Based on a simple product recommendation system blog Hbase Category: Architecture
Throughout the electricity industry's website, I stood on a user's perspective, there are many product recommendations:
One is through my search, view items that the system statistics to search, view other users to search the item to view the number of other commodities, the top-ranking recommended to me, a chestnut Dangdang:
I look at the "Hadoop The Definitive Guide", the system recommended to me by a bunch of other books:
One is through my recent searches, viewed commodity, the system I recommend some of it I think the interest of goods, Taobao a chestnut:
There are several, I feel quite interesting:
In particular this search Hadoop users, and ultimately a lot of people have bought a down jacket, this is why? Learning hadoop cheese are very cold it? Or, northerners?
- Multifarious Ha, another matter. Dir only discuss one of the next person YY Design: based on a simple commodity recommendation system Hbase of.
- Leaving aside other processes do not say, solely on the terms of this recommendation, Hbase based design, two tables on OK. A table user_item record each user to view all the goods, item_user record view for all users of an item.
- user_item: userid as revolves, as clusters and column item: itemid, data are as follows:
- user1 item:itemid timestamp=1234567891, value=item1
- user1 item:itemid timestamp=1234567892, value=item2
- user1 item:itemid timestamp=1234567893, value=item3
- user2 item:itemid timestamp=1234567894, value=item4
- user2 item:itemid timestamp=1234567895, value=item5
- user3 item:itemid timestamp=1234567881, value=item1
- user3 item:itemid timestamp=1234567832, value=item2
- user4 item:itemid timestamp=1234567843, value=item3
- user4 item:itemid timestamp=1234567854, value=item4
- user4 item:itemid timestamp=1234567895, value=item5
- ......
- item_user：itemid作为行健，列簇和列为：user:userid，数据如下:
- item1 user:userid timestamp=1234567891, value=user1
- item1 user:userid timestamp=1234567892, value=user2
- item1 user:userid timestamp=1234567893, value=user4
- item2 user:userid timestamp=1234567894, value=user3
- item3 user:userid timestamp=1234567895, value=user2
- item3 user:userid timestamp=1234567881, value=user4
- item6 user:userid timestamp=1234567832, value=user5
- item6 user:userid timestamp=1234567843, value=user6
- item8 user:userid timestamp=1234567854, value=user5
- item8 user:userid timestamp=1234567895, value=user4
- ......
- 大概的业务是：我查看《Hadoop权威指南》（item1）时，系统从item_user表中以item1作为行健查询出所有查看过item1的用户，再分别以各userid为行健，从user_item表中查询出所有查看过的商品，最后去重、统计、排序并显示。
http://f.dataguru.cn/thread-33415-1-1.html