Inside the news_sentiment_analysis
folder, including:
- News collecting from East Money (www.eastmoney.com)
- News contents summarizing with pretrained Pegasus
- Sentiment model finetuning with ChnSentiCorp dataset
- Sentiment analyzing for merged news titles and summarized contents
Inside the modeling
folder:
There are six subfolders, each corresponding to one of the six experimental groups, as follows:
LSTM | Transformer | |
---|---|---|
Title + Content | MAS_lstm_enhancement/ |
MAS_transformer_enhancement/ |
Title | MAS_lstm_enhancement_title/ |
MAS_transformer_enhancement_title/ |
Vanilla | MAS_lstm/ |
MAS_transformer/ |
Inside each experimental group subfolder:
backtrader_sequence_model.py
: Code for deep learning model to predict stock price movements.run.py
: Script to runbacktrader_sequence_model.py
.output/
: Results of the deep learning model predictions.backtrader_mystrategy.ipynb
: Code for backtesting strategies.results/
: Results of strategy backtesting.plot.py
: Code to plot strategy returns and benchmark comparisons, with image paths infigs/
.procedure/
: Records training process metrics for the top 50 stocks.
Inside the MAS-2023-code/
path:
train_procedure.py
: Code for plotting metrics for the top 50 stocks.evaluate.py
: Code for calculating metrics for the top 50 stocks.