Update README.md

README.md

@@ -1,3 +1,94 @@
----
-license: apache-2.0
----
+---
+license: apache-2.0
+language:
+  - en
+tags:
+  - climate
+  - ESG
+  - sustainable-finance
+  - sequence-classification
+base_model: climatebert/distilroberta-base-climate-detector
+metrics:
+  - f1
+  - accuracy
+---
+
+# Green Shareholder Proposal Classifier
+
+## Model Summary
+
+This model is a fine-tuned version of [`climatebert/distilroberta-base-climate-detector`](https://huggingface.co/climatebert/distilroberta-base-climate-detector), specifically designed to classify **shareholder proposals** into binary categories: green (climate/environmental) or non-green. 
+
+It was trained on a highly curated dataset of Institutional Shareholder Services (ISS) proposals, achieving an **F1 score of 0.981** on the validation set.
+
+## Model Details
+
+- **Base Model:** `climatebert/distilroberta-base-climate-detector`
+- **Task:** Binary Sequence Classification
+  - `Label 1`: Green / Climate-related proposal
+  - `Label 0`: Non-green proposal
+- **Language:** English
+- **License:** Apache 2.0 (Model weights). *Note: The dataset used for fine-tuning contains derived data subject to ISS licensing terms.*
+
+## Uses
+
+### Direct Use
+The model takes a structured text input describing a shareholder proposal and predicts whether it is conceptually focused on climate change or environmental sustainability. 
+
+**Recommended Input Format:**
+To achieve optimal performance, input text should mirror the structure of the training data:
+> "A {sponsor_type}-type sponsor has filed a shareholder proposal to a(an) {sic2_des}-sector company. This proposal requests: {resolution}. [It falls under a broader agenda class that may include items not directly relevant to this specific proposal: {AgendaCodeInformation}]"
+
+### Out-of-Scope Use
+- Applying the model to non-English texts.
+- Using the model for automated legal or compliance decision-making without human oversight.
+- Generalizing to broad ESG topics outside of strict environmental/climate scopes (e.g., social or governance issues like gender equality or animal welfare are explicitly trained as negative classes).
+
+## Training Data
+
+The model was fine-tuned on a custom stratified dataset of 1,500 manually curated ISS shareholder proposals. The dataset underwent rigorous rule-based correction to exclude tangentially environmental or purely social/governance proposals. 
+
+For full details on data sampling, text construction, and labeling rules, please refer to the **[Dataset Card](在这里填入你的数据集链接)**.
+
+- **Train split:** 1,200 examples
+- **Validation split:** 300 examples
+
+## Training Procedure
+
+### Hyperparameters
+
+The model was trained using the Hugging Face `Trainer` API with the following hyperparameters:
+
+- **Learning rate:** 2e-05
+- **Train batch size:** 16
+- **Eval batch size:** 16
+- **Seed:** 42
+- **Weight decay:** 0.05
+- **Optimizer:** AdamW
+- **Number of epochs:** 10
+
+### Training Results
+
+The model weights from **Epoch 8 (`checkpoint-600`)** were selected as the best performing based on the validation F1 score.
+
+| Epoch | Training Loss | Validation Loss | Accuracy | F1 (Binary) |
+|:---:|:---:|:---:|:---:|:---:|
+| 1 | 0.3060 | 0.0968 | 0.9667 | 0.9675 |
+| 2 | 0.0954 | 0.0898 | 0.9733 | 0.9740 |
+| 3 | 0.0956 | 0.1808 | 0.9600 | 0.9623 |
+| 4 | 0.0029 | 0.0783 | 0.9800 | 0.9805 |
+| 5 | 0.0395 | 0.1026 | 0.9800 | 0.9803 |
+| 6 | 0.0350 | 0.1308 | 0.9733 | 0.9744 |
+| 7 | 0.0094 | 0.1108 | 0.9767 | 0.9772 |
+| **8** | **0.0003** | **0.1182** | **0.9800** | **0.9806** |
+| 9 | 0.0004 | 0.1154 | 0.9767 | 0.9773 |
+| 10 | 0.0002 | 0.1229 | 0.9767 | 0.9773 |
+
+## Limitations and Bias
+
+While the model achieves high accuracy on the validation set, its performance is tightly coupled with the specific linguistic patterns and taxonomy of the ISS database (e.g., SIC-2 sector descriptions, ISS agenda codes). It may exhibit lower confidence or accuracy when processing unstructured news articles, raw corporate filings, or proposals from different jurisdictional contexts outside the US/global norm represented in the training set.
+
+## Citation
+
+If you use this model in your research, please cite the associated working paper:
+*(Citation details forthcoming)*