SlideShare una empresa de Scribd logo
1 de 30
A User Oriented
Modeling Analysis
of Cultural Backgrounds in Microblogging

Elena@Ilina.nl
Best Paper Award

http://www.asesite.org/awards/awards/164.html
2
Outline
1.
2.
3.
4.
5.

Introduction
Lewis Model of Cultures
Approach
Experimental Setup
Results

3
The Lewis Model of Cultures
Richard Lewis (2000) “When cultures collide:
Managing successfully across cultures”
Hispanic
America

MULTIACTIVE

Italy,
Portugal,
Spain

Argentina,
Brazil,
Chile,
Sub-Saharan
Mexico
Africa

USA

China

LINEARACTIVE

REACTIVE

UK

Germany,
Switzerland

Japan

Vietnam
4
Personality Traits
Multi-active

Linear-active
Talks half the time
Does one thing at a time
Plans ahead step by step
Polite but direct
Partly conceals feelings

Talks most of the time
Does several things at once
Plans grand outline only
Emotional
Displays feelings

Reactive

Listens most of the time
Reacts to partner’s action
Looks at general principles
Polite, indirect
Conceals feelings

5
Personalizing E-commerce
• Customized product descriptions
• User preferences and previous purchase history
- may not be directly available or not up to date
• Targeted advertisements
Web Site

Advertisiment Platform

http://google.com

Search Results

http://amazon.com

Web shop

http://groupon.com

Web site and e-mail

http://triggit.com

Facebook

6
Culture-oriented User Modeling
•
•
•
•

Adapting Applications to Cultural Origins
Using Social Web Data
Finding Microblogging Patterns
Creating Culture-oriented User Profiles
describing specific user preferences

When cultural background is not known,
can we find cultural cues from microblogs ?
7
Inferring User Cultural Traits
Culture-specific
User Traits
Differences in Behaviour

Microblogging
Patters

Differences in Microblogging

Adaptation
Employing User Profiles

Culture-oriented
User Modeling
Creating User Profiles

8
Content
Activity

• Tweeting Mobility (geo-locations)
• Posting on Weekends
• Friends and Followers
• User Mentions

Conversation

• URLs and Hashtags
• Automatically-detected Languages

Social

Twitter-specific Features

• Retweets and Replies
9
Example: German User
• User A from Berlin, German language specified in
Twitter Profile
• URLs and Hashtags: 49 and 4
• Automatically-detected Languages -2
• Tweeting Mobility (geo-locations) -7
• Posting on Weekends -23 out of 100
• User Mentions - 75
• Friends and Followers: 50 and 96
• Retweets and Replies: 1 and 28
Example: German User
Gute Nacht! #TWoff
Weekend’s Tweet
Language: de
tweet place: Berlin
I'm at Laroy w/ @username http://t.co/Ct0ObmPz
URLs: 0
Workday’s Tweet
Tags: 1
Language: de
Mentions: 0
tweet place: Sweden 1 w/ @username) http://t.co/8c
Tschüss Madrid :) (@ Terminal (Stockholm)
URLs:
Detected Languages:
Workday’s Tweet 1
Language: Tags: 0
es
Mentions: 1
English: 43
tweet place: Spain
German: 23
URLs: 1
Other: 34
Tags: 0
Mentions: 1

Twitter User Profile Information
Location: Germany (Berlin)
Language: German
Experimental Setup
1. Select Users
2. Collect Tweets
3. Create user profiles
4. Create a classifier
5. Evaluate performance
12
Select Users

Twitter
API

Retrieve
Users(CURL)

MySQL

13
Crawling & Data Processing
Twitter
API

Retrieve Streams
(CURL)

Performance
Report
Tests
(Matlab)

Store JSON
(java)

MySQL
(Tweets)

MySQL
(User Profiles)

Select and Store
Features (java)
Country Total Number Users Posted 100
Of Users
or More Tweets
Japan

4885

2984

Spain

4906

3119

Brazil

4910

2935

USA

1714

1316

Germany 2823

1644

1 199 800 tweets
Microblogging Patterns
Cool, factual,
planners

Hashtags
URLs
Mobility
Networking
LINEARACTIVE
(Germany,
USA)

MULTIACTIVE
(Brazil, Spain)

Courteous,
accommodating,
listeners

Weekends
Replies

Warm, emotional,
loquacious

Mentions
Retweets
Languages

REACTIVE
(Japan)

16
8
Germany
Japan
Spain
USA
Brazil

6

DE
JP
ES
US
BR

c1

4

2

0

−2

−4

−8

−6

−4

−2

0
c2

2

4

6

17
Linearactive
Reactive

Reactiv
e
2.5 (C)

MultiLinear
active
Reactive
1.1 (B)

Multi

4.1 (A)

B
A
C

18
Classification Models
1

Language Codes

Number of

LANG

DEF

3

DEF+LANG
19

• URLs
• Hashtags
• Automatically-detected
Languages
• Geo-locations Detected
• Posts on Weekends
• Friends
• Followers
• User Mentions
• Retweets
• Replies

2
Decision Tree (LANG Feature)
Language Code
>= 4.5
< 4.5
JP
>= 3.5
< 3.5
>= 2.5

< 2.5
< 1.5
< 0.5

>= 1.5
>= 0.5

BR
DE

DE

BR

ES

Language
Japanese
Spanish
Portuguese
German
English
Other

Code
5
4
3
2
1
0
20
Languages in User Profiles

Languages: Native, English, Other
21
Decision Tree (DEF Features)
Languages
>= 0.5
< 0.5
Tags

Tags
< 6.5
JP

<54.5
JP

< 13.5

>= 6.5

Mentions
>= 54.5
< 69.5
URLs
>= 26.5 BR
< 26.5
ES

>= 13.5
Mentions
>= 69.5
ES

US

22
Classification Results
Country-level
Model
1
2
3

Features
LANG
DEF
L.+D.

