Update README.md

README.md

@@ -7,4 +7,31 @@ base_model:
 pipeline_tag: text-classification
 ---
 
-Efficient BERT-based model for binary classification of task statement texts.
+# task-classifier-mini-improved2 (Legacy)
+
+> [!NOTE]  
+> **A newer, improved model is available!**  
+> We highly recommend using [loyoladatamining/task-classifier-mini-v3](https://huggingface.co/loyoladatamining/task-classifier-mini-v3) instead of this model.
+
+## Model Description
+This is a legacy BERT-based model fine-tuned for the binary classification of task statements (identifying whether a text represents a task to be done on the job). It was built on top of `prajjwal1/bert-tiny`.
+
+## Why Upgrade?
+When evaluated on the [loyoladatamining/usajobs_validation](https://huggingface.co/datasets/loyoladatamining/usajobs_validation) dataset, the newer version of the  model demonstrates higher performance:
+
+| Model | Accuracy | F-1 |
+| :--- | :---: | :---: |
+| task-classifier-mini-improved2 (This Model) | 0.8358 | 0.8253 |
+| **[task-classifier-mini-v3 (Recommended)](https://huggingface.co/loyoladatamining/task-classifier-mini-v3)** | **0.9583** | **0.9585** |
+
+## Citation
+
+If you do make use of this version, please consider citing:
+
+```
+@article{meisenbacher2025extracting,
+  title={Extracting O* NET Features from the NLx Corpus to Build Public Use Aggregate Labor Market Data},
+  author={Meisenbacher, Stephen and Nestorov, Svetlozar and Norlander, Peter},
+  year={2025}
+}
+```