This paper proposes a strategic model for recommending recipes called SARDG. The model focuses on dynamically generating knowledge tags from the current World Wide Web structure. It uses a dynamic learning approach with a Restricted Boltzmann Machine to classify the generated knowledge stack. Additionally, a lightweight machine learning Random Forest classifier is used to classify the actual recipe dataset. The model aims to create a limited knowledge graph to keep the system lightweight and relevant. Summarized recipe blogs and E-book indexes are used to build lightweight context trees, keeping them limited in scope. This allows for better computational flexibility. To determine semantic relevance, the model uses two techniques: Normalized Compression Distance and Salient Semantic Analysis. These methods effectively calculate semantic relevance in comparison to other models.