Keyword search is the most popular technique for querying the ever growing repositories of RDF graph data on the Web. However, keyword queries are ambiguous. As a consequence, they typically produce on linked data a huge number of candidate results corresponding to a plethora of alternative query interpretations. Current approaches ignore the diversity of the result interpretations and might fail to satisfy the users who are looking for less popular results. In this paper, we propose a novel approach for keyword search result diversification on RDF graphs. Our approach instead of diversifying the query results per se, diversifies the interpretations of the query (i.e., pattern graphs). We model the problem as an optimization problem aiming at selecting k pattern graphs which maximize an objective function balancing relevance and diversity. We devise metrics to assess the relevance and diversity of a set of pattern graphs, and we design a greedy heuristic algorithm to generate a relevant and diverse list of k pattern graphs for a given keyword query. The experimental results show the effectiveness of our approach and proposed metrics and also the efficiency of our algorithm.