Research On Deep Hybrid Recommendation Model Based On Weighted And Feature Fusion

The recommendation system plays a big role in solving the problem of information redundancy.The researchers tried to combine the deep learning with the recommendation system to improve the recommendation effect to some extent.However,there are still problems such as sparse data and cold start,and insufficient representation of potential features.This paper combines model fusion and feature fusion from two perspectives.Combining deep neural network with hybrid recommendation algorithm,using two methods of weighted and feature fusion to construct a deep hybrid recommendation model,this paper proposes a problem for the current recommendation system.The corresponding solution.Firstly,the commonly used recommendation algorithms are described.The principle and corresponding advantages and disadvantages of the traditional matrix decomposition algorithm are introduced in detail.The defects of the recommendation algorithm based on deep neural network are analyzed,and the corresponding improvement strategies are proposed.Secondly,for the data sparse and cold start problems,a weighted deep hybrid recommendation model is constructed.A content-based deep recommendation model is built by introducing content features.A scoring feature is formed by filling the high will matrix,and a deep recommendation model based on the score is constructed.The hybrid recommendation method fully combines the advantages of different information,and separately explores the interaction ability of content features and scoring features,and finally implements model fusion in a weighted manner,in order to improve the accuracy of recommendation by improving the feature interaction ability.Thirdly,the deep hybrid recommendation model of feature fusion is constructed for the insufficient representation of potential features.The self-encoder is used to extract the potential content features,and two different feature fusion methods are proposed.The potential hybrid features of users and projects are constructed,and the expressive abilityof content features and scoring features is merged.By analyzing the loss composition of the model,the fusion loss function is proposed to optimize the model as a whole,in order to improve the accuracy of recommendation by improving the feature expression ability.Finally,the hybrid recommendation model based on weighted and feature fusion proposed in this paper is experimentally verified on the real data set.The experimental results show that the model constructed in this paper effectively improves the accuracy of the recommendation results.