Pulse Lab Jakarta Launch Presentation

Pulse Lab Jakarta Launch / Jakarta, Indonesia / 1 October 2012

Launch Event Peluncuran 1 October 2012 / 1 Oktober 2012

BAPPENAS


Ir. Wismana Adi Suryabrata Deputy Minister for Development Funding, Bappenas


Prof. Dr. Armida S. Alisjahbana Minister of National Development

Planning Agency / Head of Bappenas


Mr. El-Mostafa Benlamlih United Nations Resident Coordinator


Robert Kirkpatrick Director, UN Global Pulse

HARNESSING THE POWER OF BIG DATA FOR

GLOBAL DEVELOPMENT

Robert Kirkpatrick

Hyperconnectivity

Airplane Flights

Internet Traffic

Telephone Calls

Social Networking

Real-time isn’t just faster. It’s different.

20th-Century tools…

Welcome to the age of

BIG DATA

Volume. Velocity. Variety….

Phoenician alphabet invented ca. 1000 B.C.E.

955 million users

400 million tweets per day….

or, 4600 per second

400 million tweets = 7 terabytes of data…

…..or120,000 file cabinets full of paper

1.3 billion

1.2 billion

955 million

309 million

238 million

If social networks were countries…

~500 million

Jakarta - 9 million tweets every day

PRIVATE SECTOR Monitor operations…in real time

Track market trends…in real time

Get customer feedback…in real time

GLOBAL DEVELOPMENT Unemployment?

Food Security?

Public Health?

Education?

Migration?

Disaster Relief?

1.  Better early warning: Faster detection of

anomalies in at the onset of a crisis allows rapid,

effective, efficient response.

2.  Real-time awareness: A more accurate and up-to-

date picture of population needs and concerns

informs better design of programs and policies;

3.  Real-time feedback: Understand sooner where

programs and policies are NOT working and

make rapid, adaptive adjustments until progress

is on track.

BIG DATA IN REAL TIME: OPPORTUNITIES

BIG DATA, PRIVACY, AND HUMAN RIGHTS

Global Pulse privacy guidelines: •  Never analyze personally identifiable data

Never analyze confidential data

•  Never seek to re-identify individuals

EXAMPLE: AIRTIME CREDIT PURCHASE DATA

Higher household

income

Lower household

income

Average # of purchases / month

Ave

rage s

ize o

f purc

hase

SIZE AND FREQUENCY PREDICT HOUSEHOLD INCOME

Lower

socioeconomic level

Higher

socioeconomic level

CALLING PATTERNS AND ECONOMIC OPPORTUNITY

Men: •  Fewer calls

•  Shorter calls

•  Smaller social network

•  More work-related calls

Women: •  More calls

•  Longer calls

•  Larger social network

•  More personal calls

MEN AND WOMEN USE THEIR PHONES DIFFERENTLY

Mobile phone locations predict cholera outbreaks

dominican

republic 11.7%

mexico 9.1%

united kingdom 7.5%

canada 7.0%

guatemala 5.5%

ecuador 5.3%

jamaica 3.3%

india 3.0%

germany 2.5%

philippines 2.4%

Using mobile phone movement to assess H1N1 response

dominican

republic 11.7%

mexico 9.1%

united kingdom 7.5%

canada 7.0%

guatemala 5.5%

ecuador 5.3%

jamaica 3.3%

india 3.0%

germany 2.5%

philippines 2.4%

TWITTER PREDICTS H1N1 INFLUENZA

GOOGLE SEARCHES FOR SYMPTOMS PREDICT DENGUE

RECENT SEARCHES FOR JOBS, COMMODITY PRICES

THREE PROOF OF

CONCEPT PROJECTS

GLOBAL PULSE 2011

Tracking Commodity Prices

through Online Retail

Advertising

• Date• Item• Price• Descrip-on

Use1scraping1technology

Connect1to1online1retailers

Find1individual1items

Calculate1PriceStats1Indexes

1 2 3 4 5Store1key1item1informa-on1in1a1database

Generating Real-Time Inflation Indices

ARGENTINA BRAZIL CHILE

COLOMBIA URUGUAY VENEZUELA

URUGUAY

Uruguay Bread

Index

(daily data)

Uruguay

Bread & Cereals CPI

(monthly data)

Online Discussions & Unemployment

Ireland

Online Discussions & Unemployment

United States

Tweets per day about food,

during Ramadan in Indonesia

Start of Ramadan

End of Ramadan

Number of tweets per month commenting on the price of rice

Wordcloud1 (sunday)

Wordcloud2 (Sunday)

Food inflation

0.47

0.44

-0.85

1.49

2.81

2.21

-0.33

-1.94

-1.9

-0.28

1.27

1.84

1.07

-0.09

Data inflation

Tweets about the

price of rice

(per month)

Official

Food Price Inflation

(monthly from 25 cities)

Tweets predict food basket inflation (rice, chilies, fish, sugar, corn, cooking oil)

GLOBAL PULSE ABOUT

Dynamic Development?

