Abstract
An individual’s conflict resolution styles can have a large impact on the decision making process of a group. This impact is affected by a variety of factors, such as the group size, the similarity of the group members, and the type of support offered by the recommender system, if the group is using one. Measuring the effect of these factors goes beyond the capability of a live user study. In this article we show that simulation-based experiments can be effectively exploited to analyse the effect of the group members’ conflict resolution styles and to help researchers to formulate additional research hypotheses, which could be individually tested in ad hoc user studies. We therefore propose a group discussion procedure that simulates users’ actions while trying to make a group decision. The simulated users adopt alternative conflict resolution styles derived from the Thomas–Kilmann Conflict Model. The simulation procedure is informed by the analysis of real users’ interaction logs with a group discussion support system. Our experiments are conducted on scenarios characterized by four group factors, namely, conflict resolution style, inner-group similarity, interaction length and group size. We demonstrate the effect of these factors on the recommendation quality. This is measured by the loss in the utility obtained by an individual when choosing the recommended group choice rather than his/her individual best choice. We also measure the difference between the highest and lowest utility that the group members obtain, in order to understand the fairness of the group recommendation identified by the system. The experimental results show (among other findings) that if group members have similar tastes then groups composed of users with the competing conflict resolution style obtain the largest utility loss, compared to groups whose members adopt the cooperative styles (accommodating and collaborating), and yet, whatever their conflict resolution styles, there is no distinct difference in their utility for the group choice (they are treated equally). Conversely, when group members have diverse preferences, the average utility loss of competing members is still the largest, but the differences in their utility is the lowest (they all get a similar but lower utility). Some of the findings of our simulation experiments also match observations made in real group discussions and they pave the way for new user studies aimed at further supporting the reported findings.