【数据科学赛】WSDM Cup 2023 #百度出题 #搜索排序 #双赛道 #无偏学习 #预训练 #已开始 #$13,000

最新推荐文章于 2024-06-26 17:11:57 发布

转载最新推荐文章于 2024-06-26 17:11:57 发布 · 295 阅读

0 ·

CC 4.0 BY-SA版权

原文链接：https://aistudio.baidu.com/aistudio/competition/detail/534/0/introduction

文章标签：

#数据挖掘 #人工智能 #机器学习

CompHub 实时聚合多平台的数据类(Kaggle、天池…)和OJ类(Leetcode、牛客…）比赛。本账号会推送最新的比赛消息，欢迎关注！

更多比赛信息见 CompHub主页[1] 或点击文末_阅读原文_

以下内容摘录自比赛主页

Part1赛题介绍

1题目

Task 1: Unbiased Learning for Web Search[2]

Task 2: Pre-training for Web Search[3]

2标签

搜索排序、预训练、无偏学习、NDCG

3主办方

4交流方式New!

由于这两场比赛缺少官方交流渠道，CompHub 首次尝试提供一个讨论的空间，方便对比赛感兴趣的各位进行交流。如果大家认为这种方式对大家有帮助或有改进建议，欢迎留言让我知道~

Task 1: Unbiased Learning for Web Search 交流链接 ↓

https://support.qq.com/products/449835/topic-detail/2354/

Task 2: Pre-training for Web Search 交流链接 ↓

https://support.qq.com/products/449835/topic-detail/2355/

另外，在CompHub主页中也能进入这两场比赛的交流空间：

Part2时间安排

两个任务安排相同：

Event	Complete Date
Oct 18 2022	Website ready and training set available for download.
Nov 11 2022	Intermediate test set release and intermediate submission starts.
Dec 11 2022	Intermediate submission ends.
Dec 16 2022	Intermediate leaderboard result announcement.
Dec 17 2022	Final test set release and final submission starts.
Jan 11 2023	Final submission ends.
Jan 15 2023	Final leaderboard result announcement.
Jan 20 2023	Invitations to top 6 teams on each sub-task for short papers.
Feb 15 2023	Short paper deadline.
Feb 27- March 3 2023	WSDM Cup conference presentation in person for top 3 teams on each

Part3奖励机制

两个任务安排相同：

Champion: One team ($2000)
Runner-up: One team ($1000)
3rd-place: One team ($500)

For those top 3 teams, we will sponsor for one conference registration ($1000) each team.

Part4赛题描述

Task 1: Unbiased Learning for Web Search

You are required to train a ranking model with the Large Scale Web Search Session Data.

The submission is not limited to the PaddlePaddle and Pre-trained Models.

Task 2: Pre-training for Web Search

You are required to pre-train a PLM with the Large Scale Web Search Session Data and finetune the PLM with the Expert Annotation Dataset

The award will only be honored to the top-3 results that use PaddlePaddle in the leaderboard.

Part5数据描述

Large Scale Web Search Session Data

Qid, Query, Query Reformulation
Pos 1, URL MD5, Title, Abstract, Multimedia Type, Click, -, -, Skip, SERP Height, Displayed Time, Displayed Time Middle, First Click, Displayed Count, SERP's Max Show Height, Slipoff Count After Click, Dwelling Time , Displayed Time Top, SERP to Top , Displayed Count Top, Displayed Count Bottom, Slipoff Count, -, Final Click, Displayed Time Bottom, Click Count, Displayed Count, -, Last Click , Reverse Display Count, Displayed Count Middle, -
Pos 2, URL MD5, Title, Abstract, Multimedia Type, Click, -, -, Skip, SERP Height, Displayed Time, Displayed Time Middle, First Click, Displayed Count, SERP's Max Show Height, Slipoff Count After Click, Dwelling Time , Displayed Time Top, SERP to Top , Displayed Count Top, Displayed Count Bottom, Slipoff Count, -, Final Click, Displayed Time Bottom, Click Count, Displayed Count, -, Last Click , Reverse Display Count, Displayed Count Middle, -
......
Pos N, URL MD5, Title, Abstract, Multimedia Type, Click, -, -, Skip, SERP Height, Displayed Time, Displayed Time Middle, First Click, Displayed Count, SERP's Max Show Height, Slipoff Count After Click, Dwelling Time , Displayed Time Top, SERP to Top , Displayed Count Top, Displayed Count Bottom, Slipoff Count, -, Final Click, Displayed Time Bottom, Click Count, Displayed Count, -, Last Click , Reverse Display Count, Displayed Count Middle, -
# SERP is the abbreviation of search result page.

Expert Annotation Dataset