An approach based on hybrid semantic model (HSM) was proposed to the solve problem raised in the retrieval process of product knowledge documentation. It expands the traditional user query to a semantic set composed of user preference, context and query, while representing the knowledge documents and user interest with an ontology based fuzzy concept. The leaves in the ontology are selected as components of the document concept vector with the weight determined by the depth of the concept in the ontology graph, the quantity of the information contained, and occurrence in the document and the whole repository. Furthermore, ontology is used to express context and query, and to construct a user preference model. Different relevancy computation methods are adopted for different retrieval models. The semantic similarity between query or user preference and documentation is computed by cosine method. The semantic similarity of context is estimated by the concept distance in the concept hierarchy. Finally, the method is shown by experimentation to be more effective than the classic vector space method.