Data fusion is the process of resolving conflicting data from multiple data sources. As the data sources are inherently heterogenous, there is a need for an expert to resolve the conflicting data. Traditional approach requires the expert to resolve a considerable amount of conflicts in order to acquire a high quality dataset. In this project, we consider how to acquire a high quality dataset while maintaining the expert effort minimal. At first, we achieve this goal by building a model which leverages the provenance of the data in reconciling conflicting data. Secondly, we improve our model by taking the dependency between data sources into account. In the end, we empirically show that our solution can significantly reduce the user effort while it can obtain a high quality dataset in comparison with traditional method.