The investigation of human activity in location-based social networks such as Twitter is one promising example of exploring spatial structures in order to infer underlying mobility patterns. Previous work regarding Twitter analysis is mainly focused on the spatiotemporal classification of events. However, since the information about the occurrence of a general event can in many cases be considered as given, one identified research gap is the exploration of human spatial behavior within specific mass events to potentially characterize underlying, locally occurring mobility clusters. One key challenge is to explore whether this noisy biased dataset can be a reliable source for the knowledge discovery of human mobility during mass events. In this paper we therefore present an advanced methodological framework, including a generative semantic topic modeling and local spatial autocorrelation approach, to observe both spatiotemporal and semantic clusters during a major sports event in Boston in the US. Our results of the observed spatiotemporally and semantically clustered tweets within the selected case study area have shown the possibility of deriving intra-urban event related mobility patterns with similar spatiotemporal movement.