Data Philanthropy?

PULSE LABS

R&D INNOVATION STRATEGY

1.  Partner with governments to establish Pulse Labs

around the world.

2.  Build world-class teams of data scientists,

engineers, and policy experts

3.  Partner with private sector for access to real-time

data and cutting edge technology

4.  Design and conduct exploratory research

5.  Build an open source toolkit for real-time

monitoring

6.  Support broad adoption of useful innovations

7.  Share everything we learn and build

Pulse Lab Network

Pulse Lab Jakarta…………October 2012

Pulse Lab Kampala……….January 2013

Other locations…………...???


Dr. Lukita Dinarsyah Tuwo Vice Minister of National Development

Planning Agency / Bappenas


Initial Research Findings

from Pulse Lab Jakarta


Indonesia & Social Media

Source:

Indonesia 5th in worldwide Twitter users & volume

90% of Indonesian Internet users are on Facebook

10-25% of Indonesian regularly access Internet, including via Internet cafes and mobile devices

Language barriers, local slang and Indonesia-specific sites present new challenges for social media analytics.


Pulse Lab Jakarta An innovation lab for real-time, Big Data analysis

Source:


Research Agenda 2012-2013

Source:

Regions of focus:

•  Jakarta

•  Jogja

•  Makassar (South Sulawesi)

•  Medan (North Sumatera)

Topic of focus:

•  Changes in social welfare, especailly as related to

food prices, fuel prices (transport and cooking) and

employment


Research Agenda 2012-2013

Source:

Approaches for Exploration:

•  Social Media & Twitter Analysis

•  Mobile CDR Analysis

•  Rapid Mobile Surveys


Research Questions 2012-2013

Source:

How are these topics discussed on Social Media?

•  Which social media sources?

•  How much relevant conversation is there?

How can we monitor how they affect Indonesians?

•  Which analytical approaches?

•  Which technological capabilities?


Pre-Launch Research Topics Initial Exploration

Source:

Food Prices

•  Staple foods (sembako), rice (beras), sugar

(gula), eggs (telur)

Fuel Prices

•  Vehicle fuel (bensin), oil fuel (BBM), LPG

(elpiji)

Unemployment

•  Unemployed (pengangguran), severance pay

(pesangon)

1

2

3


Proof of Concept Research Initial Exploration

Source:

Data Sources

•  Sources: Twitter, Facebook, Blogs, News…

•  Real-time data + 2-year historical archive

Analytical Approaches

•  Identify trending topics in online conversation

•  Assess mood and sentiment for those topics

Results

•  Correlate with official data sources (CPI,

unemployment)

•  Visualize trends & map the results

101001001001001110010011


Research Tool 1 Crimson Hexagon: ForSight

Source:


Research Tool 2 SAS Social Media Analytics and SAS Text Miner

Source:


The Signals Are Getting Stronger

Source:

0

5000

10000

15000

20000

25000

30000

35000

40000

minyak (oil) ketahanan pangan (food security) budidaya (cul>va>on) telur (eggs)

è Big increase in volume of relevant conversations over 18 months

Indonesians are increasingly using social media to discuss basic needs


So Are the Temporal Correlations

Source:

è Listening to social conversations provides insight on official data

-‐2.5

-‐2

-‐1.5

-‐1

-‐0.5

0

0.5

1

1.5

2

2.5

Social Media Food Index

BAPPENAS Food Price Index


Overview of Preliminary Findings

Source:

Food prices influenced by global events

•  Global soy bean shortage

•  Rising prices in Indonesia

•  Online complaints about affording food

•  New tool to detect impact faster?

Fuel prices affected by government policy

•  Cuts in subsidies => price rise

•  Social outcry - worry, uncertainty, anger

•  New tool to estimate vulnerability?

