The rise of the quality of life index together with the improvement of medical technology lead to a longer life expectancy. Thus a better Health Diet Recommendation Service (HDRS), especially for elderly, is needed. However, to date, there are only a few Decision Support Systems (DSS) to provide HDRS for Dietary Therapy (DT) according to user's diseases and retrieve the diet limitations.For this reasoning, we propose the Dynamic Ontology (DO) which includes Medical Ontology (MO) and Food Therapy Ontology (FTO) to build the HDRS. For ontology description and building, we refer ICD (International Classification of Diseases) and dietitian's recommendation to define and classify the diseases into MO and the foods into FTO, respectively. Moreover, we propose a curative food recommendation method, the Dietary Therapy Recommendation Mechanism (DTRM), which combines DO, Term Frequency-Inverse Document Frequency (TF-IDF), Latent semantic analysis (LSA), and Self-Organizing Map (SOM) for DT to provide the HDRS. The DTRM considers the user's physiology state and diet preference to infer user's diseases and retrieve the diet limitations according to DO. Afterward, The DTRM infers the optimum food collocation to provide relevant HDRS to user.In this paper, we design two test cases using Chinese food therapy to evaluate the DTRM. The Case 1 considers the ”soup class” to provide the HDRS by DTRM, and the Case 2 considers the ”meat class” for DT. The experimental results show that the recommendation precisions of DO-based DTRM and Static Ontology (SO)-based diet recommendation are 75.00% and 46.88% in Case 1. The recommendation precisions in Case 2 with DO and SO are 71.86% and 37.50%, respectively. Therefore, the DTRM based on DO is better than SO in both cases for DT.