Research And Implementation Of Recommendation Algorithm Based On Deep Learning And Implicit Semantic Model

With the popularization of information network and its application,the amount of information in the network environment is increasing,and people have entered the information overload era from the information shortage era.Against information overload,sear-ch engines often requires the user to enter a valid keyword to get more information,but the average user is often difficult to give accurate keywords,recommendation system solves the users'search keyword uncertainty,thus recommendation system become the present stage people to obtain useful information from the information in the ocean way again and again and again.Through the analysis of users'historical behavior records(such as rating,collecting,purchasing and browsing),the recommendation system judges users'preferences and pushes the information they are interested in to users.This paper studies the current application of deep learning in the recommendation system,analyzes the common recommendation algorithms in the recommendation system,and studies the existing problems of these algorithms.The application of Restricted Boltzmann Machine in deep learning algorithm in recommendation system is analyzed and studied.Deep learning,as a method of modeling with multi-layer neural network,is gradually emerging in the recommendation system.However,too many parameters of deep learning and too long training time are always problems that need to be solved in the development of deep learning itself.Main work of this article:(1)introducing the conditions Restricted Boltzmann Machine,for deep learning parameters,training time is too long,too much study a condition factor Deep Belief Network model,the model is based on Restricted Boltzmann Machine,and get inspired by lingo righteousness in the model makes the condition factor,Deep Belief Networks based on weighting matrix decomposition,decreasing the number of parameters and improve the convergence of the algorithm efficiency.Experiments show the feasibility of this method.(2)For data sparsity and long tail problems,this paper also proposes an SVD++ model based on Item Embedding,which improves recommendation coverage by extracting more item features and mining long tail items through item embedding.Experiments show the effectiveness of the algorithm.(3)the improved SVD++ model and conditional factor deep confidence network are mixed for recommendation.This method combines the ability of object feature extraction based on SVD++ model and the role of user feature mining based on Deep Belief Network.Based on algorithm research,experimental research is conducted,and the results show that this method is effective.On the basis of the above research,the corresponding prototype system should be designed.