Resubst.Err.
0.22
0.17
0.02

Cross-valid. Err.
0.22
0.42
0.06

Culture-level
Model
1
2
3

Features
LANG
DEF
L.+D.

Resubst.Err.
0.17
0.10
0.01

Cross-valid. Err.
0.17
0.29
0.04
23
Lang. Code
Lang. Code
Languages

< 2.5
< 0.5

Linear-active
Tags
< 8.5

>= 0.5
Lang. Code

< 1.5
Languages

Multi-active

Reactive

Multi-active

>= 1.5

>=8.5

Linear-active
Multi-active
Replies

< 38.5
Tags

>=48.5
Tags
< 15

Reactive

>= 2.5

>= 1.5

< 1.5

Replies
< 48.5

>= 4.5

< 4.5

>=15
Multi-active

< 36.5

>=38.5

>=36.5
Mentions

Multi-active

< 84.5
Weekends

< 20.5
Multi-active

>=20.5

>=84.5
Multi-active

Linear-active

Linear-active
Key Findings (Cultural Groups)
• Linear-active Users prefer sharing URLs and Hashtags,
and have larger social networks.
• Reactive users do not share so many Hashtags, they,
however, tend to Reply more than Multi-active users.
They employ the least of foreign languages, have lowest
tweeting mobility and tweet mostly on Weekends.
• Multi-active users generally employ more foreign
languages in their content.
25
Key Findings (Country Groups)
• German users share the most of Hashtags and tend to
reply;
• Users from the USA share the most of URLs, have
largest social networks than others and tweeting
mobility;
• Spanish users tend to retweet and mention other users;
• Brazilian users reply the least;
• Users from Japan tweet the most on weekends and
share the least of hashtags and user mentions, employ
the least of foreign languages and have lowest tweeting
26
mobility.
Adaptation Options
When appropriate, creating adaptive apps such as ecommerce or social network web sites to fit user
preferences for:
• sharing content;
• employing foreign languages;
• changing locality;
• communicating with other users.

27
Further Work
• Employ larger data set;
• Include more countries and add features;
• Extend our platform for other social networking
web sites;
• Recommending products/content in accord to user
cultural origings

28
Conclusions
Culture-oriented User Modeling
• Found microblogging patterns for cultural groups
• Employed them for identifying cultural origins
• Got insights on culture-oriented user modeling and
adaptation

29
Thank You
Full-text

Elena Daehnhardt

Google Scholar

Supplementary
Material

Elena@Ilina.nl
www.daehnhardt.com

30

Más contenido relacionado

Similar a A User Modeling Oriented Analysis of Cultural Backgrounds in Microblogging

Week 6 - Interactive News Editing and Producing
Week 6 - Interactive News Editing and ProducingWeek 6 - Interactive News Editing and Producing
Week 6 - Interactive News Editing and Producingkurtgessler
 
Social media strategy for activists - #WomensSummit2018 2018.06.20
Social media strategy for activists - #WomensSummit2018 2018.06.20Social media strategy for activists - #WomensSummit2018 2018.06.20
Social media strategy for activists - #WomensSummit2018 2018.06.20Alan Rosenblatt
 
UCLA X469.21 - FALL '15 WEEK 3
UCLA X469.21 - FALL '15 WEEK 3UCLA X469.21 - FALL '15 WEEK 3
UCLA X469.21 - FALL '15 WEEK 3SocialMediaUCLA
 
How donuts can help you communicate in an online world
How donuts can help you communicate in an online world How donuts can help you communicate in an online world
How donuts can help you communicate in an online world George Hulbert
 
Social Media for Learning: A Balanced Approach
Social Media for Learning: A Balanced ApproachSocial Media for Learning: A Balanced Approach
Social Media for Learning: A Balanced ApproachQuickLessons LLC
 
Managing content to enhance member value
Managing content to enhance member valueManaging content to enhance member value
Managing content to enhance member valueSteve Drake
 
AMS Training Social Media Presentation
AMS Training Social Media PresentationAMS Training Social Media Presentation
AMS Training Social Media PresentationAnnalisa Boccia
 
Smwpoland day2 final -ws7
Smwpoland day2 final -ws7Smwpoland day2 final -ws7
Smwpoland day2 final -ws7James Hutson
 
