The localness inference problem is to identify whether a person is a local resident in a city or not and the likelihood of a venue to attract local people. This information is critical for many applications such as targeted ads of local business, urban planning, localized news and travel recommendations. While there are prior work on geo-locating people in a city using supervised learning approaches, the accuracy of those techniques largely depends on a high quality training dataset, which is difficult and expensive to obtain in practice. In this study, we propose to exploit spatial-temporal-social constraints from noisy online social media data to solve the localness inference problem using an unsupervised approach. The spatial-temporal constraint represents the correlations between people and venues they visit and the social constraint represents social connections between people. In particular, we develop a Spatial-Temporal-Social-Aware (STSA) inference framework to jointly infer i) the localness of a person and ii) the local attractiveness of a venue without requiring any training data. We evaluate the performance of STSA scheme using three real-world datasets collected from Foursquare. Experimental results show that STSA scheme outperforms the state-of-the-art techniques by significantly improving the estimation accuracy.