# Content Moderation on Multi-category Chinese Content Moderation Dataset 1.0 (test)

Source: [Content Moderation on Multi-category Chinese Content Moderation Dataset 1.0 (test)](<https://www.sota2.com/research/sota/content-moderation-on-multi-category-chinese-content-moderation-dataset-1-0-test>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-04-14T09:24:29.309Z

Evaluation of Content Moderation performance on the test split of the Multi-category Chinese Content Moderation Dataset version 1.0, covering specific categories like Politics, Pornography, Violence, Bias, Gambling, and Harmlessness.

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

| Scope | Recorded value |
| --- | --- |
| Task | Content Moderation |
| Dataset (including any stated split) | Multi-category Chinese content moderation dataset 1.0 (test) |
| Metric | Politics Accuracy |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 15 of 15 |
| Coverage | All results currently returned for this benchmark |

Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol.

## Metrics

- [Politics Accuracy](<https://www.sota2.com/research/sota/content-moderation-on-multi-category-chinese-content-moderation-dataset-1-0-test.md?metric=Politics%20Accuracy>): Higher is better; unit: PERCENT.

- [Pornography Accuracy](<https://www.sota2.com/research/sota/content-moderation-on-multi-category-chinese-content-moderation-dataset-1-0-test.md?metric=Pornography%20Accuracy>): Higher is better; unit: PERCENT.

- [Violence Accuracy](<https://www.sota2.com/research/sota/content-moderation-on-multi-category-chinese-content-moderation-dataset-1-0-test.md?metric=Violence%20Accuracy>): Higher is better; unit: PERCENT.

- [Bias Accuracy](<https://www.sota2.com/research/sota/content-moderation-on-multi-category-chinese-content-moderation-dataset-1-0-test.md?metric=Bias%20Accuracy>): Higher is better; unit: PERCENT.

- [Gambling Accuracy](<https://www.sota2.com/research/sota/content-moderation-on-multi-category-chinese-content-moderation-dataset-1-0-test.md?metric=Gambling%20Accuracy>): Higher is better; unit: PERCENT.

- [Harmlessness Accuracy](<https://www.sota2.com/research/sota/content-moderation-on-multi-category-chinese-content-moderation-dataset-1-0-test.md?metric=Harmlessness%20Accuracy>): Higher is better; unit: PERCENT.

- [Average Accuracy](<https://www.sota2.com/research/sota/content-moderation-on-multi-category-chinese-content-moderation-dataset-1-0-test.md?metric=Average%20Accuracy>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Politics Accuracy | Result date | Report paper |
| --- | --- | --- | --- |
| CARO (Llama3-8B) | 89.7 | Not recorded | 69ddca117d0172140ec799b6 |
| CARO (Qwen2.5-7B) | 89.5 | Not recorded | 69ddca117d0172140ec799b6 |
| Naive SFT | 82.8 | Not recorded | 69ddca117d0172140ec799b6 |
| DeepSeek V3 | 79 | Not recorded | 69ddca117d0172140ec799b6 |
| CoT SFT+RL | 78.3 | Not recorded | 69ddca117d0172140ec799b6 |
| CoT SFT | 77.3 | Not recorded | 69ddca117d0172140ec799b6 |
| QwQ-32B | 75.4 | Not recorded | 69ddca117d0172140ec799b6 |
| Class-RAG | 74.9 | Not recorded | 69ddca117d0172140ec799b6 |
| DeepSeek R1 | 72.7 | Not recorded | 69ddca117d0172140ec799b6 |
| Agentic ICL | 66.8 | Not recorded | 69ddca117d0172140ec799b6 |
| Qwen2.5-32B-Instruct | 59.1 | Not recorded | 69ddca117d0172140ec799b6 |
| GPT-4 | 58.6 | Not recorded | 69ddca117d0172140ec799b6 |
| LLaMA3-8B | 58.5 | Not recorded | 69ddca117d0172140ec799b6 |
| Qwen2.5-7B-Instruct | 54.9 | Not recorded | 69ddca117d0172140ec799b6 |
| LLaMA-Guard-3-8B | 12 | Not recorded | 69ddca117d0172140ec799b6 |

### CARO (Llama3-8B)

Politics Accuracy: 89.7. Result ID: 69de07c647f34ce924c5be41.

[Reporting paper](<https://www.sota2.com/research/paper/69ddca117d0172140ec799b6.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69ddca117d0172140ec799b6.md>)

Recorded evaluation configuration: {"Category":"Our Model","Base Model":"Llama3-8B"}. This record does not assert independent reproduction.

### CARO (Qwen2.5-7B)

Politics Accuracy: 89.5. Result ID: 69de07c647f34ce924c5be40.

[Reporting paper](<https://www.sota2.com/research/paper/69ddca117d0172140ec799b6.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69ddca117d0172140ec799b6.md>)

Recorded evaluation configuration: {"Category":"Our Model","Base Model":"Qwen2.5-7B"}. This record does not assert independent reproduction.

