is still need more revision(dataset correctness), you could check that in our paper.
bash scripts/download.sh
bash scripts/train.sh
Need more correct data to enhance the result, this is still need more revision(for multi-label issues), you could check that in our paper.
bash scripts/download.sh
bash scripts/train.sh
only to inference(predict), this task doesn't provide model weight, need to train first to do inference here.
bash scripts/predict.sh ./path/to/test_file ./path/to/output_file