Research And Practice Of Recommendation Algorithm Based On Convolutional Neural Network

With the advent of the era of big data,the amount of information has rapidly increased exponentially.In order to better obtain information,recommender systems emerge as the times require.There are some problems in traditional recommendation algorithms,such as cold start,data sparse,and user interest changes,which need to be solved.Convolutional neural networks play a prominent role in feature processing.Therefore,the combination of convolutional neural networks and recommendation systems has become a field in this field.major trend.This paper designs a recommendation algorithm under the framework of neural network,relying on the advantages of convolutional network in feature extraction,and solves several problems existing in the above traditional recommendation algorithm.The specific work and contributions of this paper are as follows:(1)Combining the text convolutional neural network framework with the recommendation algorithm,a recommendation algorithm based on the textual convolutional neural network framework is designed.A large number of comparative experiments have been carried out on the Movielens dataset.Compared with the traditional recommendation algorithm,the accuracy and time complexity have been significantly improved.(2)This paper adds a self-attention module to the text convolutional neural network,which further enhances the network's ability to extract features and strengthens the correlation between different features.Based on this network framework,an improved CNN-A algorithm is designed,which has a significant improvement in recommendation accuracy.After comparative experiments on the Movielens dataset,the RMSE results of the CNN-A algorithm are 2.8%/2.2% higher than CNN,4.1%/3.7% higher than the traditional SVD,and 8.3% higher than the recently proposed ALS-CF and CFGAN algorithms,respectively.%/7.9% and 5.3%/4.7%..(3)A CNN-TA algorithm integrating adaptive time coefficients is designed.Since the user's interest in an item will change with time,the above CNN-A algorithm can only use the user's historical data for historical preference recommendation.To solve this problem,an adaptive time weight is designed in the CNN-TA algorithm.coefficient to meet the task requirements of online recommendation.Through comparative experiments on Movielens-100 K and Douban datasets,the HR value of the CNN-TA algorithm proposed in this paper is increased to 0.73,and the DCG value is increased to 0.53.(4)Based on the above work,this paper designs a movie recommendation system that meets the user's personalized recommendation needs.Through the analysis of functional and non-functional requirements,a recommendation system is designed,which uses CNN-A algorithm for offline recommendation and CNN-TA algorithm for online recommendation.