天池大数据竞赛----IJCAI SocInf'16 Contest-Brick-and-Mortar Store Recommendation

原创 2016年06月01日 12:42:52

标题少写了 with Budget Constraints



Brick-and-Mortar Store Recommendation with Budget Constraints
IJCAI (http://ijcai-16.org/) is a leading international conference in AI area, and often organizes various activities to promote AI innovations, e.g. hosting a sequence of competitions. In this year, Alibaba Group and Ant Financial Service Group will hold a data mining competition oflocation base services, in cooperation with IJCAI workshop SocInf 2016 (http://socinf2016.isistan.unicen.edu.ar/). Winners will be invited to present their solutions at the workshop in New York City with travelling support.

As mobile devices become ubiquitous in our daily life, location based services (LBS) have become increasingly more important. People are getting more comfortable sharing their real-time locations with various location-based services, such as navigation, car ride hailing, restaurant/hotel booking, etc. As a result, huge amount of user data has been accumulated, which ignites the excitement from the machine learning/data mining community to join in force revealing the magic in the Pandora’s box of our daily life, where the high dimension time-space complexity needs to be explored.
In this contest, we will focus on nearby store recommendation when users enter new areas they rarely visited in the past. The contest has two novelties:
First, you are supposed to investigate whether the correlation between online and on-site preference helps in recommending nearby stores.Alibaba Group owns Taobao.com and Tmall.com, the largest online retail platforms in China, serves for more than 10 million merchants and over 300 million customers. Meanwhile, the Ant financial’s Alipay offers restaurant and retail store recommendation and payment services, named Koubei, for a number of customers. A user enjoying services provided by these two groups often has a unified online account. While Taobao and Tmall have run for many years and accumulate vast consumers’ behavior data, the nearby recommendation/payment services provided by Alipay are relatively new thus with less data.
Second, a set of budget constraints is imposed on the recommender system, for example, due to service capacity or number of coupons available at the stores.
As far as we know, such contest setting is novel to the research community, though it is critically important to the blooming location based business




How to use extra data.pdf

.pdf (567KB) 

ODPS User Manual.zip

.zip (3MB) 

Tianchi User Manual.pdf

.pdf (2MB) 


.csv (5KB) 

data sets


data sets (mirror link, code: CVVz9P)


In this contest, we aim to predict users' preference on Dec., 2015 (Table 4) based on his/her online/on-site behavior between July 1 st , 2015 and Nov. 30 th , 2015 (Table 1, 2). Moreover, budget constraints are imposed on each merchant (Table 3), simulating the limited discount/coupons available.

This task involves following data accumulated on Tmall.com/Taobao.com and the app Alipay.

Remark 1: Due to both business and noise concerns, we remove data in the great promotion period. That is, Nov. 01-Nov. 20 in Table 1 and Dec.12 in Table 4. 
Remark 2: Data are biased sampled from the daily log, thus its distribution would be different from the distribution of our entire business. Nevertheless, we believe it won’t affect too much on users’ preference prediction. 
Remark 3: The `budget’ constraints in Table 3 denotes the number of coupons the merchant offers in December, which is estimated by domain knowledge. See the ``evaluation metric’’ for more detail. 
Remark 4: To prevent overfitting, predicted results associated to half of the merchant set are evaluated at this stage, while the full set will be used after May 24. That is, some merchants are directly removed from both your recommendation list and the truth without evaluation before updating.



 What’s the relationship between users in Table 1 and 2? 
A: As mention in the contest introduction, there is a huge amount of overlapping users between alipay and Taobao/Tmall. An identical user ID from Table 1 and 2 refers to a unique user, which means you may utilize users’ online behavior/relationship to improve your prediction accuracy.




天池离线赛 - 移动推荐算法(一):题目与数据解析



转自:http://sanwen8.cn/p/151ompp.html 这篇文章记录2015阿里天池大数据竞赛中,CSDN博主@wphh 的一些代码,由于代码分享时比赛正在进行中,基于规则,仅分...




关注公众号 在公众号里回复“”秘密“”两个字 返回 http://task.csdn.net/m/task/home?task_id=398 领取奖励 提示:根据公众号里的自动回复,完成...

属性动画----把图片渐渐变小不见(主函数MainActivity 页面)(XML布局)(本布局和渐变布局一样)

LinearLayout xmlns:android="http://schemas.android.com/apk/res/android" xmlns:app="http://schema...

JavaEE 6及以上版本的web.xml问题?

JavaEE 6及以上版本的web.xml问题?MyEclipse JavaEE 6版本开始web.xml突然消失不见?没这回事,只是不太必须而已,有需要的项目可以自行进行添加或在创建项目的时候点击n...

spring集成 JedisCluster 连接 redis3.0 集群

maven依赖: redis.clients jedis 2.8.0 2. 增加spring 配置 ...

Android 图片毛玻璃的实现方法

注:本文的高斯模糊只能显示,如果想要保存模糊后的图片,请参考另一篇文章:http://blog.csdn.net/fan7983377/article/details/51568059 效果...