The document discusses user modeling and personalization on Twitter. It identifies four building blocks for generating user profiles from Twitter data: 1) temporal constraints, 2) profile type, 3) semantic enrichment, and 4) weighting schemes. The presentation analyzes how these building blocks impact the characteristics of Twitter-based user profiles over time. It then evaluates how different user modeling strategies can improve personalized news recommendations based on Twitter profiles. Key findings include that entity-based profiles provide better recommendations than topic-based or hashtag-based profiles, semantic enrichment improves quality, and adapting profiles to temporal context helps, especially for topic-based profiles. The discussion considers open research questions around searching and re-using social data, and balancing personalization with ser
Vector Databases 101 - An introduction to the world of Vector Databases
#SDoW2011 Keynote: User Modeling and Personalization on Twitter
1. User Modeling and Personalization
on Twitter
SDoW, ISWC, Bonn, Oct 23, 2011
Fabian Abel
Web Information Systems, TU Delft
Delft
University of
Technology
2. #papers that use Twitter datasets
2006 2007 2008 2009 2010 2011 2012 time
User Modeling and Personalization on Twitter 2
3. Perspectives on Twitter data
Grrrr…that is gr8
http://bit.ly/47gt3
@bob
What are Bob’s personal
interests? What are his
current demands?...
User Modeling and Personalization on Twitter 3
4. What we do: Science and Engineering
for the Personal Web
domains: news social mediacultural heritage public datae-learning
Personalized Personalized
Adaptive Systems
Recommendations Search
Analysis and
User Modeling
Semantic Enrichment,
Linkage and Alignment
user/usage data
Social Web
User Modeling and Personalization on Twitter 4
5. User Modeling Challenge
Personalized News
Recommender
Profile I want my
? personalized news
recommendations!
Analysis and
User Modeling
Semantic Enrichment,
(How)
Linkage and Alignment can we infer a Twitter-
based user profile that supports
a news recommender?
User Modeling and Personalization on Twitter 5
6. QiGao Geert-Jan Houben Ke Tao
Fabian, Qi, Geert-Jan, Ke: Analyzing User Modeling on Twitter
for Personalized News Recommendations. UMAP 2011
User Modeling Framework
Building Blocks for generating valuable user profiles
User Modeling and Personalization on Twitter 6
7. User Modeling Building Blocks 1. Temporal
Constraints
Profile?
(a) time period
1. Which tweets of concept weight
the user should be
analyzed? ? (b) temporal patterns
start weekends end
Morning:
Afternoon:
time
Night:
June 27 July 4 July 11
User Modeling and Personalization on Twitter 7
8. User Modeling Building Blocks 1. Temporal
Constraints
Francesca T Sport
Schiavone 2. Profile
Profile? Type
concept weight
Francesca Schiavone won
?
# hashtag-based
French Open #fo2010 entity-based
T topic-based
French
Open # fo2010
2. What type of concepts
should represent “interests”?
time
June 27 July 4 July 11
User Modeling and Personalization on Twitter 8
9. User Modeling Building Blocks 1. Temporal
Constraints
Francesca (a) tweet-based
Schiavone 2. Profile
Profile? Type
concept weight
Francesca
Francesca Schiavone won! Schiavone
3. Semantic
http://bit.ly/2f4t7a French Open
Enrichment
Tennis
Francesca wins French Open
French
Thirty in women's Open (b) further enrichment
tennis is primordially
old, an age when
agility and desire Tennis
recedes as the …
3. Further enrich the semantics of tweets?
User Modeling and Personalization on Twitter 9
10. User Modeling Building Blocks 1. Temporal
Constraints
Profile? 2. Profile
concept weight Type
?
4. How to weight the Francesca
Schiavone
4
3. Semantic
concepts? French Open 3
Enrichment
Tennis 6
Concept frequency (TF)
4. Weighting
TFxIDF Scheme
weight(Francesca
Time-sensitive weight(French Open)
weight(Tennis)
Schiavone)
time
June 27 July 4 July 11
User Modeling and Personalization on Twitter 10
11. User Modeling Building Blocks
1. Temporal (a) time period
Constraints (b) temporal patterns
(a) hashtag-based
(b) entity-based 2. Profile
(c) topic-based Type
3. Semantic (a) tweet-based
Enrichment (b) further enrichment
(a) concept 4. Weighting
frequency Scheme
User Modeling and Personalization on Twitter 11
12. 1. Temporal (a) time period
Constraints (b) temporal patterns
(a) hashtag-based
(b) entity-based 2. Profile
(c) topic-based Type
3. Semantic (a) tweet-based
Enrichment (b) further enrichment
(a) concept 4. Weighting
frequency Scheme
Analysis
How do the user modeling building blocks impact the (temporal)
characteristics of Twitter-based user profiles?
