Research On Recommendation Algorithm Based On Deep Learning

The popularization of the Internet has promoted the rapid development of network communication,online shopping,online movie viewing,online books and other fields.As a kind of network books,electronic novels are not only of various kinds and large amount,but also of fast updating and wide audience.It is often difficult for users to quickly obtain interesting and high-quality content from a large number of online novels.Therefore,for a novel website,if a list of books that matches the user's interests and preferences can be filtered for each user from the mass of books,it will not only save user's time and improve user experience,but also increase user stickiness for novel websites and increase website revenue.It can be described as a win-win process.The recommendation algorithm has been researched and applied in many fields,but the research in the application of e-novels is still scarce.At the same time,with the rapid growth of the number of users and novels,the requirements for novel recommendation algorithms are getting higher and higher.Existing recommendation algorithms rarely take into account the characteristics of online novels and their users,and they are deficient in terms of accuracy,computational efficiency,and processing of sparse data.Deep learning technology has developed rapidly in recent years,and it is also a trend to introduce deep learning into recommendation algorithms.The existing deep learning-based recommendation algorithms still have a lot of research space,such as how to use user and item information from multiple perspectives,and how to improve interpretability.In view of the above analysis,the research work of this paper is mainly focused on the following points:1.For the problem of increasing sparse rating data,a method of using review information to make up for sparse rating is proposed.In this paper,the hybrid recommendation algorithm combining the deep learning algorithm and the PMF algorithm is applied to the e-novel recommendation,so as to overcome the shortcoming of single algorithm.Among them,the deep learning algorithm uses user and novel review information to extract user review features and novel review feature vectors,and the PMF algorithm uses rating information to extract the potential feature vectors ofusers and novels through maximum posterior estimation optimization.Finally,the user and the novel's potential characteristics and comment characteristics are fitted to the rating score.2.In view of the uneven quality of user's reviews,we putting forward that high-quality reviews can help improve the recommendation effect,while low-quality reviews are meaningless and even weaken the effect of the recommendation.In order to verify the impact of review quality on the performance of novel recommendation algorithms,this paper applies the attention mechanism to convolutional neural networks.Through the attention mechanism,the weights of different reviews are learned,the guesses are verified,the review information is more effectively used,and the accuracy and interpretability of the model are improved.3.In the experimental,the Kindle?Store?5 public dataset and the Douban novel dataset were used.First,the Douban novel data was crawled as the real dataset,and then three typical recommendation algorithms,PMF,Conv MF,and deep Co NN,were selected to compare with the algorithm in this paper.The experimental results show that the overall performance of the proposed algorithm is better than other algorithms,and it also performs relatively well on sparse data sets.The experiment also compares the model with and without the attention mechanism,thereby verifying the impact of the quality of the review information.The corresponding weight of comments calculated by the model verifies the interpretability of the algorithm in this paper.4.Based on the Spark big data computing platform and the algorithm proposed in this article,a novel recommendation system was designed and implemented,including daily recommendation,popular recommendation,online recommendation and user retrieval functions.The system function tests and performance tests shows that the system basically meets the preset requirements and has a certain practical application value.