| # ModernBERT-BizClassifier - Business Relevance Classifier | |
| ## 📌 Overview | |
| ModernBERT-BizClassifier is a **fine-tuned ModernBERT model** that classifies text from **any source** (social media, news, reports, etc.) and determines whether it is **useful for businesses**. | |
| The model distinguishes between: | |
| - **Business_Action_Direct**: Clearly relevant business information (e.g., financial reports, earnings, funding rounds). | |
| - **Business_Intelligence_Indirect**: Text that implies business relevance (e.g., regulatory news, economic trends, emerging markets). | |
| - **Not_Business_Relevant**: Text that contains no useful business insights. | |
| ## 🛠 Model Details | |
| | Hyperparameter | Value | | |
| |-----------------------|----------------| | |
| | **Base Model** | `ModernBERT-base` | | |
| | **Max Context Length** | `8192 tokens` | | |
| | **Batch Size** | `32` | | |
| | **Learning Rate** | `1e-5` | | |
| | **Weight Decay** | `0.005` | | |
| | **Epochs** | `2.0` | | |
| | **Warmup Ratio** | `0.1` | | |
| | **Precision Used** | `BF16` if CUDA available | | |
| | **Problem Type** | `Multi-label classification` | | |
| ## 📊 Evaluation Results | |
| | Metric | Score | | |
| |-----------------------------|--------| | |
| | **Final Validation Loss** | `0.2940` | | |
| | **Micro F1 Score** | `0.8049` | | |
| | **Macro F1 Score** | `0.8042` | | |
| | **Evaluation Runtime** | `3.48 seconds` | | |
| | **Samples Processed Per Second** | `239.09` | | |
| | **Steps Processed Per Second** | `7.48` | | |
| ## 🏗 Model Card | |
| - **Tags**: `business-classification`, `BERT`, `financial-analysis`, `text-classification` | |
| ```python | |
| { | |
| 1: "Business_Action_Direct", | |
| 2: "Business_Intelligence_Indirect", | |
| 3: "Not_Business_Relevant" | |
| } | |
| ``` |