Fuzzy joins are widely used in a variety of data analysis applications such as data integration, data mining, and master data management. In the context of Big Data, computing fuzzy joins is challenging due to the high computational cost required and the communication cost. While on one hand big fuzzy joins on relational data and on the other hand joins on tree-structured data have been investigated in the literature, to the best of our knowledge, combining the two is still an open problem. In this context, we study methods for leveraging distributed environments in order to compute fuzzy joins over large collections of JSON documents. Our algorithms take into account both the text-similarity of the joining data, as well as its structural similarity.