Two-sided marketplaces are an important component of many existing Internet services like Airbnb and Amazon, which have both consumers (e.g. users) and producers (e.g. retailers). Traditionally, the recommendation system in these platforms mainly focuses on maximizing customer satisfaction by recommending the most relevant items based on the learned user preference. However, it has been shown in previous works that solely optimizing the satisfaction of customers may lead to unfair exposure of items, which jeopardizes the benefits of producers. To tackle this problem, we propose a fairness-aware recommendation framework by using multi-objective optimization, Multi-FR, to adaptively balance the objectives between consumers and producers. In particular, Multi-FR adopts the multi-gradient descent to generate a Pareto set of solutions, where the most appropriate one is selected from the Pareto set. In addition, four fairness metrics/constraints are applied to make the recommendation results on both the consumer and producer side fair. We extensively evaluate our model on three real-world datasets, comparing with grid-search methods and using a variety of performance metrics. The experimental results demonstrate that Multi-FR can improve the recommendation fairness on both the consumer and producer side with little drop in recommendation quality, also outperforming several state-of-the-art fair ranking approaches.