This paper presents a framework for utilizing domain ontology and graph representation in ad-hoc document retrieval. The main task is to retrieve a ranked list of (text) documents from a fixed corpus in response to free-form keyword queries. In this work, the query and documents are modeled by enhanced graph-based representations. Ranking features are generated by matching the two representations through semantic similarity measures which consider both semantic and statistical information in documents to improve search performance. The suitability of the solution has been demonstrated through applications of document retrieval such as The learning resource repository management system and The Vietnamese online news aggregating system and The job seeking system in the field of Information Technology. The results show that the incorporation of domain ontology with semantic graph structure improves the quality of the retrieval solution compared with documents modeled by bag of words or vector space model only.