Research On Personalized Recommendation System Based On Word Embedding

The World Wide Web allows users to access various services handly such as e-commerce,social networking,online lending and online work platforms.Personalized recommender systems are necessary to consumers and suppliers.With the emergence of new products and the change of consumer behavior,personalized recommender systems can learn new patterns constantly to meet the supply demand of consumers and suppliers.Therefore,it is necessary for the research on personalized recommendation system based on word embedding.In this thesis,the corresponding recommendation algorithms are proposed for different scenarios and different data sets.The main research works are as follows:1)This thesis proposes an improved collaborative filtering algorithm with fusion segment vector.There is a disadvantage in collaborative filtering algorithm.It is not able to provide an explanation for the results of the recommendation.The text information is important to the recommendation system.Therefore,the word embedding algorithm is used to train the text information to obtain the distributed representation of the user and the product.It can reflect the similarity and the semantic information of the text.By combining the two algorithms,the disadvantage of collaborative filtering can be solved.2)This thesis proposes a hybrid recommendation technique optimized by dimension reduction.The proposed algorithm combines principal component analysis algorithm and collaborative filtering algorithm to improve the accuracy of recommendation.In collaborative filtering algorithm,the root-mean-square-difference is used to compare the recommendation accuracy,and the change of recommendation accuracy under different measurement is analyzed.The relevance dimension is reduced by using principal component analysis algorithm,and then it is weighted with the similarity of the collaborative filtering algorithm to improve the recommendation accuracy.In the experiment,the time complexity and the change of recommendation accuracy in different dimensions are analyzed.3)This thesis proposes a recommendation algorithm based on recurrent neural network.Traditional recommendation algorithms have been unable to cope with more and more recommendation scenarios.Therefore,with the development of deep learning technology,the technology has been used to solve the recommendation problem.In thisthesis,the time series data is constructed for each user and commodity according to the feature engineering.The recurrent neural network is used to predict whether the user will buy or not.The experimental results show that the improved collaborative filtering algorithm with fusion segment vector can provide explanation for the recommendation results.Compared with the collaborative filtering algorithm,the proposed hybrid recommendation technique optimized by dimension reduction can improve the recommendation accuracy by 2.85%.Recommendation algorithm based recurrent neural network has the F-Measure of 66%.Therefore,the personalized recommendation system is feasible in this thesis.The problems that remain and the future study plans are introduced at the end of this thesis.