Research On Deep Learning Recommendation Model Based On Users' Temporal Behavior

With the continuous development of the Internet and information technology,various kinds of information are confusing and chaotic,showing an explosive growth trend,and the problem of information overload is serious.As a bridge between information producers and consumers,the personalized recommendation system aims to alleviate the dilemma of information overload,help users to discover valuable information for themselves,and push information to users in a timely and accurate manner,thus realizing a win-win situation between information producers and consumers.Some recent studies have shown that deep learning techniques are used in recommendation tasks and also demonstrate their powerful capabilities.Compared with the traditional recommendation algorithm,deep learning uses deep nonlinear structure,which can better extract the feature representation of users and items,and learn the interaction between users and itemsThis thesis focuses on the time series on interaction between users and items,and performs the recommendation tasks through the time sequence of the interaction between the user and the items.For the first time,a hybrid recommendation model based on deep learning is constructed,which effectively integrates recurrent neural network,convolutional neural network and self-attention mechanism,and is applied to the Top-N recommendation task of movies.The advantage of this model is that,it reduces the stacked serial layers structure,increases the parallel layers.It can simultaneously acquire the users' long-term and short-term interest states at different moments,and weighted sum the state information of each moment to generate the final recommendation resultSubsequently,this thesis designs and implements a personalized movie recommendation system,which combines deep learning model recommendation and content-based recommendation.It can dynamically recommend movies for users according to their viewing history,and solves the problem of cold start for new users and new movies.At the same time,the system can dynamically update the database and recommendation model,which is more in line with the actual task needsFinally,this thesis carries out experiments on MovieLens dataset,and compares the proposed model with the traditional collaborative filtering method and the deep learning recommendation model proposed in recent years.The experimental results show that the proposed model has a certain improvement in various indicators,and achieves a good recommendation effect.At the same time,this thesis demonstrates the recommendation effect of the designed recommendation system,and the effect has reached the requirements for practical application.