Improve ChineseHarm-Bench dataset card with overview, usage, results, and detailed links

This PR enriches the `ChineseHarm-Bench` dataset card to provide more comprehensive information and practical guidance for users.

Key improvements include:
- **Expanded Metadata:** Added relevant tags (`harmful-content-detection`, `safety`, `moderation`, `benchmark`, `chinese`) for better discoverability.
- **Refined Links:** Clarified and added separate links for the project's GitHub repository (Code), the dedicated project website (Project Page), the associated paper (arXiv), and the Hugging Face collection.
- **Detailed Overview:** Integrated the "Overview" section from the project's GitHub README, offering a concise summary of the benchmark's purpose and key features.
- **Practical Usage Examples:** Included code snippets and instructions for performing single and batch inference, as well as evaluating predictions, making it easier for users to get started with the dataset.
- **Performance Insights:** Added the "Main Results" table, showcasing the performance of various models on the benchmark, which is crucial for researchers.
- **Community Engagement:** Added the "Contributors" section from the GitHub README.

These updates aim to make the dataset card a more valuable and self-contained resource for researchers and practitioners.

README.md

@@ -6,16 +6,23 @@ size_categories:
 - 10K<n<100K
 task_categories:
 - text-classification
+tags:
+- harmful-content-detection
+- safety
+- moderation
+- benchmark
+- chinese
 ---
 
 <h1 align="center"> ChineseHarm-bench</h1>
-<h3 align="center"> A Chinese Harmful Content  Detection Benchmark </h3>
+<h3 align="center"> A Chinese Harmful Content Detection Benchmark </h3>
 
 > ⚠️ **WARNING**: This project and associated data contain content that may be toxic, offensive, or disturbing. Use responsibly and with discretion.
 
 <p align="center">
-  <a href="https://github.com/zjunlp/ChineseHarm-bench">Project</a> •
+  <a href="https://github.com/zjunlp/ChineseHarm-bench">Code</a> •
   <a href="https://arxiv.org/abs/2506.10960">Paper</a> •
+  <a href="https://zjunlp.github.io/project/ChineseHarm-Bench/">Project Page</a> •
   <a href="https://huggingface.co/collections/zjunlp/chineseharm-bench-683b452c5dcd1d6831c3316c">Hugging Face</a> 
 </p>
 
@@ -26,6 +33,12 @@ task_categories:
   <img src="chineseharm_case.png" width="80%"/></p>
 </div>
 
+## 🌟Overview
+
+We introduce ChineseHarm-Bench, a professionally annotated benchmark for Chinese harmful content detection, covering six key categories. It includes a knowledge rule base to enhance detection and a knowledge-augmented baseline that enables smaller LLMs to match state-of-the-art performance. 
+
+The benchmark construction process is illustrated in the figure above and for more detailed procedures, please refer to our paper.
+
 ## 🌟Benchmark
 
 This folder contains the ChineseHarm-Bench.
@@ -38,6 +51,99 @@ Each file is a list of examples with:
 * `"文本"`: the input Chinese text
 * `"标签"`: the ground-truth label
 