### Naive SFT

Politics Accuracy: 82.8. Result ID: 69de07c647f34ce924c5be3b.

[Reporting paper](<https://www.sota2.com/research/paper/69ddca117d0172140ec799b6.md>)

Recorded evaluation configuration: {"Category":"Post-Training"}. This record does not assert independent reproduction.

### DeepSeek V3

Politics Accuracy: 79. Result ID: 69de07c647f34ce924c5be38.

[Reporting paper](<https://www.sota2.com/research/paper/69ddca117d0172140ec799b6.md>)

Recorded evaluation configuration: {"Category":"General LLMs"}. This record does not assert independent reproduction.

### CoT SFT+RL

Politics Accuracy: 78.3. Result ID: 69de07c647f34ce924c5be3d.

[Reporting paper](<https://www.sota2.com/research/paper/69ddca117d0172140ec799b6.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7c491e209131cea8d665.md>)

Recorded evaluation configuration: {"Category":"Post-Training"}. This record does not assert independent reproduction.

### CoT SFT

Politics Accuracy: 77.3. Result ID: 69de07c647f34ce924c5be3c.

[Reporting paper](<https://www.sota2.com/research/paper/69ddca117d0172140ec799b6.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa72781e209131cea568ff.md>)

Recorded evaluation configuration: {"Category":"Post-Training"}. This record does not assert independent reproduction.

### QwQ-32B

Politics Accuracy: 75.4. Result ID: 69de07c647f34ce924c5be37.

[Reporting paper](<https://www.sota2.com/research/paper/69ddca117d0172140ec799b6.md>)

Recorded evaluation configuration: {"Category":"General LLMs"}. This record does not assert independent reproduction.

### Class-RAG

Politics Accuracy: 74.9. Result ID: 69de07c647f34ce924c5be3e.

[Reporting paper](<https://www.sota2.com/research/paper/69ddca117d0172140ec799b6.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa764f1e209131cea6c526.md>)

Recorded evaluation configuration: {"Category":"Post-Training"}. This record does not assert independent reproduction.

### DeepSeek R1

Politics Accuracy: 72.7. Result ID: 69de07c647f34ce924c5be39.

[Reporting paper](<https://www.sota2.com/research/paper/69ddca117d0172140ec799b6.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa76a91e209131cea6e3eb.md>)

Recorded evaluation configuration: {"Category":"General LLMs"}. This record does not assert independent reproduction.

### Agentic ICL

Politics Accuracy: 66.8. Result ID: 69de07c647f34ce924c5be3f.

[Reporting paper](<https://www.sota2.com/research/paper/69ddca117d0172140ec799b6.md>)

Recorded evaluation configuration: {"Category":"Post-Training"}. This record does not assert independent reproduction.

### Qwen2.5-32B-Instruct

Politics Accuracy: 59.1. Result ID: 69de07c647f34ce924c5be36.

[Reporting paper](<https://www.sota2.com/research/paper/69ddca117d0172140ec799b6.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa767e1e209131cea6d6a7.md>)

Recorded evaluation configuration: {"Category":"General LLMs"}. This record does not assert independent reproduction.

### GPT-4

Politics Accuracy: 58.6. Result ID: 69de07c647f34ce924c5be35.

[Reporting paper](<https://www.sota2.com/research/paper/69ddca117d0172140ec799b6.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa727a1e209131cea56a1e.md>)

[Implementation](<https://github.com/openai/evals>)

Recorded evaluation configuration: {"Category":"General LLMs"}. This record does not assert independent reproduction.

### LLaMA3-8B

Politics Accuracy: 58.5. Result ID: 69de07c647f34ce924c5be34.

[Reporting paper](<https://www.sota2.com/research/paper/69ddca117d0172140ec799b6.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa75d11e209131cea6a028.md>)

[Implementation](<https://github.com/meta-llama/llama-agentic-system>)

Recorded evaluation configuration: {"Category":"General LLMs"}. This record does not assert independent reproduction.

### Qwen2.5-7B-Instruct

Politics Accuracy: 54.9. Result ID: 69de07c647f34ce924c5be33.

[Reporting paper](<https://www.sota2.com/research/paper/69ddca117d0172140ec799b6.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa767e1e209131cea6d6a7.md>)

Recorded evaluation configuration: {"Category":"General LLMs"}. This record does not assert independent reproduction.

### LLaMA-Guard-3-8B

Politics Accuracy: 12. Result ID: 69de07c647f34ce924c5be3a.

[Reporting paper](<https://www.sota2.com/research/paper/69ddca117d0172140ec799b6.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa717b1e209131cea5186d.md>)

Recorded evaluation configuration: {"Category":"Specific LLMs"}. This record does not assert independent reproduction.

Page 1 of 1. This is an index page, not the full collection.
