The use of recommendation in trust-based models has its advantages in enhancing the correctness and quality of the rating provided by mobile and autonomous nodes in MANETs. However, building a trust model with a recommender system in dynamic and distributed networks is a challenging problem due to the risk of dishonest recommendations. Dealing with dishonest recommendations can result in the additional problem of data sparsity, which is related to the availability of information in the early rounds of the network time or when nodes are inactive in providing recommendations. This paper investigates the problems of data sparsity and cold start of recommender systems in existing trust models. It proposes a recommender system with clustering technique to dynamically seek similar recommendations based on a certain timeframe. Similarity between different nodes is evaluated based on important attributes includes use of interactions, compatibility of information and closeness between the mobile nodes. The recommender system is empirically tested and empirical analysis demonstrates robustness in alleviating the problems of data sparsity and cold start of recommender systems in a dynamic MANET environment.