# A Chinese humor dataset with human vote labels

**URL:** <https://discuss.huggingface.co/t/a-chinese-humor-dataset-with-human-vote-labels/178847>\
**Category:** Research\
**Created:** [August 18, 2026, 7:24pm UTC](https://discuss.huggingface.co/t/a-chinese-humor-dataset-with-human-vote-labels/178847 "2026-08-18T19:24:34Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![Young25](https://avatars.discourse-cdn.com/v4/letter/y/b38774/32.png) [@Young25](https://discuss.huggingface.co/u/Young25)\
**Post date:** [August 18, 2026, 7:24pm UTC](https://discuss.huggingface.co/t/a-chinese-humor-dataset-with-human-vote-labels/178847/1 "2026-08-18T19:24:34Z")

</div>

Hi everyone,

I put together a small Chinese humor dataset called **Humor\_Vote** and wanted to share it in case it’s useful for anyone working on humor evaluation or preference data:

> **[Young25/Humor\_Vote · Datasets at Hugging Face](https://huggingface.co/datasets/Young25/Humor_Vote)**
>
> We’re on a journey to advance and democratize artificial intelligence through open source and open science.

## What’s in it

Each row is one humorous Chinese question with 4 candidate answers, plus human vote labels:

- `question_cn` — the question (e.g., “What’s the most useless thing you’ve ever bought?”)
- `answer_cn_0` … `answer_cn_3` — four candidate answers
- `chosen_cn` — the answer humans picked
- `answer_chosen_cn_0` … `answer_chosen_cn_3` — vote counts per answer

All questions and answers are native Chinese (not translated), so the wording and punchlines are natural.

## What I think it’s good for

1. **LLM post-training / evaluation** — the question + answers + human preference structure works directly for SFT, DPO-style preference pairs (chosen vs. the rest), reward modeling, or simply as a “can the model pick the funniest answer?” probe.
2. **Entertainment apps and games** — party icebreakers, quiz-style mini games, “vote for the funniest answer” mechanics, or humor content generation. The Q&A pairs can be used as a content library as-is.

## Quick start

```python
from datasets import load_dataset
ds = load_dataset("Young25/Humor_Vote", split="train")

```

This is a first release, so feedback on format or label quality is very welcome — and if you run any interesting experiments or benchmarks with it, I’d love to see the results. If you need more data, feel free to contact me .