Urban hotspots reflect the degree of residents' travel gathering. The study of urban hotspots has important values for urban infrastructure planning, public security and other aspects. In existing researches, single-source location data and density-based clustering algorithms are used to mine hotspots. Due to the one-sidedness of using the single-source data, the mining of hotspots based on multi-source location data fusion has become a hot topic. Multi-source location data fusion requires a quantity balance between the data sets to be fused, because several famous clustering algorithms cannot handle multi-source imbalanced data sets. To solve this problem, we propose a novel framework to mine urban hotspots. First, we construct a data imputation model for the sparse data set so that reducing the difference in quantity between two types of data sets. Then, a clustering algorithm for imbalanced data sets is proposed, and a novel evaluation metric is designed to verify the effectiveness of clustering results. The experiment uses real data sets including POI data, check-in data and GPS trajectory data. The results show that the proposed method discovers all urban hotspots formed by fused imbalanced data sets, and it is more accurate and efficient than the state-of-the-art algorithms.