SMW Poland Day 2
SMW Poland Day 2SMW Poland Day 2
SMW Poland Day 2Tom Dixon
 
Emails as Social Media: How and Why Non-Profits and Educational Orgs Should U...
Emails as Social Media: How and Why Non-Profits and Educational Orgs Should U...Emails as Social Media: How and Why Non-Profits and Educational Orgs Should U...
Emails as Social Media: How and Why Non-Profits and Educational Orgs Should U...Dan Jones
 
Introductorysocialmedia
IntroductorysocialmediaIntroductorysocialmedia
Introductorysocialmediarobweaver
 
Social media management Oct 2016
Social media management   Oct 2016Social media management   Oct 2016
Social media management Oct 2016DigiArabs
 
Evaluating yoursocialmediaprogram
Evaluating yoursocialmediaprogramEvaluating yoursocialmediaprogram
Evaluating yoursocialmediaprogramErin Flior
 
Architecting Your Global Digital Experience House - Nicole Uhlig and Derek Pa...
Architecting Your Global Digital Experience House - Nicole Uhlig and Derek Pa...Architecting Your Global Digital Experience House - Nicole Uhlig and Derek Pa...
Architecting Your Global Digital Experience House - Nicole Uhlig and Derek Pa...SDL
 
Social Media in 30 Minutes a Day
Social Media in 30 Minutes a DaySocial Media in 30 Minutes a Day
Social Media in 30 Minutes a DayAmy Sample Ward
 
Using Social Media to Promote Your Research (Translate MedTech edition)
Using Social Media to Promote Your Research (Translate MedTech edition)Using Social Media to Promote Your Research (Translate MedTech edition)
Using Social Media to Promote Your Research (Translate MedTech edition)Kirsten Thompson
 
Social media architecture
Social media architectureSocial media architecture
Social media architectureDean Da Costa
 

Similar a A User Modeling Oriented Analysis of Cultural Backgrounds in Microblogging (20)

Week 6 - Interactive News Editing and Producing
Week 6 - Interactive News Editing and ProducingWeek 6 - Interactive News Editing and Producing
Week 6 - Interactive News Editing and Producing
 
Social media strategy for activists - #WomensSummit2018 2018.06.20
Social media strategy for activists - #WomensSummit2018 2018.06.20Social media strategy for activists - #WomensSummit2018 2018.06.20
Social media strategy for activists - #WomensSummit2018 2018.06.20
 
UCLA X469.21 - FALL '15 WEEK 3
UCLA X469.21 - FALL '15 WEEK 3UCLA X469.21 - FALL '15 WEEK 3
UCLA X469.21 - FALL '15 WEEK 3
 
How donuts can help you communicate in an online world
How donuts can help you communicate in an online world How donuts can help you communicate in an online world
How donuts can help you communicate in an online world
 
Social Media for Learning: A Balanced Approach
Social Media for Learning: A Balanced ApproachSocial Media for Learning: A Balanced Approach
Social Media for Learning: A Balanced Approach
 
Managing content to enhance member value
Managing content to enhance member valueManaging content to enhance member value
Managing content to enhance member value
 
AMS Training Social Media Presentation
AMS Training Social Media PresentationAMS Training Social Media Presentation
AMS Training Social Media Presentation
 
Smwpoland day2 final -ws7
Smwpoland day2 final -ws7Smwpoland day2 final -ws7
Smwpoland day2 final -ws7
 
SMW Poland Day 2
SMW Poland Day 2SMW Poland Day 2
SMW Poland Day 2
 
Emails as Social Media: How and Why Non-Profits and Educational Orgs Should U...
Emails as Social Media: How and Why Non-Profits and Educational Orgs Should U...Emails as Social Media: How and Why Non-Profits and Educational Orgs Should U...
Emails as Social Media: How and Why Non-Profits and Educational Orgs Should U...
 
Social Media Training
Social Media TrainingSocial Media Training
Social Media Training
 
Introductorysocialmedia
IntroductorysocialmediaIntroductorysocialmedia
Introductorysocialmedia
 
Social media management Oct 2016
Social media management   Oct 2016Social media management   Oct 2016
Social media management Oct 2016
 
Smwpoland day2 final -ws5
Smwpoland day2 final -ws5Smwpoland day2 final -ws5
Smwpoland day2 final -ws5
 
Evaluating yoursocialmediaprogram
Evaluating yoursocialmediaprogramEvaluating yoursocialmediaprogram
Evaluating yoursocialmediaprogram
 
Management de communaute
Management de communauteManagement de communaute
Management de communaute
 
Architecting Your Global Digital Experience House - Nicole Uhlig and Derek Pa...
Architecting Your Global Digital Experience House - Nicole Uhlig and Derek Pa...Architecting Your Global Digital Experience House - Nicole Uhlig and Derek Pa...
Architecting Your Global Digital Experience House - Nicole Uhlig and Derek Pa...
 
