This thesis describes an extension of the opinion-integrated agent-based betting exchange simulation platform Bristol Betting Exchange (BBE). BBE is a synthetic data generator of an in-play market on a betting exchange, providing a means to generate sufficient data necessary for the discovery of profitable trading strategies using machine learning techniques. This work considers the design and implementation of several features intended to improve the realism of the BBE platform by modifying the interactions that occur between agent-bettors. A recently introduced approach to modelling opinion dynamics has been implemented, attempting to mathematically capture the observation that humans do not communicate in exact numbers as in traditional opinion dynamics models, but in language that can cause ambiguity in how opinions are communicated. This method builds upon the Bounded Confidence (opinion dynamics model by using utilising a fuzzy logic system for the determination of agent-specific interaction weights. In addition to exploring the effects of this model within BBE, we introduce a method that introduces varying degrees of ambiguity into interactions before they occur. To further increase the realism of the interactions between agents, a network structure is implemented into the population via the Watts-Strogatz method. The impact of using such a structure in selecting the agent-bettors for an interaction is explored, from a completely random process to the direct neighbours of an agent. A novel method for selecting participants based on their degree of connection to the bettor initiating the interaction is then explored to achieve a medium between these two extremes. Finally, group-level versions of the original Bounded Confidence and new Fuzzy Bounded Confidence model are implemented to compare the effects of larger group interactions with their pairwise counterparts.