Research And Design Of Collaborative Filtering Recommendation Engine Based Of Similarity Modeling And SVD Optimization

Nowadays,Machine Learning has become a very hot subject.It is also the core of Artificial Intelligence.Machine Learning is a subject across multiple fields.It's research topic has penetrated into all aspects of human life and production.The thesis applied the machine learning in the industry.Starting from dishes recommended problems,this thesis analyzed the relevant technology and theory and discussed the advantages,difficulties and feasibility of constructing similarity model of items at the present stage.This thesis analyzed the characteristics of the proposed model for the realization of the recommendation algorithm.Based on the similarity of items,this thesis designed a numerical modeling abstraction modeling process that can be quantified.In addition,the thesis gave examples for the simplified method of the model.Finally,this thesis implemented the SVD(Singular Value Decomposition)recommendation algorithm in Python language.Three similarity calculation methods were used to verify the results.The main contents of this thesis include:1.Design the modeling theory based on the similarity of items.According to the actual demand of food recommendation,a set of modeling theories based on similarity of items are designed,and some dishes are modeled.2.We made dimension-reduction simplification of the food model to meet the needs of data computing based on user similarity recommendation algorithm at this stage.3.We studied and designed the optimized collaborative filtering recommendation engine based on SVD algorithm,and made use of SVD recommendation algorithm to convert food recommendation problem into the mathematical model problem.4.The similarity and cosine similarity are calculated by Euclidean distance and Pearson correlation coefficient.The recommended results are analyzed and the recommendation algorithm is implemented successfully.The recommendation engine is facing the problem of the cold-start.The solution to the cold-start problem is to treat the recommendation as a search problem,at the same time,we can also compute the similarity with the attribute data,which is called content-based recommendation.The effect of content-based recommendation may not be better than the effect of recommendation based on collaborative filtering,but content-based recommendation can reflect more problems,which will be the direction of my next step.