+## Sample Usage (Inference)
+
+Our inference scripts support both **Huawei Ascend NPUs and NVIDIA GPUs**, enabling flexible deployment across different hardware platforms. For the full scripts and details, please refer to the [GitHub repository](https://github.com/zjunlp/ChineseHarm-bench).
+
+We release the following variants of our harmful content detection model:
+
+- [**ChineseGuard-1.5B**](https://huggingface.co/zjunlp/ChineseGuard-1.5B)
+- [**ChineseGuard-3B**](https://huggingface.co/zjunlp/ChineseGuard-3B)
+- [**ChineseGuard-7B**](https://huggingface.co/zjunlp/ChineseGuard-7B)
+
+🔹 Single Inference (Example)
+
+Run single-input inference using the ChineseGuard-1.5B model:
+
+```bash
+SCRIPT_PATH="../infer/single_infer.py"
+model_name="zjunlp/ChineseHarm-1.5B"
+text="代发短信，有想做的联系我，无押金"
+
+python $SCRIPT_PATH \
+    --model_name $model_name \
+    --text $text
+```
+
+🔸 Batch Inference (Multi-NPU or Multi-GPU)
+
+To run inference on the entire ChineseHarm-Bench using ChineseGuard-1.5B and 8 NPUs:
+
+```bash
+SCRIPT_PATH="../infer/batch_infer.py"
+model_name="zjunlp/ChineseHarm-1.5B"
+file_name="../benchmark/bench.json"
+output_file="../benchmark/bench_ChineseHarm-1.5B.json"
+
+python $SCRIPT_PATH \
+    --model_name $model_name \
+    --file_name $file_name \
+    --output_file $output_file \
+    --num_npus 8
+```
+
+> For more configuration options (e.g., batch size, device selection, custom prompt templates), please refer to `single_infer.py` and `batch_infer.py` in the [code repository](https://github.com/zjunlp/ChineseHarm-bench/tree/main/infer).
+>
+> **Note:** The inference scripts support both NPU and GPU devices.
+
+**Evaluation: Calculating F1 Score**
+
+After inference, evaluate the predictions by computing the F1 score with the following command:
+
+```bash
+python ../calculate_metrics.py \
+    --file_path "../benchmark/bench_ChineseHarm-1.5B.json" \
+    --true_label_field "标签" \
+    --predicted_label_field "predict_label"
+```
+
+## Main Results
+
+> 🔴：Without Knowledge Augmentation 🟢：With Knowledge Augmentation 🟦：Our Strong Baseline
+
+| Model | Strategy | Knowledge | Gambling | Pornography | Abuse | Fraud | Illicit Ads | Non-Violation | Macro-F1 |
+| :-------------------------: | :--------: | :-------: | :------: | :---------: | :---: | :---: | :---------: | :-----------: | :------: |
+| **Deepseek-R1** | Prompting | 🔴 | 0.82 | 0.77 | 0.84 | 0.53 | 0.65 | 0.78 | 0.73 |
+| | Prompting | 🟢 | 0.89 | 0.83 | 0.87 | 0.65 | 0.77 | 0.80 | 0.80 |
+| **O3-mini** | Prompting | 🔴 | 0.56 | 0.55 | 0.74 | 0.57 | 0.22 | 0.45 | 0.51 |
+| | Prompting | 🟢 | 0.70 | 0.55 | 0.73 | 0.60 | 0.40 | 0.46 | 0.57 |
+| **GPT-4o** | Prompting | 🔴 | 0.78 | 0.75 | 0.83 | 0.59 | 0.53 | 0.79 | 0.71 |
+| | Prompting | 🟢 | 0.89 | 0.75 | 0.82 | 0.60 | 0.75 | 0.86 | 0.78 |
+| **GPT-4o-mini** | Prompting | 🔴 | 0.57 | 0.70 | 0.71 | 0.43 | 0.40 | 0.59 | 0.57 |
+| | Prompting | 🟢 | 0.82 | 0.76 | 0.74 | 0.51 | 0.62 | 0.72 | 0.69 |
+| **Gemini 2.0 Flash** | Prompting | 🔴 | 0.72 | 0.76 | 0.84 | 0.63 | 0.52 | 0.75 | 0.71 |
+| | Prompting | 🟢 | 0.91 | 0.77 | 0.82 | 0.51 | 0.69 | 0.75 | 0.74 |
+| **Claude 3.5 Sonnet** | Prompting | 🔴 | 0.76 | 0.76 | 0.79 | 0.11 | 0.57 | 0.80 | 0.63 |
+| | Prompting | 🟢 | 0.87 | 0.81 | 0.78 | 0.36 | 0.72 | 0.78 | 0.72 |
+| **BERT-Base-Chinese** | Finetuning | 🔴 | 0.49 | 0.60 | 0.73 | 0.49 | 0.50 | 0.68 | 0.58 |
+| 🟦 | Finetuning | 🟢 | 0.74 | 0.65 | 0.76 | 0.68 | 0.68 | 0.70 | 0.70 |
+| **Qwen--2.5-0.5B-Instruct** | Prompting | 🔴 | 0.00 | 0.21 | 0.00 | 0.00 | 0.00 | 0.30 | 0.09 |
+| | Prompting | 🟢 | 0.00 | 0.11 | 0.00 | 0.00 | 0.00 | 0.30 | 0.07 |
+| | Finetuning | 🔴 | 0.35 | 0.59 | 0.72 | 0.39 | 0.44 | 0.74 | 0.54 |
+| 🟦 | Finetuning | 🟢 | 0.75 | 0.64 | 0.75 | 0.62 | 0.70 | 0.74 | 0.70 |
+| **Qwen--2.5-1.5B-Instruct** | Prompting | 🔴 | 0.22 | 0.08 | 0.62 | 0.47 | 0.00 | 0.48 | 0.31 |
+| | Prompting | 🟢 | 0.55 | 0.13 | 0.53 | 0.52 | 0.00 | 0.45 | 0.36 |
+| | Finetuning | 🔴 | 0.36 | 0.61 | 0.74 | 0.43 | 0.48 | 0.81 | 0.57 |
+| 🟦 | Finetuning | 🟢 | 0.77 | 0.71 | 0.77 | 0.70 | 0.74 | 0.79 | 0.75 |
+| **Qwen-2.5-3B-Instruct** | Prompting | 🔴 | 0.38 | 0.53 | 0.58 | 0.38 | 0.36 | 0.50 | 0.46 |
+| | Prompting | 🟢 | 0.62 | 0.55 | 0.46 | 0.58 | 0.10 | 0.49 | 0.47 |
+| | Finetuning | 🔴 | 0.47 | 0.63 | 0.77 | 0.37 | 0.49 | 0.82 | 0.59 |
+| 🟦 | Finetuning | 🟢 | 0.81 | 0.72 | 0.79 | 0.72 | 0.74 | 0.85 | 0.77 |
+| **Qwen--2.5-7B-Instruct** | Prompting | 🔴 | 0.35 | 0.58 | 0.42 | 0.09 | 0.45 | 0.56 | 0.41 |
+| | Prompting | 🟢 | 0.51 | 0.63 | 0.48 | 0.37 | 0.32 | 0.42 | 0.46 |
+| | Finetuning | 🔴 | 0.35 | 0.64 | 0.72 | 0.38 | 0.49 | 0.82 | 0.57 |
+| 🟦 | Finetuning | 🟢 | 0.82 | 0.70 | 0.75 | 0.75 | 0.75 | 0.82 | 0.77 |
+
 ## 🚩 Ethics Statement
 
 We obtain all data with proper authorization from the respective data-owning organizations and signed the necessary agreements.
@@ -65,4 +171,8 @@ Please cite our repository if you use ChineseHarm-bench in your work. Thanks!
       primaryClass={cs.CL},
       url={https://arxiv.org/abs/2506.10960}, 
 }
-```
+```
+
+## 🎉Contributors
+
+We will offer long-term maintenance to fix bugs and solve issues. So if you have any problems, please put issues to us.