Social Media in 30 Minutes a Day
Social Media in 30 Minutes a DaySocial Media in 30 Minutes a Day
Social Media in 30 Minutes a Day
 
Using Social Media to Promote Your Research (Translate MedTech edition)
Using Social Media to Promote Your Research (Translate MedTech edition)Using Social Media to Promote Your Research (Translate MedTech edition)
Using Social Media to Promote Your Research (Translate MedTech edition)
 
Social media architecture
Social media architectureSocial media architecture
Social media architecture
 

Último

Navi Mumbai Call Girls 🥰 8617370543 Service Offer VIP Hot Model
Navi Mumbai Call Girls 🥰 8617370543 Service Offer VIP Hot ModelNavi Mumbai Call Girls 🥰 8617370543 Service Offer VIP Hot Model
Navi Mumbai Call Girls 🥰 8617370543 Service Offer VIP Hot ModelDeepika Singh
 
Polkadot JAM Slides - Token2049 - By Dr. Gavin Wood
Polkadot JAM Slides - Token2049 - By Dr. Gavin WoodPolkadot JAM Slides - Token2049 - By Dr. Gavin Wood
Polkadot JAM Slides - Token2049 - By Dr. Gavin WoodJuan lago vázquez
 
presentation ICT roal in 21st century education
presentation ICT roal in 21st century educationpresentation ICT roal in 21st century education
presentation ICT roal in 21st century educationjfdjdjcjdnsjd
 
Strategies for Landing an Oracle DBA Job as a Fresher
Strategies for Landing an Oracle DBA Job as a FresherStrategies for Landing an Oracle DBA Job as a Fresher
Strategies for Landing an Oracle DBA Job as a FresherRemote DBA Services
 
TrustArc Webinar - Stay Ahead of US State Data Privacy Law Developments
TrustArc Webinar - Stay Ahead of US State Data Privacy Law DevelopmentsTrustArc Webinar - Stay Ahead of US State Data Privacy Law Developments
TrustArc Webinar - Stay Ahead of US State Data Privacy Law DevelopmentsTrustArc
 
Apidays New York 2024 - The Good, the Bad and the Governed by David O'Neill, ...
Apidays New York 2024 - The Good, the Bad and the Governed by David O'Neill, ...Apidays New York 2024 - The Good, the Bad and the Governed by David O'Neill, ...
Apidays New York 2024 - The Good, the Bad and the Governed by David O'Neill, ...apidays
 
Apidays New York 2024 - Scaling API-first by Ian Reasor and Radu Cotescu, Adobe
Apidays New York 2024 - Scaling API-first by Ian Reasor and Radu Cotescu, AdobeApidays New York 2024 - Scaling API-first by Ian Reasor and Radu Cotescu, Adobe
Apidays New York 2024 - Scaling API-first by Ian Reasor and Radu Cotescu, Adobeapidays
 
Axa Assurance Maroc - Insurer Innovation Award 2024
Axa Assurance Maroc - Insurer Innovation Award 2024Axa Assurance Maroc - Insurer Innovation Award 2024
Axa Assurance Maroc - Insurer Innovation Award 2024The Digital Insurer
 
"I see eyes in my soup": How Delivery Hero implemented the safety system for ...
"I see eyes in my soup": How Delivery Hero implemented the safety system for ..."I see eyes in my soup": How Delivery Hero implemented the safety system for ...
"I see eyes in my soup": How Delivery Hero implemented the safety system for ...Zilliz
 
Boost Fertility New Invention Ups Success Rates.pdf
Boost Fertility New Invention Ups Success Rates.pdfBoost Fertility New Invention Ups Success Rates.pdf
Boost Fertility New Invention Ups Success Rates.pdfsudhanshuwaghmare1
 
Powerful Google developer tools for immediate impact! (2023-24 C)
Powerful Google developer tools for immediate impact! (2023-24 C)Powerful Google developer tools for immediate impact! (2023-24 C)
Powerful Google developer tools for immediate impact! (2023-24 C)wesley chun
 
MINDCTI Revenue Release Quarter One 2024
MINDCTI Revenue Release Quarter One 2024MINDCTI Revenue Release Quarter One 2024
MINDCTI Revenue Release Quarter One 2024MIND CTI
 
ICT role in 21st century education and its challenges
ICT role in 21st century education and its challengesICT role in 21st century education and its challenges
ICT role in 21st century education and its challengesrafiqahmad00786416
 
AWS Community Day CPH - Three problems of Terraform
AWS Community Day CPH - Three problems of TerraformAWS Community Day CPH - Three problems of Terraform
AWS Community Day CPH - Three problems of TerraformAndrey Devyatkin
 
Apidays New York 2024 - Accelerating FinTech Innovation by Vasa Krishnan, Fin...
Apidays New York 2024 - Accelerating FinTech Innovation by Vasa Krishnan, Fin...Apidays New York 2024 - Accelerating FinTech Innovation by Vasa Krishnan, Fin...
Apidays New York 2024 - Accelerating FinTech Innovation by Vasa Krishnan, Fin...apidays
 
Strategies for Unlocking Knowledge Management in Microsoft 365 in the Copilot...
Strategies for Unlocking Knowledge Management in Microsoft 365 in the Copilot...Strategies for Unlocking Knowledge Management in Microsoft 365 in the Copilot...
Strategies for Unlocking Knowledge Management in Microsoft 365 in the Copilot...Drew Madelung
 
Apidays Singapore 2024 - Building Digital Trust in a Digital Economy by Veron...
Apidays Singapore 2024 - Building Digital Trust in a Digital Economy by Veron...Apidays Singapore 2024 - Building Digital Trust in a Digital Economy by Veron...
Apidays Singapore 2024 - Building Digital Trust in a Digital Economy by Veron...apidays
 
