The purpose is to design novel methods for exploring food- and nutrition-related data and knowledge to improve public health. The paper consists of three problems. The first one is how to obtain more robust results from statistical analysis of nutritional experimental data, especially in the case when one parameter is measured by different methods and needs to be compared over multiple samples, and each sample is measured several times. The second one is to develop rule-based NER method for an untapped domain without using of an annotated corpus, where the rules will not be associated to the characteristics of the entities of interest and its application in the dietary domain. The third one is focused on text normalization in order to improve text similarity measures of domain specific short text segments. Its application is on linking the extracted food intake to nutrient intake or reverse.