Research On Hybrid Recommendation Methodology Integrating User Preference And Context Information

In the era of mobile internet,with the popularization and promotion of 5G,the scale of users and items continues to grow.However,the emergence of recommender systems solves the problems of information filtering for users,and meets their diversified and upgraded needs.In addition,the current way of mobile application interaction has become more and more diversified.While users enjoy services provided by various application,they also leave behind a large amount of complex information.Apparently,these contextual data can directly or indirectly reflect the preference features of user.Nevertheless,the methodologies of recommender systems have achieved certain progress,there are still some limitations.One is that traditional methods usually adopt topic model to deal with review text,however,it is hard to learn the semantic features effectively.Second,the item-user interaction information is very sparse,and the recommendation approaches are poor in interpretation.The third one is that existing methods of recommendation utilizing static and independent ways to extract latent features of items and user,whereas,ignoring the correlation between them.Consequently,it is of great significance to the research of context-based methodologies of recommendation.The main work of the paper includes:(1)In this paper,the context-based methodologies of recommendation are deeply studied and analyzed.In the meanwhile,it pointed out the challenges and limitations of this type of methods.And based on this,raised a suitable solution.(2)Raised a method of deep collaborative filtering fused content-text(DCT),which was based on the contextual information.The methodology utilized two parallel neural networks to process interaction data and textual information separately,for which can get the low-dimensional feature representations.Indeed,the method employed the semantic features learned from reviewed to enhance the final item representation.In addition,a large amount of comparative experiments and hyper-parameter analyzed had done,in order to proves that DCT is availability.(3)Raised an attention-based deep hybrid recommendation method(ADH),which was based on dual attention mechanism,including local-attention and mutual-attention.The methodology utilized the convolutional neural networks to deal with the reviews of items and users respectively.And incorporated attention mechanism to the network for heightening the interpretability of the algorithm itself.Moreover,ADH also fully considered the correlation features,and employed the neural factorization machine to make the rating prediction,which can improve the performance of the recommendation.In addition,plenty of experimental comparative analysis on data of Amazon 5-core had done,which can be proved the validity of ADH.(4)Integrated two methods addressed in this paper,and applied them to the actual mobile scenarios.Then,designed and implemented a prototype application for music service recommendation.