Research On Food Collocation And Recommendation Algorithm

With the improvement of people's living standards,people have put forward new requirements for food,not only require the nutritional efficacy and therapeutic effects of the dishes,but also have higher expectation for the new dishes.People expect more innovative dishes.At present,how to create an innovative dish is mostly based on individual's experience and conception of cooking.However,it is difficult to make the combination of nutritious ingredients for ordinary person,because they are not familiar with the knowledge of nutrition.And it is hard to extract the information they need from the Internet because of the mass data of nutritional information.Therefore,how to strengthen the innovation of dishes and speed up the efficiency of innovation for different users with specific efficacy has become a research topic waiting to be solved.In this thesis,the researches on food recommendation and innovation algorithm are summarized.It is found that the current research on food recommendation mainly considered the needs of basic dietary intake of human body and user preferences.But these researches didn't consider the nutritional efficacy and therapeutic effects of the dishes.And there are few studies on the innovation of dishes.To solve these problems,a new innovation frame of dishes is established by collating and analyzing a variety of nutritional indicators,according to the relationship between ingredient and nutrients and the relationship between nutrients and efficacy.A multi-objective optimization genetic algorithm with user preference(MOGA-UP)is proposed to achieve innovation of the dishes.The algorithm includes a default random weight method to represent the user preference,an ingredient selected strategy to control the quantity ingredients,and a new expressive gene cross method to solve the problem of unwanted rapid convergence.The experimental results show that the recommendation have high index of nutrition quality,which means the algorithm is feasible and effective.Compared with NSGA-II algorithm,MOGA-UP has faster convergence speed and higher quality of the recommended results.The feasible solutions of MOGA-UP are well-distributed in the search space.To verify the therapeutic effect of MOGA-UP algorithm,BP neural network is applied to the innovation model of dishes in this thesis.A BP neural network is trained,and the therapeutic efficacy of dishes from MOGA-UP input to the network as the expression vector.It is found that some dishes are not satisfied with the expectation by comparing the gap between the output efficacy and the expected efficacy.To solve this problem,MOGA-UP algorithm is combined with neural network(BP-GA)to ensure the efficacy of innovative dishes is accordant with expected efficacy.The experimental results show that the error rate of the BP-GA recommendation is lower and the similarity is higher,compared with MOGA-UP.It shows that the BP neural network method based on genetic algorithm has higher accuracy,but BP-GA cannot calculate the index of nutrition quality,so the nutritional efficacy cannot be evaluated.Besides,BP-GA relies on high quality data with large amount and multi-objective problems.