User preferences elicitation is a key issue of location recommendation. This paper proposes an adaptive user preferences elicitation scheme based on Collaborative filtering (CF) algorithm for location recommendation. In this scheme, user preferences are divided into user static preferences and user dynamic preferences. The former is estimated based on location category information and historical ratings. Meanwhile, the latter is evaluated based on geographical information and two-dimensional cloud model. The advantage of this method is that it not only considers the diversity of user preferences, but also can alleviate the data sparsity problem. In order to predict user preferences of new locations more precisely, the scheme integrates the similarity of user static preferences, user dynamic preferences and social ties into CF algorithm. Furthermore, the scheme is parallelized on the Hadoop platform for significant improvement in efficiency. Experimental results on Yelp dataset demonstrate the performance gains of the scheme.