Text mining is widely used to infer relationships between biological entities. Most text-mining algorithms utilize a cooccurrence-based approach. The term co-occurrence denotes a relationship between two interesting entities if they appear in the same sentence. Using these approaches current studies have extracted relationships between biological entities such as disease-gene relationships. However, these approaches cannot provide specific information for inferred relationships such as the role of the gene in the disease. To overcome this limitation, we propose a novel approach for inferring disease-gene relationship that provides specific knowledge of the inferred relationships. To implement this method, we first built terms based on text analysis to extract opinion sentences that include disease-gene relationships. We then extracted these opinion sentences and inferred disease-gene relationships by using disease-related and gene-related terms in the opinion sentences. Using these extracted relationships and terms, we inferred disease-related genes and constructed a disease-specific gene network. To validate our approach, we investigated the top k (k = 20) inferred genes for prostate cancer and analyzed the constructed gene network using three network analysis measures. Our approach found more disease-gene relationships than comparable method, and inferred describable disease-gene relationships.