A heuristic strategy for persuasion dialogues Josh Murphy, Elizabeth Black, Michael Luck King’s College London josh.murphy@kcl.ac.uk May 19, 2016 Josh Murphy, Elizabeth Black, Michael Luck (KCL) A heuristic strategy for persuasion dialogues May 19, 2016 1 / 18
Overview Persuasion dialogues 1 Current methods for computing a dialogue strategy 2 Heuristic strategy 3 Results 4 Future work 5 Josh Murphy, Elizabeth Black, Michael Luck (KCL) A heuristic strategy for persuasion dialogues May 19, 2016 2 / 18
Argument-based persuasion dialogues Initial condition: Agents have conflicting views on a topic. Individual goals: At least one agent aims to persuade the other that the topic is acceptable/unacceptable. Goals of the dialogue: Attempt to resolve a conflicting view on the topic. Josh Murphy, Elizabeth Black, Michael Luck (KCL) A heuristic strategy for persuasion dialogues May 19, 2016 3 / 18
Example dialogue Global Knowledge Persuader Persuadee A B F A B F C T C T C T D E D E Josh Murphy, Elizabeth Black, Michael Luck (KCL) A heuristic strategy for persuasion dialogues May 19, 2016 4 / 18
Example dialogue Persuader Persuadee A B A F C T A C T D E Josh Murphy, Elizabeth Black, Michael Luck (KCL) A heuristic strategy for persuasion dialogues May 19, 2016 5 / 18
Example dialogue Persuader Persuadee A B A F C T D C T D D E Josh Murphy, Elizabeth Black, Michael Luck (KCL) A heuristic strategy for persuasion dialogues May 19, 2016 6 / 18
Current methods for computing a dialogue strategy AI planning [Black et al., 2014]. Mixed observability Markov decision problems [Hadoux et al., 2015] Minimax algorithm [Rienstra et al., 2013] However, none of these approaches have been shown to scale to domains with more than 10 arguments. Josh Murphy, Elizabeth Black, Michael Luck (KCL) A heuristic strategy for persuasion dialogues May 19, 2016 7 / 18
Heuristic strategy We want a strategy that can be computed in domains with many arguments, even if the strategy is not optimal. To find a time-e ffi cient strategy we consider the local topological properties of arguments graphs to determine some estimate of how beneficial an argument would be if asserted. Josh Murphy, Elizabeth Black, Michael Luck (KCL) A heuristic strategy for persuasion dialogues May 19, 2016 8 / 18
Estimating the influence of arguments To obtain an accurate estimate of how beneficial asserting an argument will be we want to take into account: Does the argument support or defend the topic? What is the estimated of influence the argument has over the topic? Josh Murphy, Elizabeth Black, Michael Luck (KCL) A heuristic strategy for persuasion dialogues May 19, 2016 9 / 18
Estimating the influence of arguments Does the argument support or defend the topic? A B C D T E F Josh Murphy, Elizabeth Black, Michael Luck (KCL) A heuristic strategy for persuasion dialogues May 19, 2016 10 / 18
Estimating the influence of arguments What is the estimated influence an argument has over the topic? A B C D E F G T H I J K Josh Murphy, Elizabeth Black, Michael Luck (KCL) A heuristic strategy for persuasion dialogues May 19, 2016 11 / 18
Estimating the influence of arguments A B C D T E F G C C D T C E F G +1/16 -1/8 +1/4 -1/2 +1 Josh Murphy, Elizabeth Black, Michael Luck (KCL) A heuristic strategy for persuasion dialogues May 19, 2016 12 / 18
Estimating the influence of arguments Global Knowledge Persuader Persuadee A B F A B F C T C T C T D E D E Josh Murphy, Elizabeth Black, Michael Luck (KCL) A heuristic strategy for persuasion dialogues May 19, 2016 13 / 18
Experimental evaluation Evaluation through simulation. Generate thousands of instances of dialogue scenarios. I Requires many argumentation frameworks to act as a domain I Ideally, from real-world sources — but, public databases not large enough for serious empirical evaluation. I So, generate random frameworks with “realistic” properties Measure if the persuader is successful when using the heuristic strategy, and how long computing the strategy takes. Use a random strategy as a lower bound on success. Josh Murphy, Elizabeth Black, Michael Luck (KCL) A heuristic strategy for persuasion dialogues May 19, 2016 14 / 18
Results The heuristic strategy is fast to compute, and e ffi ciently scales to domains with 50 arguments. Args 10 20 30 40 50 < 0.1 0.21 0.37 0.56 0.77 Time Table: Time to compute heuristic strategy (seconds). Args is the number of arguments in the domain. Josh Murphy, Elizabeth Black, Michael Luck (KCL) A heuristic strategy for persuasion dialogues May 19, 2016 15 / 18
Results The heuristic strategy has a high success rate. Figure: Percentage success rate of strategies. Error bars indicate standard error. Josh Murphy, Elizabeth Black, Michael Luck (KCL) A heuristic strategy for persuasion dialogues May 19, 2016 16 / 18
Future work Application of the strategy to more complex dialogue scenarios. I Particularly dialogues with more than two participants. The generation of argumentation frameworks for the use in simulation has a large e ff ect on the resulting dialogue. I What structures of framework exist in real-world domains? Di ff erent argumentation schemes, argument mining of di ff erent sources, models of argument (extended frameworks)... I What impact do di ff erent structures have on other argument-based systems? Dialogues, argument solvers, dynamic argumentation... Josh Murphy, Elizabeth Black, Michael Luck (KCL) A heuristic strategy for persuasion dialogues May 19, 2016 17 / 18
References E. Black, A. Coles, S. Bernardini (2014) Automated planning of simple persuasion dialogues Computational Logic in Multi-Agent Systems , LNCS vol. 8624, Springer, 87 – 104. E. Hadoux, A. Beynier, N. Maudet, P. Weng, A. Hunter (2015) Optimization of probabilistic argumentation with Markov decision models. In Proceedings of the Twenty-Fourth International Joint Conference on Artificial Intelligence , AAAI Press, 2004 – 2010. T. Rienstra, M. Thimm, N. Oren (2013) Opponent models with uncertainty for strategic argumentation. In Proceedings of the Twenty-Third International Joint Conference on Artificial Intelligence , AAAI Press, 332 – 338. Josh Murphy, Elizabeth Black, Michael Luck (KCL) A heuristic strategy for persuasion dialogues May 19, 2016 18 / 18
Recommend
More recommend
