# xxai.art · 推荐算法设计

### 开发环境搭建

确认有 <https://github.com/aier-dev/conf> 的访问权限

```
curl -s https://ghproxy.com/https://raw.githubusercontent.com/xxai-art/docker/main/dev/curl.sh | bash
```

### 表结构

启动之后访问 <http://localhost:8082>

![](https://i-01.eu.org/2023/09/SxvhYuQ.webp)

用户名 `gt` 密码 `DPUBU6Ytyx47dh`

表结构见 [init.sql](https://github.com/xxai-art/rsrv/blob/main/init/init.sql)

其中 aid 代表 action id，含义见 [rsrv/xc/src/action.rs](https://github.com/xxai-art/rsrv/blob/main/xc/src/action.rs)

[aier-art/rec](https://github.com/aier-art/rec/blob/main/src/main.rs)&#x20;

会读取数据库，更新每个图的质量分

注意代码中有根据图片的宽度做降权，越宽的图片展示面积越大，越容易被点击，要降权

推荐算法也需要注意这一点

[rsrv/xws/src/db/rec.rs](https://github.com/xxai-art/rsrv/blob/main/xws/src/db/rec.rs)&#x20;

会根据用户每次点击和收藏，创建新的推荐序列

[web/src/db/recPool.coffee](https://github.com/xxai-art/web/blob/main/src/db/recPool.coffee)&#x20;

会用下面的算法，抽取 41 个推荐序列+1 个全局推荐，来输出用户图片流

被点击的推荐流会被再次置顶

如果被推荐的图片被点击或收藏，会形成一个点击链条，用点击链条来再次做推荐

推荐是用websocket异步推送给前端的，如果后端有离线算法，无需点击也可以主动向浏览器推送新的推荐序列（离线算法最好也归因出每个推荐流的前一次点击，来形成推荐链条）

[web/src/lib/sampling.coffee](https://github.com/xxai-art/web/blob/main/src/lib/sampling.coffee)&#x20;

这是随机抽取的算法，魔改版的斐波那契数列


---

# Agent Instructions: Querying This Documentation

If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://xxai.gitbook.io/zh/xxai.art-tui-jian-suan-fa-she-ji.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.