Research On E-Commerce Recommendation System Based On Hybrid Recommendation Algorithm

With the rapid development of Internet technology,the number of Internet users worldwide has increased rapidly,and e-commerce has become more and more popular.Consumers tend to shop online more than physical stores.However,while online shopping brings convenience to people's lives,it is followed by an explosion of information.How users select their desired products from a wide variety of products has become the current e-commerce website.The primary problem that needs to be addressed in raising sales is that search engines perform well on this issue.However,this search is based on the user's initiative.When the user is not clear about their needs,this search engine completely loses its effect.At this time,the role of the recommendation system is reflected.The recommendation system is considered to be a more personalized solution.Therefore,the recommendation system is becoming more and more important in modern electronic commerce systems.In order to meet the user's need for passive acceptance of recommendations,it is increasingly important to study recommender systems in order to provide users with the principle of providing quality and superior services.This dissertation makes in-depth research on user's characteristic information,user interaction behavior,product's characteristic information,recommendation engine,and related technologies in data mining.The "large data volume","cold start",and "data sparseness" existing in the recommendation system of modern business websites are analyzed.In response to the first two problems,this paper first reduces the problem of large data volume into a large user group problem and then uses the k-means clustering algorithm to reduce the large user groups into homogeneous user groups.In response to the latter,this paper proposes This problem is mitigated by a combination of a collaborative filtering algorithm and a content-based recommendation algorithm within a homogeneous user group.This thesis mainly studies the following aspects:1.In-depth study of the principle of the recommendation system,the recommendation system is divided into information preprocessing module andrecommendation engine module to study separately.Then use the idea of ? ?demographic recommendation algorithm to model the user.Then,by studying the interaction between users and shopping mall websites,we use weighted method to convert user behavior into preference score.At the commodity modeling stage,we studied the related technologies in the field of information retrieval and used the vector space model to model the commodities.2.In the research of the recommendation engine of this paper,first use the k-means clustering algorithm to divide users into small user groups with similar attributes according to the user model,which alleviates the problem of “large data volume”,and then uses the intra-group user ratings to rank Ways to ease the "cold start" problem.3.A detailed study of the collaborative filtering algorithm and content-based recommendation algorithm.The difference between the real similarity and the calculated similarity when using different similarity calculation schemes in the two algorithms is compared by an example,and the respective optimal similarity calculation schemes are selected for the two algorithms.The reasons for the "data sparseness" affecting the accuracy rate were analyzed.Based on this,a combination of two kinds of collaborative filtering algorithm and content-based recommendation algorithm is put forward.4.In the experimental stage of this paper,the first experiment was conducted with the accuracy and recall rate as indicators,and the optimal weights for the overall combination of the two algorithms in scheme one were determined.In the second experiment,six kinds of algorithms use the average absolute error as the evaluation index,which proves that the two combinations proposed in this thesis can improve the recommendation accuracy of the recommendation engine.It also proves from the aspect that the method proposed in this paper for the "cold start" homogenous group score order can alleviate the "cold start" problem.