A Research Of Classifying The Compatibility Between Foods

With the development of computer technology and artificial intelligence technology,more and more fields begin to apply the technology of machine learning.Among them,traditional Chinese medicine proposed the theory of that medicine equals food and there is compatibility between foods.In this paper,the machine learning algorithm will be applied to this field,we collect some information on the corpus on internet in order to develop a model for the classification of the compatibility between food and food.In principle,the classification of compatibility between food is also a kind of knowledge representation method.The rise in recent years,a lot of knowledge representation learning method,one of the most famous new and widely used is the transE algorithm and its derivative algorithm,transE algorithm and its derivative algorithm but the principle is not suitable for processing a kind of relationship is reflexive,while compatibility between food is just a reflexive relationship.A and B are compatible means B compatible A.In order to solve this problem,this paper presents a new feature relation representation method,and puts forward two methods of data augmentation for this feature,we can make these data augmentation methods to produce more relation representation on the same pair of food.In training stage,we can through the extended data set to get more samples to improve the classification accuracy;in the prediction stage,we can use the same food to generate multiple feature representation then obtain multiple answers and voting among them,so as to improve the accuracy of the classifier.Because the studies are very rare,so there is no public dataset on internet now.The word embedding model used in the paper and the dataset of compatibility between foods used by the algorithm was built by our own,generating the dataset of compatibility between foods is by fusing the data from crawling multiple sites through the network.The corpus we using to build word embedding is from food in wikipedia and health blog,etc.We devise experiments and do it on the dataset in order to find out the influence of two data augmentation methods and which method is better.And study the relationship of accuracy with dim of word embedding,different classifier model,etc.After all,we get many classifiers and try to aggregate them together,we devise three method of aggregation and make things better.