About Our Recommender System

18
Silver Egg Technology Co., Ltd. About Our Recommender System Kimikazu Kato, Chief Scientist

description

The material presented at the spring session of Operations Research Society Japan.

Transcript of About Our Recommender System

Page 1: About Our Recommender System

Silver Egg Technology Co., Ltd.

About Our Recommender System

Kimikazu Kato, Chief Scientist

Page 2: About Our Recommender System

1

Table of Contents

• About myself

• About the company and its business

• Survey on related researches

• Conclusion

Page 3: About Our Recommender System

2

About Myself

Kimikazu Kato

• Ph.D in computer science, background in

mathematics

• Joined Silver Egg as a Chief Scientist in Nov. 2012

• Experiences in numerical computation

– 3D CAD, geometric computation

– Computer graphics

– Partial differential equation

– Parallel computation, GPGPU

• Now designing the core of recommender system

Page 4: About Our Recommender System

3

About Silver Egg Technology

≪著書≫「One to Oneマーケティングを超えた

戦略的Webパーソナライゼーション」

(出版社:日経BP社 発売:2002年5月)

「ASP・SaaS・ICTアウトソーシングアワード2009」

ASP・SaaS部門「委員長特別賞」受賞

第8回(2010)、第9回(2011) 「デロイト 日本テクノノロジー Fast50」受賞

Book written by CEO Silver Egg Technology

Established: September, 1998

CEO: Tom Foley

COO: Junko Nishimura

Capital: ¥78 Million

Main Services:

Recommender System

Online Advertisement Service

Page 5: About Our Recommender System

4

Ranking

Additional

Cross-sell

Combination

XXXXXX XXXXXXXXX 3,800円

Recommender System

No.1 No.2 No.3

Recommender system proposes the items best fit for individuals’ needs.

Good recommender system provides a comfort for online shopping

experiences and improves customer loyalty.

Page 6: About Our Recommender System

5

Aigent Mail

Aigent Personalized

Search Transaction

Mail Event Driven

Mail Recommender

Mail

Recommender

Aigent Recommender Recommado

Aigent LPO

Aigent Gadget Portal

HotView Retargeting ad

Aigent On-Demand Printing

Traffic inflow Service Conversion Retention

Pre-access On-access Post-access

Consistent behavior targeting

Consistent user behavior targeting from “traffic inflow” to “retention” is

essential for improving sales and profit.

Aigent Suite (Real Time Recommender Platform)

Silver Egg Technology provides smart targeting technology which enables optimization of online marketing

Page 7: About Our Recommender System

6

Media Dashboard

ブティック

百貨店

TVショッピング

通販カタログ

アパレル

Aigent Suite

Consumer

Merchandizer

Shows ads of items to promote to the target users

Retailer

Recommendation for up-sell and cross-sell

Discovery in a media site

- Timestamp - Geographic information - Use behavior - Demands - Contexts (search words) To the shopping site

To the site they are interested

HotView

Aigent

Interaction of Advertisement and Recommender

Ad contents based on users behaviors in shopping sites are more likely to attract attentions and effectively lead users back to those sites

-Registers items to promote - Checks performance

Page 8: About Our Recommender System

7

Mechanism

Client’s EC site

“Who bought what”

“Who is browsing what”

Aigent server

“What should be recommended”

Characteristics:

• Real time response

• Implemented as an add-on (cost efficient)

Code snippet to connect

with AIgent

+

Stored and analyzed

Respond in real time

ASP service Batch update of inventory

Page 9: About Our Recommender System

8

Consulting Services

• Just showing the result of mathematical

computation is not enough

• To extract optimal sales, parameters should be

tuned by hand

– Statistical co-relation is not all that matters.

• Sometimes recommendations should reflect some

“intention”

– According to policy, strategy, etc.

• Continuous monitoring and A/B testing

Page 10: About Our Recommender System

9

About recommendation algorithms

• Collaborative filtering

• Fruitful methods as a result of Netfilx Prize

– Neighborhood Models

– Matrix factorization

– Restricted Boltzmann Machines

Page 11: About Our Recommender System

10

Netflix Prize

The Netflix Prize was an open competition for the best collaborative filtering

algorithm to predict user ratings for films, based on previous ratings

— Wikipedia

Netflix provided open data for this competition

Closed in 2009

Page 12: About Our Recommender System

11

Movie Rating Prediction

W X Y Z

A 5 4 1 4

B 4

C 2 3

D 1 4 ?

user

movie

Each user gives rating to the movies they saw

Is it possible to predict the rating of unknown user/movie pair?

Ratings are expressed as a sparse matrix.

A zero value of the matrix doesn’t really mean “zero” but “unknown”

Page 13: About Our Recommender System

12

Probabilistic Matrix Factorization

Regarding ratings are expressed by small number of components

noise

Approximate only the non-zero elements

𝐴 𝑈𝑇 𝑉

Page 14: About Our Recommender System

13

According to Bayes’ theorem,

Minimize this objective function

Page 15: About Our Recommender System

14

Rating vs Purchase

W X Y Z

A 5 4 1 4

B 4

C 2 3

D 1 4 ?

W X Y Z

A 1 1 1 1

B 1

C 1

D 1 1 ?

user

movie

user

item

Movie rating Purchase recommendation

Predicts the rating for the user and

movie pair.

Predicts how likely the user buy the

item

The matrix includes negative feedback

(Some movies are rated as “boring”) No negative feedback

(No reason is given for missing elements)

=> Strong bias toward 1

Only one kind of value for known elements

=> Gives more degree of freedom

A method successful in movie rating prediction is not

useful for recommendation of usual shopping site.

Page 16: About Our Recommender System

15

Solutions

• Regard a zero element as a negative feedback

– Too ad hoc but better than naïve PMF

• Assume a certain ratio of zero elements becomes

one at the optimum [Sindhwani et al. 2010]

– Assign other variables to zero elements and solve a

relaxed optimization

– Experimentally outperform the “zero-as-negative”

method.

V.Sindhwani et al., One-Class Matrix Completion with Low-Density Factorizations. In Proc. of ICDM

2010: 1055-1060

Page 17: About Our Recommender System

16

Minimize

Subject to:

Solve this relaxed problem for non-negative variables

Page 18: About Our Recommender System

17

Conclusion

• Scientific approach is important

– Math really makes money

• But that alone is not enough for real business

• Engineering matters

– Efficient platform and easy-to-deploy mechanism

• Hand tuning part always remains

– Consulting for parameter tune is essential