Teachers and instructional designers frequently incorporate collaborative learning approaches into their e-learning environments. A key factor of collaborative learning that may affect learner outcomes is whether the collaborative groups are assigned project topics randomly or based on a shared interest in the topic. This is a particularly important question for adults, whose performance can depend strongly on how closely the project topic relates to their professional goals. In this study involving an on-line course for 103 professionals we divided the learners into two parallel sections that differed in how they would be assigned to groups to perform collaborative tasks on different topics. In one section, learners were assigned randomly into groups and were given task topics randomly as well. In the other section, they were assigned to a group based on whether they shared a common interest in the topic given to that group. We used Social Network Analysis and Content Analysis to assess the level of collaboration in quantitative and qualitative terms (knowledge construction). Both groups showed a low level of knowledge construction and similar levels of centrality degree and learning performance. However, more learners participated in the collaborative tasks if groups had been assigned based on learners' topic preferences. Our findings suggest that forming groups of adult learners based on topic preferences in on-line environments can increase the number of learners that collaborate, but it does not necessarily improve learner performance.