Research On Recommendation Algorithms Based On Side Information And Deep Learning

With the continuous development of the Internet,the recommendation algorithm has gradually become the main technical means for major enterprises to solve information overload.From listening to music、watching news、browsing videos to buying goods,the recommendation system is already ubiquitous,and the core of the recommendation system is the recommendation algorithm.The existing recommendation algorithm mainly calculates the similarity based on the user or the item,so as to recommend the user,but this is far from meeting the needs of users.At the same time,with the accumulation of user browsing logs and the improvement of item information,there are more and more auxiliary information related to recommendation.These auxiliary information are collectively called side information.How to effectively combine side information with traditional recommendation algorithm has become the current Research hotspot in the recommended field.The existing research has the following two problems: First,the traditional recommendation algorithm mainly fits the user's rating data,even if some user information or item information is used.Extract shallow user features or item features,and the existing recommendation system contains a lot of structured and unstructured data.If we can mine the deep features of the data well,we can better model users and items,and recommend more representative items for users;the second is that the cold start problem is prominent and the data sparseness is more serious.Collaborative Filtering(CF),a well-known approach in producing recommender systems,has achieved wide use and excellent performance not only in research but also in industry.However,problems related to cold start and data sparsity have caused CF to attract an increasing amount of attention in efforts to solve these problems.Traditional approaches adopt side information to extract effective latent factors but still have some room for growth.Due to the strong characteristic of feature extraction in deep learning,many researchers have employed it with CF to extract effective representations and to enhance its performance in rating prediction.Based on this previous work,we propose a probabilistic model that combines a stacked denoising autoencoder and a convolutional neural network together with auxiliary side information(i.e,both from users and items)to extract users and items' latent factors,respectively.Extensive experiments for four datasets demonstrate that our proposed model outperforms other traditional approaches and deep learning models making it state of the art.The main research results of this paper are as follows:(1)Digging deep into the potential features of side information,and proposed a method of dividing and treating different side information.The traditional recommendation algorithm mainly recalls around the similarity.The mining of side information is not deep enough,and the interaction feature between side information is also the shortcoming of the traditional recommendation algorithm.In this paper,different feature extraction models are designed for different side information,and the potential features of side information are better explored.(2)Combined with the outstanding performance of deep learning in feature extraction,a hybrid depth recommendation algorithm PHD(a Probabilistic model of Hybrid Deep collaborative filtering)is proposed based on side information.The PHD algorithm mainly uses four basic models: stack denoising self-encoder,convolutional neural network,word vector model and probability matrix decomposition.It can well handle user-based side information and item-based side information.Dealing with the absence of both,has certain versatility and practical value.Experiments verify that the coded PHD can alleviate the cold start problem and data sparseness to some extent.At the same time,the paper also optimizes and parallelizes the PHD model,which further accelerates the training and convergence of the PHD model.(3)Based on the PHD algorithm,a film recommendation system with good scalability is designed.The system mainly includes functions such as user login,user rating and user feedback,and collects side information based on users and movies.Experimental verification based on real data sets indicates PHD.The model has a good recommendation effect.This paper first introduces the research background and key research issues of the recommendation algorithm,and gives a brief overview of the recommendation model based on the traditional method and the recommendation model based on deep learning.Then the specific research on the side information and the deep learning recommendation model is carried out.Based on this,a PHD model based on auxiliary stack denoising autoencoder and convolutional neural network is proposed.Then,a complete data experiment is carried out on four real data sets.Finally,we design a relatively complete movie recommender system based on PHD model and look forward to future research work.