Design And Implementation Of Recommender System Based On Hybrid Recommendation Technology

With the explosive development of Internet in recent years,massive information and its continued growth make the information heavily overloaded,which results in a significant reduction in the utilization of information,and the challenges in accessing to valuable information.Based on this,recommender system as a separate branch has attracted the attention of academia and industry.Recommender system can effectively solve the problem of users' selection caused by information overload.Unlike the search engine,recommender system pushes the potentially valuable information to user actively based on an analysis of the user's historical data.Traditional recommender system is mostly based on a single data source and based on a single or simple mix of strategies,which can't solve the problems in sparse matrix,cold start,recommended real-time and so on.And in this big data era,recommender system is always built on platform with large-scale data,thus combining with big data technology then building a stable and efficient recommender system has been a hot topic in recommended research.In this paper,a hybrid recommendation method has been designed based on the study of traditional recommendation algorithms,the method takes into account multiple data sources,which include user information,item attribute,explicit or implicit rating,through this method it can fix the problems like sparse matrix and cold start in recommender system.Then,in this paper a recommender system has been built and implemented on Hadoop and Storm based on the hybrid recommendation method,and it generates recommended list by cross-mixing Hadoop's and Storm's results.Also,user's behavior has been taken into consideration to generate on-line recommendation based on Storm's sliding window,which can greatly improve real-time in recommendation.And in this paper,several related technologies have been researched and used in the recommender system like Flume,and Kafka.Experiments show that the hybrid recommendation method proposed in this paper effects better comparing to traditional recommendation algorithms,and combining with on-line recommendation,the hybrid recommender system can better meet user's real-time needs.