The classification of city activities on social media is a complex problem based on semantic analysis for smart city modeling. Urban area characteristics are analyzed through hierarchical tasks where location properties are gathered along with municipality records and city demographics. Actually, urban locations are graded in social media since multimodal spatiotemporal features incorporating real-time topics in user activities are indexed. Their popularity, their quality, their customer based interaction are evaluated and shared as city insights in different platforms. As a data source, social communication leads to a gap between unstructured information and online city trends. People who share their activities in social media bring their sentiments as multimodal opinions in several formats which result in a complex big data. City semantics are coupled with city demographics and citizen feelings in urban locations. Therefore, new methods are required to support municipality records in smart city classification. In this study, we have focused on smart city classification tasks in two metropolitan areas; Istanbul and Ankara. We have fetched five different types of locations from Foursquare and Twitter platforms. A pre-trained Turkish language model has been used to perform the microblogging scoring. After the preprocessing steps, we have applied the tuned language model onto our Turkish dataset. Five different language models have been compared using statistical evaluation metrics. Our results show that location microblogging would be a premising benchmark in city semantics. The accuracy rates were found above 90% in three different classes. We conclude that a spatial distribution of user reviews leads to offer new metrics in smart city measurement.