Manulife - Insurer Transformation Award 2024
Manulife - Insurer Transformation Award 2024Manulife - Insurer Transformation Award 2024
Manulife - Insurer Transformation Award 2024The Digital Insurer
 
Emergent Methods: Multi-lingual narrative tracking in the news - real-time ex...
Emergent Methods: Multi-lingual narrative tracking in the news - real-time ex...Emergent Methods: Multi-lingual narrative tracking in the news - real-time ex...
Emergent Methods: Multi-lingual narrative tracking in the news - real-time ex...Zilliz
 
Web Form Automation for Bonterra Impact Management (fka Social Solutions Apri...
Web Form Automation for Bonterra Impact Management (fka Social Solutions Apri...Web Form Automation for Bonterra Impact Management (fka Social Solutions Apri...
Web Form Automation for Bonterra Impact Management (fka Social Solutions Apri...Jeffrey Haguewood
 

Último (20)

Navi Mumbai Call Girls 🥰 8617370543 Service Offer VIP Hot Model
Navi Mumbai Call Girls 🥰 8617370543 Service Offer VIP Hot ModelNavi Mumbai Call Girls 🥰 8617370543 Service Offer VIP Hot Model
Navi Mumbai Call Girls 🥰 8617370543 Service Offer VIP Hot Model
 
Polkadot JAM Slides - Token2049 - By Dr. Gavin Wood
Polkadot JAM Slides - Token2049 - By Dr. Gavin WoodPolkadot JAM Slides - Token2049 - By Dr. Gavin Wood
Polkadot JAM Slides - Token2049 - By Dr. Gavin Wood
 
presentation ICT roal in 21st century education
presentation ICT roal in 21st century educationpresentation ICT roal in 21st century education
presentation ICT roal in 21st century education
 
Strategies for Landing an Oracle DBA Job as a Fresher
Strategies for Landing an Oracle DBA Job as a FresherStrategies for Landing an Oracle DBA Job as a Fresher
Strategies for Landing an Oracle DBA Job as a Fresher
 
TrustArc Webinar - Stay Ahead of US State Data Privacy Law Developments
TrustArc Webinar - Stay Ahead of US State Data Privacy Law DevelopmentsTrustArc Webinar - Stay Ahead of US State Data Privacy Law Developments
TrustArc Webinar - Stay Ahead of US State Data Privacy Law Developments
 
Apidays New York 2024 - The Good, the Bad and the Governed by David O'Neill, ...
Apidays New York 2024 - The Good, the Bad and the Governed by David O'Neill, ...Apidays New York 2024 - The Good, the Bad and the Governed by David O'Neill, ...
Apidays New York 2024 - The Good, the Bad and the Governed by David O'Neill, ...
 
Apidays New York 2024 - Scaling API-first by Ian Reasor and Radu Cotescu, Adobe
Apidays New York 2024 - Scaling API-first by Ian Reasor and Radu Cotescu, AdobeApidays New York 2024 - Scaling API-first by Ian Reasor and Radu Cotescu, Adobe
Apidays New York 2024 - Scaling API-first by Ian Reasor and Radu Cotescu, Adobe
 
Axa Assurance Maroc - Insurer Innovation Award 2024
Axa Assurance Maroc - Insurer Innovation Award 2024Axa Assurance Maroc - Insurer Innovation Award 2024
Axa Assurance Maroc - Insurer Innovation Award 2024
 
"I see eyes in my soup": How Delivery Hero implemented the safety system for ...
"I see eyes in my soup": How Delivery Hero implemented the safety system for ..."I see eyes in my soup": How Delivery Hero implemented the safety system for ...
"I see eyes in my soup": How Delivery Hero implemented the safety system for ...
 
Boost Fertility New Invention Ups Success Rates.pdf
Boost Fertility New Invention Ups Success Rates.pdfBoost Fertility New Invention Ups Success Rates.pdf
Boost Fertility New Invention Ups Success Rates.pdf
 
Powerful Google developer tools for immediate impact! (2023-24 C)
Powerful Google developer tools for immediate impact! (2023-24 C)Powerful Google developer tools for immediate impact! (2023-24 C)
Powerful Google developer tools for immediate impact! (2023-24 C)
 
MINDCTI Revenue Release Quarter One 2024
MINDCTI Revenue Release Quarter One 2024MINDCTI Revenue Release Quarter One 2024
MINDCTI Revenue Release Quarter One 2024
 
ICT role in 21st century education and its challenges
ICT role in 21st century education and its challengesICT role in 21st century education and its challenges
ICT role in 21st century education and its challenges
 
AWS Community Day CPH - Three problems of Terraform
AWS Community Day CPH - Three problems of TerraformAWS Community Day CPH - Three problems of Terraform
AWS Community Day CPH - Three problems of Terraform
 
Apidays New York 2024 - Accelerating FinTech Innovation by Vasa Krishnan, Fin...
Apidays New York 2024 - Accelerating FinTech Innovation by Vasa Krishnan, Fin...Apidays New York 2024 - Accelerating FinTech Innovation by Vasa Krishnan, Fin...
Apidays New York 2024 - Accelerating FinTech Innovation by Vasa Krishnan, Fin...
 
