Research On Sequential Recommendation Model Based On Attention Mechansim And Deep Learning

Due to the explosive growth of information,people’s demand for the effectiveness and accuracy of information acquisition is also getting higher and higher.Recommendation systems are one of the ways to help people find the resources they need from massive information.The sequential recommendation model simulates the evolution trend of user interest based on user historical behavior information,which is one of the popular research directions in the field of recommendation systems,but there are still the following shortcomings that need to be improved.For example,only modeling the interaction between users and items ignores the influence of the popularity of items on user behavior;the user’s behavior sequence is random and uncertain,and it is difficult to efficiently capture the dependencies existing in the user behavior sequence;it is difficult to segment the problem of characterizing users’ long-term preferences in the obtained fixed-length behavior subsequences.In view of the problems above,the main research contents of this thesis include:1)Aiming at the problem that the sequence recommendation algorithm only models the user’s historical behavior sequence and cannot recommend some popular items for the user,a sequence recommendation algorithm based on the Item Popularity Evolution Network(IPEN)is proposed.The network introduces various attributes of users and items to generate Embedding vectors to improve the expression effect of the vectors;it uses the self-attention mechanism to model the user sequence interacting with the item to be recommended,extracts the popularity vector of the item,and taps the potential interest of the target user.The comparative experimental results show that the fusion of item popularity information and user interest can improve the accuracy and real-time performance of the algorithm.2)Aiming at the problem that the recurrent neural network model defaults to the interdependence between adjacent user behaviors,which makes the user behavior sequence prone to wrong dependencies,a sequence recommendation algorithm based on User Interest Evolution Network(UIEN)is proposed.The network uses multiple convolution kernels of different scales to convolve the user behavior sequence to obtain a user interest vector sequence containing multiple cascade relationships,and then input it into the Gated Recurrent Unit(GRU)network to capture user interest.changing trend.The experimental results show that the algorithm reduces the probability of generating incorrect behavior dependencies and improves the recommendation effect of the algorithm.3)Aiming at the problem that most sequence recommendation models only model behavior sub-sequences of fixed length,and it is difficult to describe users’ long-term interests,a sequence recommendation based on Search User-Long-Interest Evolution Network(SUEN)is proposed.algorithm.The algorithm first filters the long user behavior sequence and finds the behavior subsequence with the highest correlation with the item to be recommended;then models the user behavior subsequence to extract the user’s long-term interest.The experimental results show that the algorithm can effectively deal with long sequence data.Finally,this paper proposes a hybrid recommendation model that integrates item popularity and users’ long-term interests.Experiments show that the hybrid model has a good recommendation effect.