With the constant-depth research on the relationship between food and disease, relying on diet modification has been proved to make the disease condition improve or even cure. However, most of the research is currently based on how to better concoct recipes while ignoring the recipe expression. This paper first proposed two new recipe expression methods based on 2-tuple food-energy feature and 3-tuple food-energy-relationship feature using the energy value and the grey relational degree, which not only can solve the sparsity and binarization problems of the recipe matrix, but also can reflect the realistic energy relationship of different food in a recipe and the relationship of the same food in different recipes. Then, we used alternating least squares (ALS) for recipe concoction for the first time. Moreover, the non-smooth sparse matrix is introduced for better concocting recipes in the ALS-based model. Experiments on real-life data showed that the two recipe expression methods made 1.6% and 3.3% improvements respectively and the ALS algorithm can be applied well to the concoction of recipes.