User Modeling and Personalization on Twitter 12
13. Dataset
more than:
20,000 Twitter users
2 months
Available online:
10,000,000 tweets
http://wis.ewi.tudelft.nl/umap2011/ Assange,
WikiLeaks founder, Julian
under arrest in London
75,000 news articles
Nov 15 Dec 15 Jan 15 time
User Modeling and Personalization on Twitter 13
14. Profile
Size of user profiles Type
~5% of the users
entity-based do not make use of
hashtags
hashtag-based
profiles are empty
Entity-based user
modeling succeeds
topic-based for 100% of the
hashtag-based users
User Modeling and Personalization on Twitter 14
15. Semantic
Impact of Semantic Enrichment Enrichment
More distinct topics
further enrichment
further enrichment per profile
(e.g. exploiting links)
(e.g. exploiting links)
More distinct entities
per profile
Exploiting external
resources allows for
Tweet-based
significantly richer
Tweet-based user profiles
(quantitatively)
entity-based user profiles
topic-based user profiles
User Modeling and Personalization on Twitter 15
16. Temporal
User Profiles change over time Constraints
Hashtag-based profiles
# Example:
change stronger than
d1-distance: entity-based and topic-
old new
?
based profiles
music
T difference between
current profile and past tennis older the profile
The
profile the more it differs from
football
the current profile
User Modeling and Personalization on Twitter 16
17. Temporal
Temporal patterns of user profiles Constraints
2
1. Weekend
profiles differ
weekday vs. weekend profiles
significantly from
d1(pweekday, pweekend) weekday profiles
2. the difference
is stronger than
between day and
day vs. night profiles night profiles
d1(pday, pnight)
topic-based user profiles
User Modeling and Personalization on Twitter 17
18. Observations
• Semantic enrichment allows for richer user profiles
• Profiles change over time: fresh profiles seem to better
reflect current user demands
• Temporal patterns: weekend profiles differ significantly
form weekday profiles
User Modeling and Personalization on Twitter 18
19. 1. Temporal (a) time period
Constraints (b) temporal patterns
(a) hashtag-based
(b) entity-based 2. Profile
(c) topic-based Type
3. Semantic (a) tweet-based
Enrichment (b) further enrichment
(a) concept 4. Weighting
frequency Scheme
Evaluation
How do the user modeling building blocks impact the quality of
Twitter-based profiles for personalized news recommendations?
And can we benefit from the findings of the analysis to improve
recommendations?
User Modeling and Personalization on Twitter 19
20. Twitter-based Profiles for Personalization
• Task: Recommending news articles (= tweets with URLs
pointing to news articles)
• Recommender algorithm: cosine similarity between user
profile and tweets
5.5 relevant
• Ground truth: re-tweets of users tweets per user
• Candidate items: news article tweets posted during
evaluation period
5529 candidate news articles
Recommendations = ?
P(u)= ?
time
1 week
User Modeling and Personalization on Twitter 20
21. Profile
Overview: Type
Performance of User Modeling strategies
Topic-based strategy
improves S@10
# significantly
Entity-based
T strategy improves
the recommendation
quality significantly
(MRR & S@10)
User Modeling and Personalization on Twitter 21
22. Impact of Semantic Enrichment Semantic
Enrichment
T
Tweet-based
Further enrichment
Further semantic enrichment (exploiting links) improves the
quality of the Twitter-based profiles!
User Modeling and Personalization on Twitter 22
23. Impact of temporal characteristics Temporal
Constraints
Adapting to temporal context helps?
Selection of temporal
T start yes weekends end
constraints depends on
type of user profile.
no
Recommendations = ?
•Topic-based profiles:
time
adapting to temporal
context is beneficial
•Entity-based profiles:
startcomplete yes
T startfresh end
long-term profiles
perform better
no Recommendations = ?
complete: 2 months fresh: 2 weeks time
User Modeling and Personalization on Twitter 23
24. Observations
• Best user modeling strategy: Entity-based > topic-based
>hashtag-based
• Semantic enrichment improves recommendation quality
• Adapting to temporal context helps for topic-based
strategy
User Modeling and Personalization on Twitter 24
26. Semantic Web Engineering Perspective
on Twitter (and other social) data
Model of the application,
What is the
Applications e.g. news categories
…that understand and
actual impact of leverage Social Web data sports -> tennis
mining and
integrating social
translate
data on the
&
application?
integrate
Evaluate! Mining Semantics
Social Web vocabulary,
data
e.g. Twitter language
Social Web # fo2011
User Modeling and Personalization on Twitter 26
27. 1. Search on Twitter
Questio
n
compose
answer
Answer
translate between
query and Twitter
vocabulary
How can we find “information” in social (micro-
)streams? How can personalization help?
see also TREC Microblogging Task: http://trec.nist.gov/data/tweets/
User Modeling and Personalization on Twitter 27
28. 2. Re-using Twitter data in other applications
Applications
…that understand and
leverage Social Web data
translate & integrate
between application and
Twitter vocabulary
What kind of knowledge can we learn
from users’ micro-blogging activities
and how can we (re-)use it for what
types of applications?
User Modeling and Personalization on Twitter 28
29. Example: improving product
recommendations with Twitter data?
dbpedia:Mark_Haddon
dbpedia:Dog
dbpedia:Food
I would never eat
dogs!
User Modeling and Personalization on Twitter 29
30. 3. Personalization and Serendipity
Cross UM dataset:
f.abel@tudelft.nl
Profile Cross-system UM: get complete
picture about a person
Reasoning: what type of things Narcissus
could surprise and interest a
person?
How can we balance between
personalization and serendipity?
User Modeling and Personalization on Twitter 30
31. Thank you!
Twitter: @fabianabel
User Modeling and Personalization on Twitter 31
Hinweis der Redaktion
large dataset of more than 10 million tweets and 70,000 news articles
1. Translate between “information need” and Twitter vocabulary and 2. compose answer out of several tweets