Social media predicts official statistics

•  New tool to derive price index?

1

2

3


The Way Forward

Source:

Building up Pulse Lab capabilities:

•  Build interdisciplinary R&D team that consists of

Government, private sector, academia, NGOs and

development partners.

•  Develop a framework for ensuring data privacy

Developing Partnerships:

•  Secure access to new sources of real-time data,

particularly with mobile carriers.

•  Secure access to state-of-the-art technology for analysis

•  Translate analysis technologies into Bahasa and other local

languages

•  Collaborate with BIG and CSIRO on geo-location projects.


Next Steps

Source:

•  Through key partnerships and projects, continue to

explore the usefulness of new digital data sources and

new analytical tools.

•  After gaining 2 years’ of experience with the Pulse Lab

approach, assess whether this approach can contribute

usefully to improvements in social protection, early

warning, and Government Development Planning

Dr. Vivi Yulaswati, MSC Director of Social Protection and Welfare / Bappenas

Social Media Research: Methods and Preliminary Results


OUTLINE

Source:

1. Mining Social Media: Topics & Research Questions 2. Research Tools & Results •  Crimson Hexagon •  SAS

3. Discussion


TOPICS FOR EXPLORATION

Source:

1. Food Prices staple food (sembako), rice (beras), sugar (gula), eggs (telur)

2. Fuel Prices

vehicle fuel (bensin), oil fuel (BBM), LPG (elpiji)


BIG QUESTIONS

Source:

•  How are food and fuel prices in Indonesia being discussed on Social Media?

•  How can we monitor how they affect Indonesians? •  Can social media serve as a real-time approximation of

official statistics?


OUTLINE

Source:


3. Discussion

Crimson Hexagon: Methods �Human Training 1

beras OR telur OR gula

Posi>ve

Neutral

Nega>ve

Irrelevant

Choose Keywords

Define Categories

Label Examples

Select Data Sources

Step

I love when mum cooks eggs instead of tempeh… Hope prices rise more!

If tofu price goes higher, I can’t afford any more :(

èHuman labels a few examples to teach machine.

Crimson Hexagon: Methods �Machine Classifica9on 2

beras OR telur OR gula

Posi>ve

Neutral

Nega>ve

Irrelevant

Match Keywords

Detect Categories

Examine Results

Consume Data Sources

Step

èMachine analyzes millions of tweets automa>cally.

+4%

-‐1%

-‐1%

+3%

Food Prices: Global Influence�

23 -‐ 28 July Spikes of comments on tempe

and tofu

25 -‐ 27 July DPR urges government to reduce soy bean import duty and give

subsidy.

23 July Global soy bean price hike. Indonesian produc>on of tempe and tofu cut back.

10 July – 2 August 2012 Global soy bean shortage. ‘tahu’ and ‘tempe’ become hot topics in Bahasa tweets

10-‐22 July: Not many men>ons of ‘tempe’ and ‘tahu’

10 July – 2 August 2012 Global soy bean shortage. ‘tahu’ and ‘tempe’ become hot topics in Bahasa tweets

Food Prices: Example Tweets �“nyokap ngomel-‐ngomel harga tempe tahu naik :D” “mom’s grumbling because the tempe tahu price’s up :D”

-‐-‐ Nega9ve to price rise (Jul 26, 2012) “udeh.. kebanyakan ngomong lu dit, buruan nonton sono ntar keburu harga kedelai naik lagi...” “stop it.. you talked too much, go watch it before the soy bean price up again...”

-‐-‐ Realize price rise, no emo9on (Jul 26, 2012)

“Harga Kedelai Naik, Semes>nya Pemerintah Beri Subsidi” “Soy bean price rise, Govt. supposed to give subsidy”

-‐-‐ Announcement with no sen9ment (Jul 28, 2012)

“Harga tempe naik.., hmmm, kenapa kedelai impor?, memangnya tanah Indonesia ga subur untuk kedelai? #curhatpenggemartempe” “Tempe price’s up.., hmmm, why we import soy bean?, is Indonesian land inferCle to plant soy? #tempe-‐lover’s-‐vent-‐out”

-‐-‐ Confused reac>on to price rise (Jul 28, 2012)

“disaat harga tahu dan tempe naik, alhamdulillah mama masih bikin tahu dan tempe balado :D” “during this tofu and tempe price hike, thanks God mom sCll can cook spicy tahu and tempe :D“

