This study addresses the social-cognitive interactions that occur in code-switching and integrates social and cognitive factors from a usage-based perspective. It investigates code-switching in two different interaction modes -- speech and writing -- to consider not only the influence of social factors (social networks and attitudes) but also the relative cognitive processing load. The data were gathered from 40 Chinese-English bilinguals in London, derived from interviews and written data on their most active social media site, SinaWeibo. Their socio-biographical data were collected via a questionnaire. A multivariate analysis shows that, rather than there being a simple dominance of either the social or the cognitive factors, there is an interplay between the two. A speaker's code-switching corresponds to his/her previous exposure through social networks, but personal attitudes, e.g. a positive view of English, can override network-based predictions of use. Crucially, however, we only see attitudes exerting this significant effect within the domain of contexts with low cognitive processing demand (e.g. asynchronous writing). The findings of this study show that personal preference can indeed override language usage in interactive networks, but such effect is constrained by individual differences in cognitive capacities of processing, which in turn relates back to the frequency of usage which automatizes processing.