Strategies for Unlocking Knowledge Management in Microsoft 365 in the Copilot...
Strategies for Unlocking Knowledge Management in Microsoft 365 in the Copilot...Strategies for Unlocking Knowledge Management in Microsoft 365 in the Copilot...
Strategies for Unlocking Knowledge Management in Microsoft 365 in the Copilot...
 
Apidays Singapore 2024 - Building Digital Trust in a Digital Economy by Veron...
Apidays Singapore 2024 - Building Digital Trust in a Digital Economy by Veron...Apidays Singapore 2024 - Building Digital Trust in a Digital Economy by Veron...
Apidays Singapore 2024 - Building Digital Trust in a Digital Economy by Veron...
 
Manulife - Insurer Transformation Award 2024
Manulife - Insurer Transformation Award 2024Manulife - Insurer Transformation Award 2024
Manulife - Insurer Transformation Award 2024
 
Emergent Methods: Multi-lingual narrative tracking in the news - real-time ex...
Emergent Methods: Multi-lingual narrative tracking in the news - real-time ex...Emergent Methods: Multi-lingual narrative tracking in the news - real-time ex...
Emergent Methods: Multi-lingual narrative tracking in the news - real-time ex...
 
Web Form Automation for Bonterra Impact Management (fka Social Solutions Apri...
Web Form Automation for Bonterra Impact Management (fka Social Solutions Apri...Web Form Automation for Bonterra Impact Management (fka Social Solutions Apri...
Web Form Automation for Bonterra Impact Management (fka Social Solutions Apri...
 

