Summary: In this talk we present an innovation to current VAA design which is based on the introduction of a social network element. We refer to this new type of online tool as a Social Voting Advice Application (SVAA). SVAAs extend VAAs by providing (a) community-based recommendations, (b) comparison of users’ political opinions and (c) a channel of user communication. In addition, SVAAs, enriched with data mining modules, can operate as citizen sensors recording the sentiment of the electorate on issues and candidates. Drawing on VAA datasets generated by the Preference Matcher research consortium, we evaluate the results of the first VAA –Choose4Greece– which incorporated social voting features and was launched during the landmark Greek national elections of 2012. We demonstrate how a Social VAA can provide community based features and, at the same time, serve as a citizen sensor. Evaluation of the proposed techniques is realized on a series of datasets collected from various VAAs, including Choose4Greece. The collection is made available on-line in order to promote research in the field.
Authors: Ioannis Katakis, Nicolas Tsapatsoulis, Fernando Mendez, Vasiliki Triga, Constantinos Djouvas
A Journey Into the Emotions of Software Developers
Social Recommendations in Voting Advice Applications
1. Social Recommendations
in
Voting Advice Applications
Ioannis Katakis
Fernando Mendez
University of
Athens
University of Zurich
Nicolas Tsapatsoulis
Vasiliki Triga
Costas Djiouvas
Cyprus University of
Technology
2. Summary
Provide Community Recommendations
“How do people with similar ideas vote?”
Machine Learning and Collaborative Filtering
VAA Datasets
Embedded in recent VAAs
Users “Like” social recommendation
Researchers “Like” the data insight
Ioannis Katakis, Social Recommendations in VAAs
2
3. Idea
Provide social (community) recommendation (advice)
Original VAA
Which party
share similar
opinions with
me?
Ioannis Katakis, Social Recommendations in VAAs
Social VAA
How do voters
that share similar
opinions with me
chose to vote?
3
5. How do they work?
Identify similar items
Identify similar users
Collaborative filtering (item based – user based)
Ioannis Katakis, Social Recommendations in VAAs
5
6. Data Classification – Supervised Learning
Analyze data (examples) > Learn to predict classes
Orange
Learn “Hidden” Function
𝑓 𝑋 → {𝑂𝑟𝑎𝑛𝑔𝑒, 𝐴𝑝𝑝𝑙𝑒}
Apple
feature values
(e.g. color, shape, size, weight, etc.)
Ioannis Katakis, Social Recommendations in VAAs
6
8. Data Clustering – Unsupervised Learning
Identify groups of
similar items
Similarity?
Euclidean Distance
Algorithms?
k-Means, EM, etc.
Ioannis Katakis, Social Recommendations in VAAs
8
9. Modeling the VAA problem as ML problem
Features : 30 Questions (totally disagree,…, totally agree)
Class Labels : Vote Intention (political parties)
Examples: Users already in the database
Ioannis Katakis, Social Recommendations in VAAs
9
10. Evaluation
On real VAA datasets
Train – Test split (10 fold cross validation)
Train the dataset on x% of the data
Evaluate (test) on the rest (100-x)%
Ioannis Katakis, Social Recommendations in VAAs
10
11. Approaches
Party coding (not social)
How VAAs currently work.
Voter-Party opinion
similarity
Average voter
Average the profiles of the
voters of each party
separately
Ioannis Katakis, Social Recommendations in VAAs
11
13. Results – basic approaches
Social Approaches > Party Coding
Ioannis Katakis, Social Recommendations in VAAs
Data: Greece, 2011
13
14. Results – various classifiers
Support Vector Machines – Best Predictive Performance
Collaborative Filtering - Fast + Accurate
Ioannis Katakis, Social Recommendations in VAAs
14
15. Results – various datasets
Party-Coding < SMO in all datasets
Difference between datasets maybe correlated with number of
parties, training data size, community agreement
Ioannis Katakis, Social Recommendations in VAAs
15
16. … in the VAA
Also in… Cyprus 2013, Germany 2013, …
Ioannis Katakis, Social Recommendations in VAAs
16
17. What users think…
Like button
likes
satisfaction =
likes + dislikes + neutral
Users seem to like more the social recommendations
Ioannis Katakis, Social Recommendations in VAAs
17
18. What else? – Attribute Selection
Information Gain: ΙG D, a = H D − H T a
H : information entropy
Ioannis Katakis, Social Recommendations in VAAs
18
19. What else? – Data Clustering
Ioannis Katakis, Social Recommendations in VAAs
19
20. Conclusions
Applied Machine Learning Algorithms to VAA data
… to provide social-based advice
… gain data insight
Social-based advice is more accurate than profile matching
VAA users seem to like this feature
Ioannis Katakis, Social Recommendations in VAAs
20
21. More…
Katakis, I.; Tsapatsoulis, N.; Mendez, F.; Triga, V.; Djouvas, C.,
"Social Voting Advice Applications - Definitions, Challenges,
Datasets and Evaluation," IEEE Transactions on Cybernetics
Thank you for
your attention!
www.katakis.eu
ioannis.katakis@gmail.com
@iokat
www.preferencematcher.org
Ioannis Katakis, Social Recommendations in VAAs
21
Notes de l'éditeur
The idea follows the recent trend of recommendation systems which are actually software applications (usually web applications) tha recommend us items based on previous preference. So good reads is web site that you can enter the books you have read and rate them and it