Roadmap and Next Steps as of 4/5 - Githubissues

current12 / Stat-222-Project

2 stars 0 forks source link

Roadmap and Next Steps as of 4/5 #72

Closed ijyliu closed 2 months ago

ijyliu commented 3 months ago

Implement Tree-Based, Bagging and/or Boosting Classifier

Use Autogluon results as guidance
Better explore the hyperparameter space (start from optimized hyperparameters Autogluon found)
Permutation tests for feature importance

See Medium Presets items: https://github.com/current12/Stat-222-Project/tree/main/Output/Modelling/Autogluon

Network of Firms and Graph Neural Networks

Following original paper, first step is a network of firms
We've so far used transformer-based NER to get company mentions which are interesting on their own and also possibly usable for a network

https://github.com/current12/Stat-222-Project/blob/main/Output/All%20Data%20EDA/NLP%20EDA%20-%20NER%20on%20Company%20Names/Company%20Mentions%20Distribution%20No%20Title.png

Also have Doc2Vec representations of each call to use in line with original paper (cosine similarity threshold is used there to make graph edges)

Improve NLP Features (Especially Sentiment)

Find or create models fine-tuned on earnings calls
Possibly integrate annotation project data to create a feature