1) The document discusses predicting the semantic category of hashtags on Twitter based on how they are collectively used. 2) It finds that different semantic categories (e.g. idioms, technology) exhibit distinct usage patterns, and that informational coverage and changes in follower networks over time best distinguish the categories. 3) Models using pragmatic features related to how hashtags and their discussion networks evolve over time can accurately predict hashtag semantic categories, and outperform models using only lexical features from text. Pragmatic features have advantages in being language-independent and applicable beyond text.

1. TU Graz - Knowledge Management Institute
1
Philipp Singer Rio de Janeiro, 2013-05-13 Meaning as Collective Use
Meaning as Collective Use:
Predicting Semantic Hashtag Categories on Twitter
Lisa Posch, Claudia Wagner, Philipp Singer, Markus Strohmaier
Knowledge Management Institute and Know Center
Graz University of Technology, Austria

2. TU Graz - Knowledge Management Institute
2
Philipp Singer Rio de Janeiro, 2013-05-13 Meaning as Collective Use
Motivation
Twitter
ContentPragmatics?

3. TU Graz - Knowledge Management Institute
3
Philipp Singer Rio de Janeiro, 2013-05-13 Meaning as Collective Use
Semantic Hashtag Category
Hashtags
Semantic Category?
Conference?
Meaning is use [Wittgenstein]
Content: narrow lexical context of a wordMeaning of a word is defined by the variety of uses to which the word is putPragmatics of a word – how a hashtag is used by a large group of users
Politics?
Technology?

4. TU Graz - Knowledge Management Institute
4
Philipp Singer Rio de Janeiro, 2013-05-13 Meaning as Collective Use
Pragmatics
• structural patterns of social connections
• Is the stream consumed by the same users that contribute to it?
• Are social connections distributed evenly?
• How much do the patterns change over time?
• ...
• the structural context in which a hashtag occurs
• How democratically is a hashtag used?
• How conversational are tweets of a hashtag stream?
• ...

5. TU Graz - Knowledge Management Institute
5
Philipp Singer Rio de Janeiro, 2013-05-13 Meaning as Collective Use
Research Questions
1. Do different semantic categories of hashtags reveal
substantially different usage patterns?
2. To what extent do pragmatic and lexical properties
of hashtags help to predict the semantic category of
a hashtag?

6. TU Graz - Knowledge Management Institute
6
Philipp Singer Rio de Janeiro, 2013-05-13 Meaning as Collective Use
Dataset
 Twitter
 Semantic categories
[Romero et al.]
technology games
idioms music
sports celebrity
political movies
#factaboutme
#followfriday
#dontyouhate #iloveitwhen
#nevertrust
#iwish #omgfacts
#oneofmyfollowers
#rememberwhen
#wheniwaslittle
D. M. Romero, B. Meeder, and J. Kleinberg. Differences in the mechanics of information diffusion across topics: idioms, political hashtags, and complex contagion on Twitter. In Proceedings
of the 20th international conference on World wide web, WWW '11, pages 695{704, New York, NY, USA, 2011. ACM.

7. TU Graz - Knowledge Management Institute
7
Philipp Singer Rio de Janeiro, 2013-05-13 Meaning as Collective Use
Dataset
 three parts
 time frames of four weeks
 hashtag stream tweets
 social structure of authors
D. M. Romero, B. Meeder, and J. Kleinberg. Differences in the mechanics of information diffusion across topics: idioms, political hashtags, and complex contagion on Twitter. In Proceedings
of the 20th international conference on World wide web, WWW '11, pages 695{704, New York, NY, USA, 2011. ACM.
Static features Dynamic
features

8. TU Graz - Knowledge Management Institute
8
Philipp Singer Rio de Janeiro, 2013-05-13 Meaning as Collective Use
Features
 Static Pragmatic Measures
 author, follower, followee, friend entropies
 measure democracy of distributions
 author-follower, -followee, -friend overlaps
 measures if stream is consumed and produced by same users
 informational, hashtag, retweet, conversational coverages
 measures the nature of messages
 Dynamic Pragmatic Measures
 symmetric KL divergence for authors, followers, followees, friends
 measure how stable the social structure of a stream is
 Lexical Measure
 term frequency

9. TU Graz - Knowledge Management Institute
9
Philipp Singer Rio de Janeiro, 2013-05-13 Meaning as Collective Use
Do different semantic categories of
hashtags reveal substantially different
usage patterns?