A User Modeling Oriented Analysis of Cultural Backgrounds in Microblogging

Notas del editor

  1. @article{ilina2012user, title={A User Modeling Oriented Analysis of Cultural Backgrounds in Microblogging}, author={Ilina, Elena}, journal={HUMAN JOURNAL}, volume={1}, number={4}, pages={166--181}, year={2012} } Full-text is at: http://ojs.scienceengineering.org/index.php/human/article/view/43/18
  2. On this slide you see the outline of my presentation.The main question I was concerned with washow different cultural groups could be identified on microblogs like Twitter.In this context, we can ask ourselves, why is it important to understand user cultural backgrounds?How could such knowledge be exploited for adaptive applications?For addressing these questions, I used a sociological study called the Lewis Model of Cultures.This model describes cultural communication differences, which I also tried to find in Twitter micro-blogs.
  3. The Lewis model is represented as a triangle where apexes show extreme cultural dimensions.The cultural dimensions are linked with countries of origin and explain persons’ communication attitudes.Multi-active people from Hispanic America and Brazil focus on dialogs with other people and generally display their feelings.In opposite, Reactive people from Vietnam tend to conceal their feelings, they are generally very polite and good listeners.Linear-active people from Germany and Switzerland are generally great organizers and focus on planning activities.All countries in between these extremes have mixture of cultural dimensions.Note that this is a general model – some users might be “outliers” and will not fit in their “county-stereotypes”.
  4. http://www.google.nl/imgres?imgurl=http://www.crossculture.com/UserFiles/Image/LMR-table-new.gif&amp;imgrefurl=http://senseinzanzibar.wordpress.com/page/3/&amp;usg=__DT4TjH1NW71F5RrJycaRiY1260I=&amp;h=379&amp;w=500&amp;sz=25&amp;hl=en&amp;start=0&amp;sig2=pYvsvRCCvLYx4bv5iFL6Pw&amp;zoom=1&amp;tbnid=TMMQEAt9p-zxQM:&amp;tbnh=96&amp;tbnw=126&amp;ei=WEI-UIXMD-bP0QWi-IHICQ&amp;itbs=1&amp;iact=hc&amp;vpx=74&amp;vpy=106&amp;dur=44&amp;hovh=196&amp;hovw=258&amp;tx=119&amp;ty=150&amp;sig=115419424352996972569&amp;page=1&amp;ndsp=2&amp;ved=1t:429,r:0,s:0,i:52In The Lewis model, cultural dimensions are associated with personality traits.I found this idea appealing to me.I have met a number of talkative multi-active persons and am fascinated by their ability to talk eloquently and using gestures a lot.However, for linear-active people encounters with such multi-active persons can be sometimes overwhelming. Also, impatience and concealed feelings of linear-active persons can be perceived as offensive and cold to multi-active persons.Lewis explains that most of us are multi-active, but businesses mostly address linear-active customers. This can lead to lost sales.We could employ different strategies for targeting persons from different cultures.When a customer is multi-active, we could focus on more emotional side of the product rather than just listing factual characteristics appealing more to linear-active persons.
  5. E-commerce web sites like Amazon can benefit from customised product descriptions and behaviour targeted advertisements.User location, languages used and previous purchases can be collected from history logs.When user is new to the system, such information is not available.As a solution, user preferences could also be collected from the social web and micro-blogs.Targeted ads are already created using data on user preferences collected from social networking sites.For instance, one company (http://triggit.com/) uses Facebook for placing their advertisements and collecting user data from a web statistics service.
  6. Deleted {At this point, you might ask yourself:Why would there be a need to identify user cultural origins? And how could this be exploited in information and web systems?} Information on the cultural origin of a user is important for improving the user experience in adaptive applications, requiring for instance Location of an userA functionality or particular Design preferences.Often, such information is not available, for example because a user is new to the system or has not given details.In this case, user characteristics and preferences can be mined from the social web.From the Twitter micro-blogs employed in my experiments, I could find out amongst others User locationsLanguages usedSocial connections. Further, as we will see on next slides, our experiments has shown that users from different cultural groups blog differently.Cultural microblogging patterns could then be employed to build culture-oriented user profiles.
  7. Now, the question arises on how we could learn about user traits based on the cultural background of a user?For this, I propose to derive behavioral patterns by mining the microblogging activities of users.With known microblogging patterns, I then identify a user as belonging to a particular cultural group.This allowed me to create user profiles with preferences information on culture-specific user traits, to be used in the adaptation process.
  8. I assumed that Twitter data can be used for finding cultural patterns in microblogging behavior.Indeed, with millions of active users and easy to access open profiles, Twitter is an excellent platform to study microblogging behavior.The features I have analyzed include amongst others usage of URLshashtagsuser mentions languages employedThey were grouped into the following feature groups: Content-based featuresActivity-based featuresSocial Network-based features Conversation-based features These features were defined as indicators for cultural patterns.It is important to mention that only the Twitter features usage for the selected user groups was analyzed, no private information was stored or shared as result of experiments.
  9. select userid where languages&gt;2 and countries&gt;2 and origin like &quot;germany&quot; limit 5;select test,urls,tags,languages,countries,weekend,users,friends,followers,userRetweet,userReply from features where userid=&quot;452248921” and test=5; (test=5)select id,username,origin,language,location,timezone from User_culture where id=&quot;452248921&quot;;Consider the following example for feature selection:Take a user from Berlin, Germany, having as preferred language German in his profile.In randomly selected 100 of his tweets we found 49 and 4 URLs and hashtags respectively.In his content, we automatically detected two languages, English and German.Each tweet’s meta-data can contain also geo-coordinates or the place of the tweeting.We identified 7 different countries in the meta-data of 100 tweets.The user also tweeted most of the time on workdays.He had 50 friends and 96 followers. In his 100 tweets we found 75 mentions of other users.Out of 100 posts, he had 1 retweet and 28 replies to other users.
  10. select count(language),language from languages where userid=&quot;452248921&quot; group by language;select tweets_culture.id,tweets_culture.content,languages.language from tweets_culture,languages where tweets_culture.id=languages.id order by rand() limit 5;Mode=1112Here we see three tweets posted by our user, in English and German.Since tweets are short, informal and include hashtags or URLs, automatic language detection is challenging.For detecting languages, a threshold was used, disregarding a specific language when lower than 5.Similarly, we deal with locations identified with the help of geonames web service and Google Maps. A location is disregarded when it is detected less than 5 times out of the 100 tweets for a user.
  11. I will now come to the experimentation setup, data collection process and present the results.In the experiments I try to find microblogging behavioral differences for people from different cultural origins. Next, found microblogging patterns are used to classify users in their respective user groups.The experimental setup consists of the five main steps:Users whose tweets originate from respective geographic locations were selected.Next, their tweets were collectedAnd, user profiles based on the meta-data and content of the tweets were createdTo classify users into their respective cultural user groups I used decision treesAs a final step, I have assessed the classification performance.
  12. This slide shows the initially defined geo-coordinated for five selected countries: Japan, United States, Brazil, Spain and Germany.We used these coordinates to identify users tweeting around the selected places.And, also users tweeting from intermediate locations defined by the bounding geo-coordinates box.We employed Streaming Twitter API to retrieve Users for the 5 countries analyzed. After running the CURL for several days, we had a list of users and their preferred languages defined in the Twitter profile.The information was further stored in the MySQL database.Matlab mode=1111https://maps.google.com/maps/ms?ie=UTF8&amp;hl=en&amp;t=h&amp;oe=UTF8&amp;msa=0&amp;msid=103387750622659154041.0004537732c8262c4ae94crawling.kmlselect location, count(userid) as n from features_final3 where origin=&apos;usa&apos; group by location;South San Francisco
