inferring topical attributes of users in the twitter...

36
Inferring Topical Attributes of Users in the Twitter Social Network Saptarshi Ghosh Post-doctoral researcher Max Planck Institute for Software Systems, Germany socialnetworks.mpi-sws.org Dynamics On and Of Complex Networks 2014

Upload: others

Post on 25-Sep-2020

1 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Inferring Topical Attributes of Users in the Twitter ...perso.uclouvain.be/jean-charles.delvenne/index... · Inferring Topics of Interest of a User! Prior attempts # Use NLP on tweets

Inferring Topical Attributes of Users in the Twitter Social Network

Saptarshi Ghosh Post-doctoral researcher Max Planck Institute for Software Systems, Germany socialnetworks.mpi-sws.org Dynamics On and Of Complex Networks 2014

Page 2: Inferring Topical Attributes of Users in the Twitter ...perso.uclouvain.be/jean-charles.delvenne/index... · Inferring Topics of Interest of a User! Prior attempts # Use NLP on tweets

Joint work with … MPI-SWS, Germany

n  Md. Bilal Zafar

n  Juhi Kulshrestha

n  Mainack Mondal

n  Krishna Gummadi

IIT Kharagpur, India

n  Parantapa Bhattacharya

n  Naveen Sharma (currently at Univ. Washington)

n  Niloy Ganguly

Page 3: Inferring Topical Attributes of Users in the Twitter ...perso.uclouvain.be/jean-charles.delvenne/index... · Inferring Topics of Interest of a User! Prior attempts # Use NLP on tweets

Profile of a Twitter user

Page 4: Inferring Topical Attributes of Users in the Twitter ...perso.uclouvain.be/jean-charles.delvenne/index... · Inferring Topics of Interest of a User! Prior attempts # Use NLP on tweets

Twitter social network n  u à v: user u follows user v

n  u subscribes to the tweets posted by v

n  u is a follower of v

n  v is a following of u

Page 5: Inferring Topical Attributes of Users in the Twitter ...perso.uclouvain.be/jean-charles.delvenne/index... · Inferring Topics of Interest of a User! Prior attempts # Use NLP on tweets

Twitter – an information network

n  “We’re not a social network, we’re an information network” – Michael Abbott, VP Engineering, Twitter

n  Twitter slogan q  What are you doing? à changed to à q  Discover what’s happening right now, anywhere in the world

n  Millions of users rely on Twitter to discover real-time content on various topics

Page 6: Inferring Topical Attributes of Users in the Twitter ...perso.uclouvain.be/jean-charles.delvenne/index... · Inferring Topics of Interest of a User! Prior attempts # Use NLP on tweets

Discovering information in Twitter n  Plethora of information

q  Over 600 million users q  Over 400 million tweets posted daily

n  Infeasible for an individual user to discover interesting information on her own

Page 7: Inferring Topical Attributes of Users in the Twitter ...perso.uclouvain.be/jean-charles.delvenne/index... · Inferring Topics of Interest of a User! Prior attempts # Use NLP on tweets

Our research

n  Objective: enable Twitter users find information relevant to their interests

n  How to locate interesting information on a topic?

q  Identify topical experts

n  Who should be provided with information on a topic? q  Identify users who are interested in the topic

Page 8: Inferring Topical Attributes of Users in the Twitter ...perso.uclouvain.be/jean-charles.delvenne/index... · Inferring Topics of Interest of a User! Prior attempts # Use NLP on tweets

Identifying topical experts Inferring topical interests of users Application: Interaction among experts and interested users

Page 9: Inferring Topical Attributes of Users in the Twitter ...perso.uclouvain.be/jean-charles.delvenne/index... · Inferring Topics of Interest of a User! Prior attempts # Use NLP on tweets

Features used by prior approaches

Page 10: Inferring Topical Attributes of Users in the Twitter ...perso.uclouvain.be/jean-charles.delvenne/index... · Inferring Topics of Interest of a User! Prior attempts # Use NLP on tweets

Our proposal …

n  Utilize social annotations to infer topics of expertise of popular Twitter users q  How does the Twitter crowd describe a user?

n  Social annotations obtained through Twitter Lists q  A feature by which one can create a named group

containing some group related users

Page 11: Inferring Topical Attributes of Users in the Twitter ...perso.uclouvain.be/jean-charles.delvenne/index... · Inferring Topics of Interest of a User! Prior attempts # Use NLP on tweets

How Lists work ?

