Rakesh Agrawal Technical Fellow Search Labs, Microsoft Research – Silicon Valley.

31

Transcript of Rakesh Agrawal Technical Fellow Search Labs, Microsoft Research – Silicon Valley.

Page 1: Rakesh Agrawal Technical Fellow Search Labs, Microsoft Research – Silicon Valley.
Page 2: Rakesh Agrawal Technical Fellow Search Labs, Microsoft Research – Silicon Valley.

Rakesh AgrawalTechnical Fellow

Search Labs, Microsoft Research – Silicon Valley

Whither Search?

Page 3: Rakesh Agrawal Technical Fellow Search Labs, Microsoft Research – Silicon Valley.

Outline

Current state of affairEvolving searchSearch labs projects

Page 4: Rakesh Agrawal Technical Fellow Search Labs, Microsoft Research – Silicon Valley.

Outline

Current state of affairEvolving SearchSearch Labs projects

Page 5: Rakesh Agrawal Technical Fellow Search Labs, Microsoft Research – Silicon Valley.

Does Search Work? Navigational Queries

Pseudo- Navigational Queries

Page 6: Rakesh Agrawal Technical Fellow Search Labs, Microsoft Research – Silicon Valley.

My Recent SearchesCar GPS around $300Four day trip to Bhutan from Delhi to visit important Buddhist places

Page 7: Rakesh Agrawal Technical Fellow Search Labs, Microsoft Research – Silicon Valley.

Best Car GPS Around $300

Game Consol

es

Party Site

Page 8: Rakesh Agrawal Technical Fellow Search Labs, Microsoft Research – Silicon Valley.

Bhutan From Delhi in 4 Days

Page 9: Rakesh Agrawal Technical Fellow Search Labs, Microsoft Research – Silicon Valley.

User Expectations

Page 10: Rakesh Agrawal Technical Fellow Search Labs, Microsoft Research – Silicon Valley.

Observations

Search queries are not grammatically correct questions, but they are not bags of words eitherQuery terms are often more than strings of charactersData often has structure or structure can be derivedSearch can span multiple sessions over several daysSearch often provides entry point for browsing and search and browsing are inter-mixedExpectations from search are increasing

Page 11: Rakesh Agrawal Technical Fellow Search Labs, Microsoft Research – Silicon Valley.

Outline

Current state of affairEvolving searchSearch labs projects

Page 12: Rakesh Agrawal Technical Fellow Search Labs, Microsoft Research – Silicon Valley.

From Search to Search-driven Applications

HealthEducation

Humanity’s greatest advances are not in its discoveries – but in how those discoveries are applied to reduce inequity.

Bill Gates Harvard Commencement. June

7, 2007

Page 13: Rakesh Agrawal Technical Fellow Search Labs, Microsoft Research – Silicon Valley.

Humane Computing

“Is it right? Is it just? Is it in the interest of mankind?” Woodrow Wilson. May 30, 1919.

Applications to benefit individuals and society

Page 14: Rakesh Agrawal Technical Fellow Search Labs, Microsoft Research – Silicon Valley.

Changing Nature of Disease

New Challenge: chronic conditions- illnesses and impairments expected to last a year or more, limit what one can do and may require ongoing careIn 2005, 133 million Americans lived with a chronic condition (up from 118 million in 1995)

Deaths due to infectious diseases

118

125

133

141

149

157

164

171

100

120

140

160

180

1995 2000 2005 2010 2015 2020 2025 2030

Year

Num

ber

of P

eopl

e W

ith

Chr

onic

Con

ditio

ns (m

illio

ns)

Page 15: Rakesh Agrawal Technical Fellow Search Labs, Microsoft Research – Silicon Valley.

Technology TrendsTremendous simplification in the technologies for effortlessly capturing useful personal information

Dramatic reduction in the cost and form factor for personal storage

Cloud Computing

Page 16: Rakesh Agrawal Technical Fellow Search Labs, Microsoft Research – Silicon Valley.

Personal Health Analytics

Page 17: Rakesh Agrawal Technical Fellow Search Labs, Microsoft Research – Silicon Valley.

Personal Data Mining

Charts for appropriate demographics?

Optimum level for Asian Indians: 150 mg/dL(much lower than 200 mg/dL for Westerners)

Due to elevated levels of lipoprotein(a)*

Distributed computation and selection across millions of nodes

Privacy and security

*Enas et al. Coronary Artery Disease In Asian Indians. Internet J. Cardiology. 2001.

Page 18: Rakesh Agrawal Technical Fellow Search Labs, Microsoft Research – Silicon Valley.

India’s Education System: 1951-20021951 1981 2002(*)

Literacy Percentage 18.33% 43.57% 65.38%

Educational InstituionsPrimary 209671 494503 664041Upper Primary 13596 118555 219626High/Hr. Seconday & Inter & Pre Junior College 7416 51573 133492

Enrollements (in millions)Primary 19.2 73.8 113.9Upper Primary 3.1 20.7 44.8High/Hr. Seconday & Inter & Pre Junior College 1.5 11 30.5

Dropout Rates (%) NA 82.5 66

Teachers (in '000)Primary 538 1363 1928Upper Primary 86 669 1157High/Hr. Seconday & Inter & Pre Junior College 127 926 1777

Pupil Teacher RatioPrimary 20 38 43Upper Primary 20 33 34High/Hr. Seconday & Inter & Pre Junior College 21 27 34

Public Expenditure (% of GDP) 0.64% 2.92% 4.02%

Significant achievements, but problems remain …

Page 19: Rakesh Agrawal Technical Fellow Search Labs, Microsoft Research – Silicon Valley.