  13. On this slide we see the simplified Lewis Model of Cultures.Lewis explains, thatlinear-active persons are cool and factual plannersmulti-active are generally warm, emotional and loquaciousreactive persons have different time perceptions, and generally are dialog and people-centeredBased on this, I established the following hypothesises:linear-active prefer using hashtags and URLs to organize their Tweeting posts and share factual information with other users (SUPPORTED)multi-active persons might have greater social networks (NOT SUPPORTED), retweet and reply the most (SUPPORTED), and might employ more foreign languages in their posts compared to other users (SUPPORTED)reactive persons might tweet mostly on weekend and reply a lot (SUPPORTED)Further results showed thatlinear-active persons have larger contacts network on Twitter and have greater tweeting mobility. They tweet from different locations the most.This finding could be explained with the use of Twitter for business purposes (advertisement) by Americans and Germans (which could be further investigated) Overall, a different behavior between the analyzed user groups was found.
  14. The next two slides present clusters of user groups by countries and cultural groups. They show the differences between user groups in the respect of the features set which includes hashtags, user mentions, URLs and other aforementioned elements.Based on Multivariate Analysis of variance, two variables are used to distinguish between user groups. These variables are calculated from the means of features analyzed.Here we can see for instance, that the German user group considerably overlaps with other user groups. This appears to reflect the behavioral similarities between the German user groups and others.In contrast, Japan is depicted at the lower portion, showing that users from Japan appear to behave quite differently from other user groups.
  15. This slide shows the clusters for the user groups according to the Lewis model. On the culture-level, variable c1 helps to separate the reactive user group depicted in the red cluster from the other two clusters, multi-active users and linear-active users. Considering the analyzed features set, this indicates that reactive users from Japan behave indeed differently on Twitter. The distance between group means for Reactive and two other user groups is greater than between linear-active and multi-active user groups.Having determined the distance between the means of the user groups, I was intrigued if this feature set could be used to predict users’ belonging to a respective user group.
  16. For this, I created three main classification models.The first model used language codes to predict users’ groupsThe second one (Default) was used to classify users according to their usage of the features listed on the rightThe third model is a combination of the previous two models.
  17. The first decision tree is based on the languages defined in the Twitter user profile.A language code is assigned to each language (as shown on the right). In case a language does not match one of these five, the code value is zero.Interestingly we see that the lower two branches define Brazilian and German user groups, instead of users from the US.Why does this happen? And what does this imply?
  18. As it appears, many users from Brazil and Germany defined English as their preferred language in the Twitter User Profile.The decision tree trained on languages defined in the user profile does not give good prediction results.And, not surprisingly, about a fifth of users were not classified correctly to their respective country group.If we would assume that users who defined English in their user profiles were from the US, we would misclassify large fractions of Brazilian and German users.
  19. I tried to solve this issue by introducing Twitter-specific features.Knowingly, users from a defined country and cultural group employ Twitter features differently.For instance, Japanese users refrain from tagging and mentioning other users, while users from the US use URLs the most.The experiments showed that a decision tree for predicting user groups based on the ten features mentioned before can be created.
  20. We see that the testing error improved slightly compared to the language code feature of the previous test. However, the cross validation error increased two-fold.The decision tree based on the Default features overfits the training set. It seems that the tree structure is sensitive to the training set.Since I would like to have a better performance on the new data set, I combined the aforementioned feature set to get a new decision tree based on the DEF features and language codes.This helped in further decreasing the testing and cross-validation errors.
  21. To achieve the smallest cross-validation error, the decision tree can be cut back (pruned) to the number of nodes.Having 14 terminal nodes, it is possible to achieve more than 90% accuracy (96).As can be seen from the decision tree shown on this slide, the Number of hashtags, Replies, User mentions, Posts published on weekends, and languages identified in user content are useful features next to the language code defined in the user profile.With these features I can associate users to their respective cultural group of the Lewis model.
  22. This slide shows the key findings on the Lewis Cultural Dimensions for the users analyzed.Linear-active users from the US and Germany tend to share URLs and hashtagsAnd have larger social networks in average.This finding is important to keep in mind when considering building social networking applications which enable sharing content and creating social networks.For instance, German users could be provided with easy access to a hashtag sharing functionality.Since Japanese users employ the least of foreign languages and have the lowest tweeting mobility, locality features could be less noticeable in an user interface.In contrast, multi-active users could be provided with more flexible locality functionality to assist in languages selection.
  23. The same, microblogging behavior patterns could also be exploited on a country group level.Since US and German users tend to share URLs and Hashtags, it could be further investigated if sharing hashtags and URLs is used for organizing content or promotional purposes. Already the finding that linear-active people have also the largest social networks appears to support this assumption.It is well known that Twitter and other social networks are widely used for marketing and other business purposes.This is why the implications of different cultural behaviors is paramount for globally-oriented businesses.
  24. It could be suggested to employ social networking patterns in order to find out user preferences towards sharing content such as hashtags and URLs, or localization and conversation needs.This could allow suitable adaptation strategies for applications such as e-commerce dealing with customers from different cultural backgrounds
  25. In a nutshell, I described an approach to mine cultural patterns from microblogs.Ten features on Twitter were analyzed to understand user attitudes in sharing content, social networking preferences and activities.With this, interesting microblogging patterns for more than 10 000 users from five selected countries were found.This enabled me to predict user origins on country level and the user’s cultural dimension in the Lewis model.In our future work, I would like to employ a larger user dataset and include more countries in the analysis.Adding more features and more social networks into the analysis might be very useful to gain a better understanding of cultural differences of a user behavior online.The next target would be to recommend products or content in accord to user cultural origins
  26. The main aim of the study was to employ user microblogging activities for creating culture-oriented user profiles.We found microblogging patterns for the selected user groups.I then employed cultural microblogging patterns for identifying cultural origins of users.These cultural differences can be taken into account when implementing adaptive applications or social networking web sites.Finally, I provide insights on culture-oriented modeling and further adaptation.
  27. Thank you very much.I will be happy to answer your questions.