Data Web services allow users to access information provided by different companies. Web users often need to compose different Web services to achieve a more complex task that can not be fulfilled by an individual Web service. In addition, user preferences are becoming increasingly important to personalize the composition process. In this paper, we propose an approach to compose data Web services in the context of preference queries where preferences are modelled thanks to fuzzy sets that allow for a large variety of flexible terms such as "cheap", "affordable" and "fairly expensive". Our main objective is to find the top-k data Web service compositions that better satisfy the user preferences. The proposed approach is based on an RDF query rewriting algorithm to find the relevant data Web services that can contribute to the resolution of a given preference query. The constraints of the relevant data Web services are matched to the preferences involved in the query using a set of matching methods. A ranking criterion based on a fuzzyfication of Pareto dominance is defined in order to better rank the different data Web services/compositions. To select the top-k data Web services/compositions we develop a suitable algorithm that allows eliminating less relevant data Web services before the composition process. Finally, we evaluate our approach through a set of experiments.