Page 12: Inferring Topical Attributes of Users in the Twitter ...perso.uclouvain.be/jean-charles.delvenne/index... · Inferring Topics of Interest of a User! Prior attempts # Use NLP on tweets

Using Lists to infer topics for users

n  If user u is an expert / authority in a certain topic q  u is likely to be included in several Lists q  List names / descriptions provide valuable semantic cues

to the topics of expertise of u

Page 13: Inferring Topical Attributes of Users in the Twitter ...perso.uclouvain.be/jean-charles.delvenne/index... · Inferring Topics of Interest of a User! Prior attempts # Use NLP on tweets

Topics inferred from Lists

linux, tech, open, software, libre, gnu, computer, developer, ubuntu, unix

politics, senator, congress, government, republicans, Iowa, gop, conservative

celebs, actors, famous, movies, comedy, funny, music, hollywood, pop culture

Page 14: Inferring Topical Attributes of Users in the Twitter ...perso.uclouvain.be/jean-charles.delvenne/index... · Inferring Topics of Interest of a User! Prior attempts # Use NLP on tweets

Who-is-who service for Twitter http://twitter-app.mpi-sws.org/who-is-who/

Inferring Who-is-who in the Twitter Social Network ACM Workshop on Online Social Networks 2012, ACM Computer Communication Review 2012

Page 15: Inferring Topical Attributes of Users in the Twitter ...perso.uclouvain.be/jean-charles.delvenne/index... · Inferring Topics of Interest of a User! Prior attempts # Use NLP on tweets

Identify topical experts: methodology

n  Given a query (topic)

n  Identify experts on the topic using Lists

n  Rank identified experts w.r.t. given topic q  Relevance of expert to topic q  Popularity of expert

Page 16: Inferring Topical Attributes of Users in the Twitter ...perso.uclouvain.be/jean-charles.delvenne/index... · Inferring Topics of Interest of a User! Prior attempts # Use NLP on tweets

Cognos n  Search system for topical experts in Twitter

http://twitter-app.mpi-sws.org/whom-to-follow/

Cognos: Crowdsourcing Search for Topic Experts in Microblogs, ACM SIGIR Conference 2012

Page 17: Inferring Topical Attributes of Users in the Twitter ...perso.uclouvain.be/jean-charles.delvenne/index... · Inferring Topics of Interest of a User! Prior attempts # Use NLP on tweets

Cognos results for “politics”

Page 18: Inferring Topical Attributes of Users in the Twitter ...perso.uclouvain.be/jean-charles.delvenne/index... · Inferring Topics of Interest of a User! Prior attempts # Use NLP on tweets

Cognos results for “stem cell”

Page 19: Inferring Topical Attributes of Users in the Twitter ...perso.uclouvain.be/jean-charles.delvenne/index... · Inferring Topics of Interest of a User! Prior attempts # Use NLP on tweets

Evaluation of Cognos q  Cognos compared with Twitter Who To Follow

q  Evaluator shown top 10 results (anonymized) by both systems q  Evaluator judges which is better / both good / both bad q  Judgment by majority voting

q  Cognos results judged equally good or better for 60% topics

Page 20: Inferring Topical Attributes of Users in the Twitter ...perso.uclouvain.be/jean-charles.delvenne/index... · Inferring Topics of Interest of a User! Prior attempts # Use NLP on tweets

Identifying topical experts Inferring topical interests of users Application: Interaction among experts and interested users

Page 21: Inferring Topical Attributes of Users in the Twitter ...perso.uclouvain.be/jean-charles.delvenne/index... · Inferring Topics of Interest of a User! Prior attempts # Use NLP on tweets

Inferring Topics of Interest of a User n  Prior attempts

q  Use NLP on tweets posted by a user or received by a user q  Limitation: tweets often contain day-to-day conversation

n  Our approach

q  If a user is subscribing to several experts on a topic, she is very likely to be interested in that topic

q  Identify which topical experts a user follows, and infer the topics of expertise of those experts

Page 22: Inferring Topical Attributes of Users in the Twitter ...perso.uclouvain.be/jean-charles.delvenne/index... · Inferring Topics of Interest of a User! Prior attempts # Use NLP on tweets

Who likes what n  System to infer topical interests of users in Twitter

http://twitter-app.mpi-sws.org/who-likes-what/

Inferring User Interests in the Twitter Social Network, ACM Conference on Recommender Systems (RecSys) 2014

Page 23: Inferring Topical Attributes of Users in the Twitter ...perso.uclouvain.be/jean-charles.delvenne/index... · Inferring Topics of Interest of a User! Prior attempts # Use NLP on tweets

Topical interests of Lada Adamic

Page 24: Inferring Topical Attributes of Users in the Twitter ...perso.uclouvain.be/jean-charles.delvenne/index... · Inferring Topics of Interest of a User! Prior attempts # Use NLP on tweets

