Computational models of trust have been proposed for use in ubicomp environments for deciding whether to allow customers to pay with an e-purse or not. In order to build trust in a customer, a means to link transactions using the same e-purse is required. Roughly, trust is a result of knowledge. As the number of transactions increases, the resulting increase in knowledge about the user of the e-purse threatens privacy due to global profiling. We present a scheme (and its prototype) that mitigates this loss of privacy without forbidding the use of trust for smoothing payment by giving the opportunity to the user to divide trust (i.e. transactions) according to context (e.g. location, user's current activity or subset of shops).