Persuasion is an important and yet complex aspect of human intelligence. When undertaken through dialogue, the deployment of good arguments, and therefore counterarguments, clearly has a significant effect on the ability to be successful in persuasion. A key dimension for determining whether an argument is good is the impact that it has on the concerns of the intended audience of the argument (e.g., the other participant(s) in the dialogue). In this paper, we investigate how we can acquire and represent concerns of a participant, and her preferences over them, and we show how this can be used for selecting good moves in a persuasion dialogue. We provide results from empirical studies showing that: (1) we can gather preferences over types of concern; (2) there is a common understanding of what is meant by concerns; (3) participants tend to make moves according to their preferences; and (4) the persuader can use these preferences to improve the persuasiveness of a dialogue.

Computational models of argument are being developed to capture aspects of how persuasion is undertaken. Recent proposals suggest that in a persuasion dialogue between some agents, it is valuable for each agent to model how arguments are believed by the other agents. Beliefs in arguments can be captured by a joint belief distribution over the arguments and updated as the dialogue progresses. This information can be used by the agent to make more intelligent choices of move in the dialogue. Whilst these proposals indicate the value of modelling the beliefs of other agents, there is a question of the computational viability of using a belief distribution over all the arguments. We address this problem in this paper by presenting how probabilistic independence can be leveraged to split this joint distribution into an equivalent set of distributions of smaller size. Experiments show that updating the belief on the split distribution is more efficient than performing updates on the joint distribution.

Abstract. Hidden-Mode Markov Decision Processes (HM-MDPs) were proposed to represent sequential decision-making problems in non-stationary environments that evolve according to a Markov chain. We introduce in this paper Hidden-Semi-Markov-Mode Markov Decision Processes (HS3MDPs), a generalization of HM-MDPs to the more realistic case of non-stationary environments evolving according to a semi-Markov chain. Like HM-MDPs, HS3MDPs form a subclass of Partially Observable Markov Decision Processes. Therefore, large instances of HS3MDPs (and HM-MDPs) can be solved using an online algorithm, the Partially Observable Monte Carlo Planning (POMCP) algorithm, based on Monte Carlo Tree Search exploiting particle filters for belief state approximation. We propose a first adaptation of POMCP to solve HS3MDPs more efficiently by exploiting their structure. Our empirical results show that the first adapted POMCP reaches higher cumulative rewards than the original algorithm. However, in larger instances, POMCP may run out of particles. To solve this issue, we propose a second adaptation of POMCP, replacing particle filters by exact representations of beliefs. Our empirical results indicate that this new version reaches high cumulative rewards faster than the former adapted POMCP and still remains efficient even for large problems.

Participants in dialogical argumentation often make strategic choices of move, for example to maximize the probability that they will persuade the other opponents. Multiple dimensions of information about the other agents (e.g., the belief and likely emotional response that the other agents might have in the arguments) might be used to make this strategic choice. To support this, we present a framework with implementation for multi-criteria decision making for strategic argumentation. We provide methods to improve the computational viability of the framework, and analyze these methods theoretically and empirically. We finally present decision rules supported by the psychology literature and evidence using human experiments.