Topical interests of Duncan Watts

Page 25: Inferring Topical Attributes of Users in the Twitter ...perso.uclouvain.be/jean-charles.delvenne/index... · Inferring Topics of Interest of a User! Prior attempts # Use NLP on tweets

Evaluation of the approach n  Accuracy

q  Compared with topic modeling (Labeled LDA) on tweets posted / received by a user

q  Topics inferred by proposed approach judged significantly better by human volunteers

n  Scalability

q  Enables inference of interests of more than 77% of all users in Twitter

Page 26: Inferring Topical Attributes of Users in the Twitter ...perso.uclouvain.be/jean-charles.delvenne/index... · Inferring Topics of Interest of a User! Prior attempts # Use NLP on tweets

Topical interests of Duncan Watts

Interests not same as expertise

Topical expertise of Duncan Watts

Page 27: Inferring Topical Attributes of Users in the Twitter ...perso.uclouvain.be/jean-charles.delvenne/index... · Inferring Topics of Interest of a User! Prior attempts # Use NLP on tweets

Identifying topical experts Inferring topical interests of users Application: Interaction among experts and interested users

Page 28: Inferring Topical Attributes of Users in the Twitter ...perso.uclouvain.be/jean-charles.delvenne/index... · Inferring Topics of Interest of a User! Prior attempts # Use NLP on tweets

Topical groups in Twitter Topical Groups = Experts + Seekers

Experts: Users who have expertise on the topic Seekers: Users who are interested in the topic

@BarackObama Expert on Politics

@BarackObama

Seeker on Basketball

Page 29: Inferring Topical Attributes of Users in the Twitter ...perso.uclouvain.be/jean-charles.delvenne/index... · Inferring Topics of Interest of a User! Prior attempts # Use NLP on tweets

Identifying topical groups at scale n  Crawled data for first 38 million users in Twitter

n  88 million Lists, 1.5 billion social links

n  Identified 36 thousand topical groups

Deep Twitter Diving: Exploring Topical Groups in Microblogs at Scale, ACM Conference on Computer Supported Cooperative Work and Social Computing (CSCW) 2014

Page 30: Inferring Topical Attributes of Users in the Twitter ...perso.uclouvain.be/jean-charles.delvenne/index... · Inferring Topics of Interest of a User! Prior attempts # Use NLP on tweets

Diversity: Topics and Group Size

Page 31: Inferring Topical Attributes of Users in the Twitter ...perso.uclouvain.be/jean-charles.delvenne/index... · Inferring Topics of Interest of a User! Prior attempts # Use NLP on tweets

A Small Number of Very Popular Groups

Page 32: Inferring Topical Attributes of Users in the Twitter ...perso.uclouvain.be/jean-charles.delvenne/index... · Inferring Topics of Interest of a User! Prior attempts # Use NLP on tweets

Thousands of Specialized Niche Groups

Page 33: Inferring Topical Attributes of Users in the Twitter ...perso.uclouvain.be/jean-charles.delvenne/index... · Inferring Topics of Interest of a User! Prior attempts # Use NLP on tweets

Detecting topical groups n  We followed a semantic approach to identify topical

groups

n  Could community detection algorithms be used on the social network to detect them?

n  Applied BGLL / Louvain algorithm on the Twitter social network to identify communities

Page 34: Inferring Topical Attributes of Users in the Twitter ...perso.uclouvain.be/jean-charles.delvenne/index... · Inferring Topics of Interest of a User! Prior attempts # Use NLP on tweets

Detecting topical groups

n  BGLL largely unable to detect topical groups, especially the smaller ones (on niche topics)

n  Topical groups do not have good structural quality

on which most community detection algorithms rely

n  Difference explained by a sociological theory on how groups form in a social network

Page 35: Inferring Topical Attributes of Users in the Twitter ...perso.uclouvain.be/jean-charles.delvenne/index... · Inferring Topics of Interest of a User! Prior attempts # Use NLP on tweets

Why do groups form?

Identity Based Groups

Low Reciprocity

Low Personal Interactions High Topicality of discussions

Examples: audience at a conference, topical groups

Bond Based Groups

High Reciprocity

High Personal Interactions Low Topicality of discussions

Examples: family, personal friends

Common Identity and Bond Theory - Prentice, Personality and Social Psychology Bulletin, 1994

Page 36: Inferring Topical Attributes of Users in the Twitter ...perso.uclouvain.be/jean-charles.delvenne/index... · Inferring Topics of Interest of a User! Prior attempts # Use NLP on tweets

Summary n  How to identify topical expertise / interests of users?

q  Proposed crowdsourcing based methods

n  Identified topical groups in Twitter at scale

n  Ongoing work q  Developing search / recommender systems to help users

find important content relevant to their interests

Thank You !