Research On Recommendation Algorithms Based On Convolutional Neural Network

In recent years,more and more data on the Internet can be acquired,multi-source heterogeneous data including images,texts,tags,and scores contain rich user behavior information and personalized demand information.However,because these data usually have the characteristics of diverse sources,uneven distribution,different structures,multi-dimensional and large-scale but valuable information is sparse.Traditional recommendation algorithms are difficult to effectively use these multi-source heterogeneous data and valid extract information from it.Deep learning technology has developed rapidly in recent years,and has achieved good results in various tasks.It has outstanding advantages in the processing of multi-source heterogeneous data.Recommendation algorithms based on deep learning technology have emerged and have been applied in multiple areas,Such as e-commerce,location services,etc.Convolutional neural network is an extremely important research direction in deep learning.It has outstanding performance in image processing and natural language processing,and can accurately extract key information in text and image.In recent years,convolutional neural networks have been successfully applied to recommendation algorithms and have achieved good recommendation results.However,the current research on recommendation algorithms based on convolutional neural networks is only in its infancy.On the one hand,the time when the convolutional neural network is applied to the recommendation algorithm is still short.There are no research results to systematically verify the influencing factors affecting the convolution recommendation model,and what is the relationship between the recommendation effect and the selection of the influencing factors;on the other hand,there are many different techniques for deep learning,and each technology is good at different directions.It can combine the convolutional neural network with other deep learning techniques to build a recommendation model to make up for the inadequacies between different technologies and improve The performance of the proposed algorithm;At last,time information has a huge impact on users and items.A good recommendation algorithm should be able to predict the user preferences and the life cycle of the item,so that the recommendation algorithm has long-term effectiveness.However,the current convolution recommendation algorithm is far from sufficient in the above three aspects,and requires more in-depth and more extensive research in order to achieve better results.This paper starts with the development of the recommendation algorithm.Our system verifies the influence of some influencing factors on the performance of the convolution recommendation algorithm,such as word vector dimension and word frequency.And the same time,we designs the dynamic convolution recommendation model for the shortcomings of current convolution recommendation algorithm on user preference and dynamic trend prediction of item life cycle and improve the performance of the recommended algorithm and make it effective for a long time.In the end,we used the real dataset published by Amazon to conduct experiments.In the experiment,we tested the influence of various influencing factors on the recommendation effect,and got a universally applicable law.The experiment was carried out later,and the dynamics of adding the time model were verified.The convolution recommendation network has better performance than the traditional recommendation algorithm and the static convolution recommendation algorithm.