20
Recommendation algorithm is a product 孙孙 孙孙孙 [email protected]

孙超 - Recommendation Algorithm as a product

Embed Size (px)

Citation preview

Page 1: 孙超 - Recommendation Algorithm as a product

Recommendation algorithm is a

product孙超 刘凯义

[email protected]

Page 2: 孙超 - Recommendation Algorithm as a product

Abstract

In most cases, we find the similarity between two users depend on the preference of items. But in some cases, we can define the similarity by the preference of different recommended way, and also different algorithm.

Page 3: 孙超 - Recommendation Algorithm as a product

Summary

How to display the algorithm The relationship between real products

and recommended algorithm Algorithm’s algorithm

Page 4: 孙超 - Recommendation Algorithm as a product

Algorithm as implicit product

Page 5: 孙超 - Recommendation Algorithm as a product

Algorithm as implicit product

algo1 algo2 algo3

algo4 algo5 algo6

Page 6: 孙超 - Recommendation Algorithm as a product

Why not think about…? Think about hybrid algorithm? Dose the customer like our algorithm? The personalize algorithm? The algorithm’s algorithm?

Page 7: 孙超 - Recommendation Algorithm as a product

How to produce

1) different model 2) different dataset 3) different parameter 4) different algorithm 5) hybrid algorithm

Page 8: 孙超 - Recommendation Algorithm as a product

Algorithm’s algorithm

K-Nearest Neighbor algorithm(knn)

AprioriContent-basedUser-basedItem-basedVector cosine

Page 9: 孙超 - Recommendation Algorithm as a product

The relationship

Dataset

Team1

Team2

Team3

Product1

Product2

Product3

Product4

Product5

Product6

Product7

Product8

UserA

UserB

Page 10: 孙超 - Recommendation Algorithm as a product

DatasetUser id Item id Time Algo id

4027065 10310198 2009-11-30 23:49:07 100025

4027065 10882081 2009-11-30 23:52:48 101025

3292669 10814423 2009-11-30 23:00:43 101025

3292669 10026349 2009-11-30 23:05:43 200003

3765231 10896495 2009-11-30 23:39:01 102175

3765231 10023192 2009-11-30 23:14:34 200503

3765231 10018038 2009-11-30 23:04:53 201801

3977917 10023488 2009-11-30 23:46:24 102175

4008825 10093427 2009-11-30 23:28:28 102175

4008825 10031710 2009-11-30 23:16:29 201801

4010098 10300130 2009-11-30 23:20:44 200003

4010098 10320031 2009-11-30 23:20:46 200003

Page 11: 孙超 - Recommendation Algorithm as a product

DatasetUser id Item id Time Algo id

1(Bob) 1( 青花瓷 ) 2009-11-30 23:49:07

1(user_base)

1(Bob) 2( 十年 ) 2009-11-30 23:52:48

1(user_base)

2(Linda) 2( 十年 ) 2009-11-30 23:00:43

1(user_base)

2(Linda) 5( 双截棍 ) 2009-11-30 23:05:43

2(item_base)

3(Lucy) 1( 青花瓷 ) 2009-11-30 23:39:01

3(conten_base)

3(Lucy) 3( 富士山下 ) 2009-11-30 23:14:34

4(apriori)

3(Lucy) 2( 十年 ) 2009-11-30 23:04:53

5(other)

3(Lucy) 4( 天黑黑 ) 2009-11-30 23:46:24

3(conten_base)

4(Tom) 4( 天黑黑 ) 2009-11-30 23:28:28

2(item_base)

4(Tom) 5( 双截棍 ) 2009-11-30 23:16:29

4(apriori)

5(Peter) 6( 花木兰 ) 2009-11-30 23:20:44

5(other)

5(Peter) 3( 富士山下 ) 2009-11-30 23:20:46

4(apriori)

Page 12: 孙超 - Recommendation Algorithm as a product

Binary datasetuser_ba

seitem_bas

econten_b

aseother apriori

Bob 1 0 0 0 0

Linda 1 1 0 0 0

Lucy 0 0 1 1 1

Tom 0 1 0 0 1

Peter 0 0 0 1 1

Page 13: 孙超 - Recommendation Algorithm as a product

Dispatcher

User ID Other user ID Algorithm Similarity

Bob Linda 0.589723

Bob Tom 0.279055

Linda Tom 0.279055

Lucy Tom 0.227848

Lucy Peter 0.481507

Tom Peter 0.279055

Page 14: 孙超 - Recommendation Algorithm as a product

User base User id User id Similarity

Bob Linda 0.279055

Bob Lucy 0.416997

Linda Lucy 0.197322

Linda Tom 0.310667

Linda Tom 0.219675

Lucy Peter 0.219675

Page 15: 孙超 - Recommendation Algorithm as a product

Item baseItem id Item id Similarity

青花瓷 十年 0.463457

青花瓷 富士山下 0.256949

青花瓷 天黑黑 0.256949

十年 富士山下 0.209798

十年 天黑黑 0.209798

十年 双截棍 0.253659

富士山下 天黑黑 0.256949

富士山下 花木兰 0.43935

天黑黑 双截棍 0.310667

Page 16: 孙超 - Recommendation Algorithm as a product

Content base

青花瓷 菊花台 晴天 七里香

十年 K 歌之王 背包 从何说起

富士山下 爱情转移 好久不见 在你身边

天黑黑 遇见 开始懂了 花木兰

双截棍 稻香 断了的弦 霍元甲

木兰情 天黑黑 爱情证书 原点

Page 17: 孙超 - Recommendation Algorithm as a product

AprioriK-Items Min-Sup Confidence

青花瓷 , 十年 30% 青花瓷十年 =100%

青花瓷 , 十年 30% 十年青花瓷 = 66.7%

Page 18: 孙超 - Recommendation Algorithm as a product

Other一直走 张倩

带我飞 林志玲

叹金莲 阿朵

把握你的美 江映蓉

看月亮爬上来 张杰

Page 19: 孙超 - Recommendation Algorithm as a product

Data stream

Page 20: 孙超 - Recommendation Algorithm as a product

End