Research And Implementation Of Personalized Recommendation System Based On User Behavior

As the capacity,complexity,and dynamics of network information continue to increase,recommendation systems have become a key solution to overcome information overload and are widely adopted by many online services including e-commerce,online news,and social media sites.The main research content of the personalized recommendation system is to model the user's preferences based on the user's past interactions(e.g,comments,ratings,and clicks).In this thesis,based on the personalized recommendation system of user behavior,the corresponding recommendation algorithm is studied.The main research work is as follows:1)A user comment model construction based on improving word2 vec method.The improvement of the skip-gram algorithm in word2 vec is divided into two parts: one is the entity,the other is the non-entity,and the non-entity is added to the negative sample of the target entity.By iterating over each sentence,the entity is embedded in the entity vector space.A user-review model is constructed using an improved word embedding algorithm to obtain a distributed representation of users and commodities,and similar products are recommended from the characterization.2)A recommendation algorithm for user product model fusion is proposed.Firstly,the data is pre-processed and analyzed to extract the characteristics of the user's goods.Secondly,the model is constructed for the high-latency users through user characteristics and user behavior.The user's product pairs are modeled according to the characteristics of the user's product behavior.Finally,the user model and the user product model are combined to predict the future purchase of goods.3)Recommendation algorithm based on neural network.Firstly,the rating matrix is constructed for the interactive user product pairs.Then the constructed features are used as the input layer of the neural network and the full connection layer is used as the training layer.Finally,the output layer is how much the user likes the product.The product with a high score is predicted to be recommended to the user.Experiments show that the improved word vector method can realize the construction of user comment model,which can solve the limitation of collaborative filtering for text recommendation.The recommendation algorithm based on user product model fusion can improve the accuracy of high-latency users.Compared to a single model,the recommended algorithm for model fusion can increase the recommendation accuracy by 0.05% on average;the neural network-based recommendation algorithm can embed the scoring matrix into the representation and improve the recommendation accuracy.The personalized recommendation system proposed in this paper has certain feasibility.