Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,43 @@
|
|
| 1 |
---
|
| 2 |
license: apache-2.0
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
license: apache-2.0
|
| 3 |
+
datasets:
|
| 4 |
+
- financial_phrasebank
|
| 5 |
+
- pauri32/fiqa-2018
|
| 6 |
+
- zeroshot/twitter-financial-news-sentiment
|
| 7 |
+
language:
|
| 8 |
+
- en
|
| 9 |
+
metrics:
|
| 10 |
+
- accuracy
|
| 11 |
+
pipeline_tag: text-classification
|
| 12 |
+
tags:
|
| 13 |
+
- finance
|
| 14 |
---
|
| 15 |
+
|
| 16 |
+
|
| 17 |
+
We collects financial domain terms from Investopedia's Financia terms dictionary, NYSSCPA's accounting terminology guide
|
| 18 |
+
and Harvey's Hypertextual Finance Glossary to expand RoBERTa's vocab dict.
|
| 19 |
+
|
| 20 |
+
Based on added-financial-terms RoBERTa, we pretrained our model on multilple financial corpus:
|
| 21 |
+
|
| 22 |
+
- Financial Terms
|
| 23 |
+
- [Investopedia's Financia terms dictionary](https://www.investopedia.com/financial-term-dictionary-4769738)
|
| 24 |
+
- [NYSSCPA's accounting terminology guide](https://www.nysscpa.org/professional-resources/accounting-terminology-guide)
|
| 25 |
+
- [Harvey's Hypertextual Finance Glossary](https://people.duke.edu/~charvey/Classes/wpg/glossary.htm)
|
| 26 |
+
- Financial Datasets
|
| 27 |
+
- [FPB](https://huggingface.co/datasets/financial_phrasebank)
|
| 28 |
+
- [FiQA SA](https://huggingface.co/datasets/pauri32/fiqa-2018)
|
| 29 |
+
- [SemEval2017 Task5](https://aclanthology.org/S17-2089/)
|
| 30 |
+
- [Twitter Financial News Sentiment](https://huggingface.co/datasets/zeroshot/twitter-financial-news-sentiment)
|
| 31 |
+
- Earnings Call
|
| 32 |
+
2016-2023 NASDAQ 100 components stocks's Earnings Call Transcripts.
|
| 33 |
+
|
| 34 |
+
|
| 35 |
+
In continual pretraining step, we apply following experiments settings to achieve better finetuned results on Four Financial Datasets:
|
| 36 |
+
|
| 37 |
+
1. Masking Probability: 0.4 (instead of default 0.15)
|
| 38 |
+
2. Warmup Steps: 0 (deriving better results than models with warmup steps)
|
| 39 |
+
3. Epochs: 1 (is enough in case of overfitting)
|
| 40 |
+
4. weight_decay: 0.01
|
| 41 |
+
5. Train Batch Size: 64
|
| 42 |
+
6. FP16
|
| 43 |
+
|