-‐-‐ Posi9ve/unaffected by price rise (Jul 30, 2012)

Food Prices: Crisis Passes �

14 -‐ 21 Aug Ramadhan / Idul Fitri

10 July – 25 August 2012 Ramadhan / Idul Fitri

Food men>ons not as high as during ‘tempe’ & ‘tahu’ crisis

23 July -‐ 2 Aug ‘tempe’ and ‘tahu’ hot debate

Food Prices: The Next Topic�

14 -‐ 21 August ‘daging’ (meat) and

‘sembako’ (staple food) become main topics

“Tempe tahu udh naik harga, skrng daging menyusul +_+“ “Tempe tahu price has hiked, now meat follows +_+ ”

-‐-‐ Aug 16, 2012

“Dedek jgn keseringan minta dibua>n mie telor ya. harga sembako naik. mamah susah.” “LiFle kid, please don’t ask to cook noodle egg too oGen. The staple food price’s hiking. Mom’s troubled.”

-‐-‐ Aug 16, 2012

“Langsung belanja, harga daging naik ! mentok sekilo 30rb. #lol “ “Let’s go shopping now, meat price hike! It reached Rp 30.000 per kg. #lol ”

-‐-‐ Aug 14, 2012

Food Prices: Social vs Official �

-‐1 -‐0.5

0 0.5 1

1.5 2

February March April May June July August

Spikes in social media conversa>ons about food prices…

… are reflected in infla>on sta>s>cs measured by govt.

CPI, % change

Fuel Prices: Local Event�

18 March – 7 April Fuel subsidy cut plan and

protests: ‘bbm’, ‘makanan’ & ‘sembako’ trending topics

“BBM naik , harga makanan juga naik x_x “ “Fuel price has hiked, now food follows x_x ”

-‐-‐ Mar 27, 2012

“Bagi rakyat kecil bkn masalah u/ mngurangi pnggunaan BBM, tp efek domino dr knaikan BBM (harga pangan, trnsportasi umum naik) yg memberatkan” “For poor, reducing fuel usage is not the problem, but the domino effects of fuel price rise (food, public transportaCon price rise) is”

-‐-‐ ” Mar 30, 2012

“Waduh bbm naik harga makanan naik saya bs agak kurusan ntar” “Whoa.. Fuel price rise, the food price rise, then I will become slimmer”

-‐-‐ Mar 26, 2012

Fuel Prices: Timeline�

25 Mar: Some DPR members agree to increase fuel price. Protests began

several days before.

22 Feb: News of government plan to increase subsidized fuel price

27 Mar: Massive protests against fuel price rise in Jakarta.

31 Mar: Govt. announces it will restore subsidies.

Few men>ons of “Fuel” on Twiner

15 Feb – 15 April 2012 Government plan to reduce fuel subsidies triggered

protests

Comparing Crises �

23 July -‐ 2 Aug ‘tempe’ and ‘tahu’ hot debate

18 Mar -‐ 7 Apr Fuel subsidy cut plan and

protests against it

Tweets about food

Social Protection Programs �

20 September ‘12: Jakarta governor elec>on,

2nd round

‘Social Protec>on Program’ –men>ons gepng higher over >me, esp. near Jakarta Governor elec>on

11 July ‘12: Jakarta governor elec>on,

1st round

OUTLINE

Source:


3. Discussion

SAS Analytic Workflow�

Internet Conversa>on

Global Pulse Relevance

Filter

Sen>ment, Mood & Influence

Topic & Geography Categories

Interac>ve Dashboard

1) Over 200,000 new Indonesian language documents per day

3) Capture sen>ment and mood for Bahasa

2) Extract conversa>ons about rice, cooking oil, fuel, employment, etc.

4) Detect loca>on, price, availability, specific govt. programs, etc.

