Se ha denunciado esta presentación.
Utilizamos tu perfil de LinkedIn y tus datos de actividad para personalizar los anuncios y mostrarte publicidad más relevante. Puedes cambiar tus preferencias de publicidad en cualquier momento.

Ambient Intelligence Services Personalization via Social Choice Theory

638 visualizaciones

Publicado el

Ambient Intelligence Services Personalization via Social
Choice Theory

Publicado en: Ciencias
  • Sé el primero en comentar

  • Sé el primero en recomendar esto

Ambient Intelligence Services Personalization via Social Choice Theory

  1. 1. Ambient Intelligence Services Personalization via Social Choice Theory December 2014 Emilio Serrano, Pablo Moncada, Mercedes Garijo and Carlos A. Iglesias {emilioserra,pmoncada,mga,cif} Intelligent Systems Group Technical University of Madrid 1
  2. 2. Outline • Outline.  Motivation.  How easy is (automatic) voting?  Target system  Users’ preferences theories  VoteSim and case study  Experimental results  Conclusion 2
  3. 3. Motivation • Ambient Intelligence (AmI) focuses on adapting to people’s needs and particular situations. • Important question: what happen when resources are shared and there are conflicts between users’ preferences?  AmI systems may not only have to offer a good service considering users’ preferences,  but also to make a decision in an attempt to maximize the users’ welfare when they access shared services • Some examples in hotels:  screening rooms, decoration based on dynamic decorative panels, dance clubs, heated swimming pools, etcetera. 3
  4. 4. Motivation 2 • Accessing shared resources has not been explored in the AmI specialized literature  Just knowing the “who” and the “where” in a non intrusive manner is tough  But still a great research question for more intrusive or future scenarios 4
  5. 5. Motivation 3 • Agreement technologies (ATs) and multi-agent systems (MASs) have studied this in depth.  multi-issue negotiations, concurrent negotiations, strategy- proof mechanisms, argumentation, auctions, voting, etcetera  social choice theory seems straightforward.  The evaluation of alternative methods of collective decision-making  There aren’t good arguments for “The football match is okay, but I’d rather watch game of thrones” 5
  6. 6. Motivation 4 • Primary hypothesis: Social choice theory in Ambient Intelligence systems can improve significantly users’ satisfaction when accessing shared resources  … but the social choice theory, has mainly focused on theoretical works which deal with political elections  … even when the most popular MASs books include a social choice chapter • Research methodology based on agent based social simulations is employed to support this hypothesis and to evaluate these benefits.  what are the benefits of using a voting system in an intelligent environment?; what are the most suitable voting systems?; and, what differences does this case present when compared to political elections? 6
  7. 7. How easy is (automatic) voting? • There are a large number voting systems  Plurality, range, Borda, approval, cumulative, etcetera  Cumulative: Each voter is given k votes, which can be cast arbitrarily. • Voting may involve strategy  In UK we have the Labor (left-wing party), Liberal (center-left) and Conservative (right wing) parties, what a very left-wing voter should vote in a district with a bias towards the Conservative party?.  (Preferences + Voting method) may not be enough to vote  e.g. cumulative voting • Everybody knows that voting leads to ignored minorities…  But this may happen for majorities Podemos > IU > PSOE IU> PSOE > Podemos PSOE > Podemos > IU 7
  8. 8. How easy is (automatic) voting? 2 • No silver bullet!, the best group decision policy depends on:  Voting method  Users strategies  The users and their preferences when using services  The environment and how the services are deployed  The possible services configurations  The number of repeated services • Methods and tools are needed a make a decision on how to maximize social welfare  …and to minimize the maximum time without a wanted configuration/service. 8
  9. 9. Target system 9
  10. 10. Users’ preferences theories iduser sports sitcom Doc. soap Cart. news reality quiz movie u1 1 4 4 9 0 2 3 4 3 u2 9 0 4 7 7 3 8 6 10 u3 9 1 6 10 2 6 2 5 10 u4 2 10 2 3 9 5 9 7 10 u5 6 7 9 3 6 5 1 5 1 u6 0 10 4 3 1 9 4 1 2 u7 10 7 2 7 5 2 6 5 0 • I recommend you to go to room Sx •(more users in your cluster) •E.g. u2, try to vote where u3 is and avoid the place where u6 is. 10 (It would be nice to combine several services: you like TV here but temperature is not your cup of tea)
  11. 11. VoteSim and case study • VoteSim is an agent based social simulator designed for:  finding out the best group decision policy for a specific case  exploring manners of forming coalitions and their effects  estimating the usability of a vote system  studying security issues such as the weakness to tactic vote  validating software applications for group decision • VoteSim is based on the UbikSim simulator developed in collaboration with the University of Murcia  • Let us observe the use of VoteSim for a hotel where there are shared TVs in the hall 11
  12. 12. Experimental results Forming coalitions is more important than the voting method! 12
  13. 13. Experimental results 2 • Range voting always gets the best satisfaction… • but pretty bad worst wait (in this scenario) and terrible usability • …a proposed method gets the best balance among metrics • …but usability as bad as range voting • In this scenario, I would choose an approval method  Kind of like “Tinder” but without getting to flirt  But it is a design decision, VoteSim is given for new experiments in new scenarios 13
  14. 14. Conclusion • This paper concludes, by sound and reproducible results, that the use of social choice theory in AmI systems can improve considerably users’ satisfaction when accessing shared resources. • There are significant differences between the AmI case and the political elections • Different metrics are proposed to estimate welfare • The novel voting algorithm called exchange of weights is the most balanced voting system considered • The use of greedy pre-selection mechanisms based on Euclidean or Manhattan distances is more important than the voting method • To ensure the reproducibility of the experimental results given in this paper, the free and open-source toolVoteSim • Future works: more voting methods in VoteSim, more cluster techniques, more complex preferences: • More in: Evaluating social choice techniques into intelligent environments by agent based social simulation. In: Information Sciences , 286 (0), pp. 102–124, 2014, ISBN: 0020- 0255, (Impact Factor 2013, 3.893, Q1). 14