Problems of School Level Education in IndiaPoor performance• 39% dropouts in primary, additional 15.6% in secondary, additional 11.7% in

higher secondary

• Pass out ratio is 50% at Class X and majority of them pass in 3rd division

• Less than 8% finish all schooling to qualify for a college education

Poorly trained teachers• 51% of primary teachers are higher secondary or below

• Only 44% have received in-service training

• Absence of learning material for teachers to update their knowledge

Poor teacher-student ratios• Ratio in primary is 1:43, secondary and Higher secondary is 1:34. About 9%

of primary schools have a teacher-student ratio > 1:100

• 1.4% of primary schools have no teachers, 19% have only one teacher for all classes

Poor quality of material• Poor quality of textbooks, out-dated curriculum

Source: IBM Report on Improving India’s Education System through Information Technology, 2005

Page 20: Rakesh Agrawal Technical Fellow Search Labs, Microsoft Research – Silicon Valley.

1. Define goal.2. Find the highest-

leverage approach.3. Discover the ideal

technology for that approach.

4. In the meantime, make the smartest application of the technology on-hand.

1. Quality education to all.2. New pedagogy.3. Individualized learning

with teacher as a discussant.

4. Internet-based mass collaboration to help teachers teach better and improve the educational infrastructure.

Framework Application to Education

Attacking Complex Problems*

* Bill Gates. Harvard Commencement. June 7, 2007

Page 21: Rakesh Agrawal Technical Fellow Search Labs, Microsoft Research – Silicon Valley.

Education Web• Participation of experts,

teachers, parents and students in the development and revisions of curricula

• Sharing and collaborative development of lectures, assignments, tests, etc.

• Tools for capturing feedback on textbooks (errors, better explanations, supplementary readings)

• Collaborative translation and localization of educational material

Page 22: Rakesh Agrawal Technical Fellow Search Labs, Microsoft Research – Silicon Valley.

Outline

Current state of affairEvolving searchSearch labs projects

Page 23: Rakesh Agrawal Technical Fellow Search Labs, Microsoft Research – Silicon Valley.

23

web mining

information retrieval

machine learning

data management

algorithmsprivacy

inconsistent data

parallel miningranking

link analysis

query processing

computational economics

game theory

Search Labs

NLP

Invent next in Internet search and

applications

Page 24: Rakesh Agrawal Technical Fellow Search Labs, Microsoft Research – Silicon Valley.

Best car GPS around $300

Best car GPSaround $300

Category = “Auto GPS”Price = approx(300)Order By ReviewRank

Structure and Semantics in Data and QueriesInsights on user behavior from massive data miningFrom ranking to decision makingTask-orientation

Lincoln: Experimental Commerce Search Engine

Page 25: Rakesh Agrawal Technical Fellow Search Labs, Microsoft Research – Silicon Valley.

Customer Data Data Services

Search

ShoeQueen

<div id=results> </div>

<javascript />

ShoeQueen ShoeQueen

submit

ShoeQueen Data

Query/Click Logger

Query, Click,

App-ID

Advertising

clicksquery

Symphony Runtime

Rev Sharing Component

“Trail running shoes”

Query

Items

1. Collect initial results2. Add additional data3. Generate HTML

ShoeQueen Config

+

Semi-structured Query Processing

forum/review pages images ads

Advertising

Advertising

Web Image

Advertising … 3rd PartyProprietary

VideoNews

Symphony

Enable non-developers to create and monetize custom search applications that combine their data and knowledge with Search services.

Page 26: Rakesh Agrawal Technical Fellow Search Labs, Microsoft Research – Silicon Valley.

EduWayTools for creating and updating content (Wikipedia++)Trust and authoritativeness of contentPersonalization of search to find the material suitable for one’s own style of teachingBootstrapping and incentives

Page 27: Rakesh Agrawal Technical Fellow Search Labs, Microsoft Research – Silicon Valley.

Concluding Remarks

27

Search is becoming an essential “utility”Need to develop new foundations and abstractions to take search to next levelAcademia can (and must) play a leading role

Page 28: Rakesh Agrawal Technical Fellow Search Labs, Microsoft Research – Silicon Valley.

Thank you!

28

Search Labs’ mission is to invent next in Internet search and

applications

Page 29: Rakesh Agrawal Technical Fellow Search Labs, Microsoft Research – Silicon Valley.

What I Ended up Buying?

Page 30: Rakesh Agrawal Technical Fellow Search Labs, Microsoft Research – Silicon Valley.

Last Resort!

Page 31: Rakesh Agrawal Technical Fellow Search Labs, Microsoft Research – Silicon Valley.

How About Travel Sites?