5) Explore results and correlate with official sta>s>cs to official BPS sta>s>cs : Consumer Price Index (CPI) for 12 common foods

anxious, confident, confused, hos>le, sad, happy (-”-) ;-) ((+_+)) :@ :( :)

hnp://youtube.com/watch?v=MK-‐aHVfQ3BI

Indonesian Data Density�

è Data density across archipelagos of Indonesia

0

5000

10000

15000

20000

25000

30000

35000

40000

minyak (oil) ketahanan pangan (food security) budidaya (cul>va>on) telur (eggs)

The signals are growing stronger�èBig increase in volume of relevant conversations over 18 months

Indonesians are increasingly using social media to discuss basic needs

B �

-‐2.5

-‐2

-‐1.5

-‐1

-‐0.5

0

0.5

1

1.5

2

2.5

Social Media Food Index

BAPPENAS Food Price Index

So are the temporal correlations �èListening to social conversations provides insight on official data

Social Media Food Index BPS Food Price Index

Regression Weights The predic>ve power of each conversa>on topic

Linear Regression Analysis �

Infla>on (CPI)

Agricultural Cul9va9on (budidaya pertanian)

Food Security (ketahanan pangan)

Cooking Oil (minyak goreng)

Eggs (telur)

-‐0.1%

0.9%

0.1%

-‐0.5% 0.89

Social Media Topics BAPPENAS Sta>s>cs

Correla9on Coefficient 0 = no rela>onship 1 = perfect correla>on

=

OUTLINE

Source:


3. Discussion

ADVANTAGES

Source:

•  Real-time detection of trends to understand community concerns & coping mechanisms

•  Detect quick response to social protection plans and

policies •  Augment and complement official statistics with real

time indicators

CAVEATS

Source:

•  Current social media users do not represent the entire population => can’t measure direct response to poverty issues

•  Twitter content good for immediate, short-term trends => does not capture long-term aspirations and intentions

•  Machine classification methodologies need human supervision => are not able to deal with sarcasm, and complex human communications

•  Correlation ≠ Causation •  Noise, rumors & reliability

NEXT STEPS

Source:

New applications –  health –  natural disasters –  intention to migrate

New datasets

–  additional social media sources –  mobile phone call detail records (CDRs)

New research projects ready to start

–  UNICEF (child welfare and education) –  ILO (female labor condition and employment access)


Oktorialdi MA, PhD Head of PUSDATIN, Bappenas


Pulse Lab Operations & Roadmap


Source:

Innovation Lifecycle Phase 1 – Research

PHASE 1 – Research Phase

Carry out a series of “Proof of Concept” projects exploring

the utility of various sources of new digital data:

•  Exploratory research on social media, mobile data

(CDRs) and other relevant sources

•  Development or adaptation of technology tools for

analysis

•  Conduct analysis of historical data, and compare

against official statistics, to look for correlations and

proxy indicators

•  Publication and sharing of findings


Source:

Innovation Lifecycle Phase 2 – Prototype

PHASE 2 – Prototype Phase

If Proof-of-Concept phase proves that there are useful

insights within the new data source, built prototypes for

real-time monitoring:

•  Obtain sustainable access to digital data source in real-

time

•  Develop technology prototype for ongoing analysis

•  Sample from 4 cities: Medan, Jakarta, Jogjakarta,

Makassar


Source:

Innovation Lifecycle Phase 3 – Implementation

PHASE 3 – Use of Data in Real-Time

Decision Making and Planning

If prototypes are useful to the work of line ministries and

programme planning, mainstream and operationalize the

technique and the tools:

•  Mainstreaming use of new indicator/data source

•  Expand to larger-scale operational data analysis after a

year

•  Operationalize data monitoring software, hardware and

human expertise


Source:

YEAR 1 ROADMAP (2012 – 2013)

² Move to permanent Lab space

² Series of Proof-of-Concept research projects and prototypes

² Forge new partnerships around data, technology and

expertise

² Development of legal and data privacy frameworks

² Organize workshops & meet-ups on Big Data for

Development

² Knowledge sharing & publication of findings from research

projects


Source:

Work with Pulse Lab Jakarta Research Project Partners

•  Explore and test new digital data sources that could

contribute to real time program monitoring, or

complement an existing monitoring system.

•  Test data sources that could serve as proxy

development indicators.


Source:

Partner with Pulse Lab Jakarta Innovation Partners

•  Data Sharing Partners

•  Research Partners

•  Technology Partners

•  Host a Training Session

•  Academic Fellowships

•  Funding Support


Q&A


Innovation Showcase CSIRO SAS International Mercy Corps Microsoft LIFE Nokia Life PoliticaWave


BAPPENAS


Thank You!

Pulse Lab Jakarta Launch Presentation

Technology

pulse lab kampala

pulse labsaround

aboutglobal pulse

identifiable data

confidential data

terabytes of data

data philanthropy

power of big data