10. TU Graz - Knowledge Management Institute
10
Philipp Singer Rio de Janeiro, 2013-05-13 Meaning as Collective Use
Usage Patterns
 Pragmatic fingerprints
 Differences between categories
 Statistically significant?
 Pairwise comparison of categories
 Mann-Whitney-Wilcoxon-Test

11. TU Graz - Knowledge Management Institute
11
Philipp Singer Rio de Janeiro, 2013-05-13 Meaning as Collective Use
Results: Usage Patterns
 With p < 0.05:
 26 statistical significances
 Best distinguishable categories: idioms, technology
 Most discriminative features: informational coverage,
KL divergences for followers, authors, and friends

12. TU Graz - Knowledge Management Institute
12
Philipp Singer Rio de Janeiro, 2013-05-13 Meaning as Collective Use
Results: Usage Patterns

13. TU Graz - Knowledge Management Institute
13
Philipp Singer Rio de Janeiro, 2013-05-13 Meaning as Collective Use
Preliminary observations
 Pragmatic features can help to distinguish semantic
categories
 Idioms and technology exhibit more distinct usage
patterns than other semantic categories
 Informational coverage and KL divergence are the
most discriminative features

14. TU Graz - Knowledge Management Institute
14
Philipp Singer Rio de Janeiro, 2013-05-13 Meaning as Collective Use
To what extent do pragmatic and lexical
properties of hashtags help to predict the
semantic category of a hashtag?

15. TU Graz - Knowledge Management Institute
15
Philipp Singer Rio de Janeiro, 2013-05-13 Meaning as Collective Use
Hashtag Prediction
 Classify temporal snapshots of hashtag streams into
their correct semantic categories
 By analyzing how they are used over time
 Extremely Randomized Trees
 Stratified 6-fold Cross Validation
 Baseline (randomly permuted categories 100 times)

16. TU Graz - Knowledge Management Institute
16
Philipp Singer Rio de Janeiro, 2013-05-13 Meaning as Collective Use
Hashtag Prediction Models
 Static Pragmatic
 Dynamic Pragmatic
 Combined Pragmatic
 Lexical
 Combined Pragmatic and Lexical

17. TU Graz - Knowledge Management Institute
17
Philipp Singer Rio de Janeiro, 2013-05-13 Meaning as Collective Use
Results: Hashtag Prediction

18. TU Graz - Knowledge Management Institute
18
Philipp Singer Rio de Janeiro, 2013-05-13 Meaning as Collective Use
Results: Hashtag Prediction
 Feature Ranking
 Information Gain
1. Informational coverage
2. KL divergence followers
3. KL divergence friends
4. Hashtag coverage
5. Friend entropy

19. TU Graz - Knowledge Management Institute
19
Philipp Singer Rio de Janeiro, 2013-05-13 Meaning as Collective Use
Discussion
 Lexical features perform better
 But lexical features exhibit limitations
 text and language dependent
 only for settings with textual content
 Pragmatic features have advantages
 rely on usage information
 independent of the type of content
 may also be computed for social video or image streams
 multi-language corpora

20. TU Graz - Knowledge Management Institute
20
Philipp Singer Rio de Janeiro, 2013-05-13 Meaning as Collective Use
Conclusions & Implications
 Collective usage of hashtags reveals information
about their semantics
 Further insights necessary; especially for domains
where no textual content is available
 Pragmatic features can supplement lexical features

21. TU Graz - Knowledge Management Institute
21
Philipp Singer Rio de Janeiro, 2013-05-13 Meaning as Collective Use
Thanks for your attention!
Pragmatic features can play a vital role in
supplementing or replacing lexical features!