The increased availability of interdependent heterogeneous data generated from different sources is fostering the incorporation of semantic knowledge-based graphs and ontologies in information management and search applications. Most of the existing Information Retrieval systems mainly focus on the semantic analysis of the information contained in heterogeneous data. In their results, they provide documents as query answers without considering (i) detailed information regarding relevant granularity levels of the documents, and most importantly (ii) dependencies between the documents or parts of the documents. To overcome these limitations, we propose a graph-based search and ranking algorithm within a generic framework that retrieves the data in the form of a novel augmented data structure for query answers, which we call hybrid molecules. The latter consist of well-defined subgraphs representing relevant contextual information regarding domain-specific information coupled with structural information related to the document. This improves the search results and reduces users’ efforts in tracking and interpreting them. Experiments conducted on real world data corpus using projects from the building construction industry validate the effectiveness of our approach.