Community-based courses, driven by connectivist learning theory, provide learners with a social environment that supports collaborative problem solving (CPS). The purpose of this study is to investigate the characteristics of potential collaborators before they engage in CPS. This study analyzed a community-based course, a cMOOC with a total of 145 learners participated in the thematic learning stage and 52 of the 145 chose to participate in the CPS. The interaction pattern of the collaborators and the indirect effects of individual attribute-based (exogenous) effects and network structure-based (endogenous) effects on the interaction pattern were examined using the quadratic assignment procedure (QAP) and the stochastic actor-oriented model (SAOM). Results suggest that substantial interaction activeness and interaction attractiveness are key patterns of the collaborators' social interactions before CPS, and there is some interaction homogeneity among the collaborators. Some individual attribute-based effects, such as age, gender, and occupation, will reduce the early interaction attractiveness of the collaborators. The reciprocity, transitivity, and cumulative advantages of the network structure-based effects will gradually increase collaborators' interaction attractiveness; in contrast, the network structure-based effects simultaneously limit collaborators' interaction activeness. This study provides a theoretical foundation and practical experience regarding encouraging learners' deep collaboration in community-based courses.