Study On Recommendation Algorithm Based On Collaborative Filtering And Deep Learning

With the rapid development of the Internet,people can easily obtain the information resources what they want,but at the same time it also brings the problem of information overload.Recommendation system is an important way to solve the problem of information overload.Today,the recommendation system is applied to many scenes such as our entertainment,social network,and online shopping.But,in the actual experience,it still has many problems such as the system's “cold start” and data sparseness.This paper improves the recommendation accuracy of the collaborative filtering algorithm by improving the Pearson correlation coefficient similarity calculation formula.In addition,deep learning is applied to the recommendation system,which solved the problems in the recommendation system such "cold boot" and data sparseness.This paper mainly did the following work.This paper focuses on the recommendation principle of collaborative filtering algorithm,and makes a detailed analysis of its recommended process.In the user-based collaborative filtering,the cosine similarity and Pearson correlation coefficient were used respectively to implement the algorithm on real data sets.Based on the same data sets,the item-based collaborative filtering was implemented by Pearson correlation coefficient.Through the analysis of the recommendation results,it is found that the collaborative filtering algorithm does not consider the influence of the number of items jointly evaluated on the recommendation result,and the recommendation accuracy still has room for improvement.Therefore,the Pearson correlation coefficient is improved in this paper by introducing the penalty factor,and the number of jointly evaluated items is taken into account.The similarity is reduced by using the penalty factor for users or commodities that have less evaluation items.Through experimental comparison,it is verified that the improved Pearson correlation coefficient formula can obviously improve the recommendation accuracy.In order to solve the problems of the recommendation system such as "cold boot",and make the recommendation system more intelligent,the deep learning method is introduced.Firstly,the VGG16 model of deep learning is used to extract features from the grabbed clothes images,and the extracted features are used to calculate the similarity between the images for making a recommendation.Through the analysis of the recommendation results,it is found that there is room for improvement in the recommendatory accuracy which calculate by using the image features extracted by the model.In order to make the features extracted by the model more specific,a convolutional neural network model is established and trained by the grabbed data which was classified.Experimental analysis is performed on the factors that influence the model classification results,and the optimal model weights are preserved.The obtained weights is used to extract features from the clothes images,and the extracted features were used for similarity calculation,and the similar clothes images is found for recommendation.By analyzing the results of the recommendation and comparing it with the recommended results calculated by the features which is extracted by VGG16 model,it is verified that the extracted features by the established model can make a better recommendation.