In recent years, and due to the advances of the field of information extraction, a vast amount of web data has surfaced. This data typically comes from heterogeneous sources that may belong to the same domain, but have different schemas. This hinders the potential to integrate these sources and exploit their semantic properties. These data sources are extracted from the web and represented using a relational model, in which they are represented by schemas and tables, or a semantic web model, in which they are represented as RDF knowledge bases. Approaches to integrate heterogeneous data sources, regardless of how they are represented, usually try to infer their semantics automatically based on syntax, which is a difficult task in the absence of human guidance. Even approaches that try to involve a human element in these approaches rely on resolving uncertainties before exposing the data sources to users by explicitly asking the user to give her feedback directly over the mediated schema of the relational data sources or over similar entities that are linked in different RDF knowledge bases. This delays making use of these data sets until a perfect integration of the data sources is achieved. In our work, we utilize user feedback while the system is being used to improve the quality of the answers to the user queries. Specifically, in the relational model, we utilize user feedback to apply changes to the mediated schema and mappings used by the query processor to find answers from heterogeneous tables. Also, in the RDF domain, we utilize user feedback to improve the quality of the owl:sameAs links between equivalent entities from heterogeneous RDF data sets.