Research And Implementation Of Recommendation System Based On Two Graph Network Structure Andfusion Of Context Aware Information

Due to the rapid development of Internet technology,personalized recommendation system has been the concern of many scholars at home and abroad.At present,many recommendation algorithms have been proposed,such as content-based,collaborative filtering and recommendation algorithm based on two graph network structure,etc.Among them,two graph network structure recommendation algorithm,which is mainly based on the relationship between the user and the item selection,due to not restricted by the type of the object and high hit rate,is more and more widely used.However,there are still some problems in the existing recommendation algorithms.In this paper,we study these problems and make the following improvements:First,we propose a weighted method,which is based on the traditional two graph network structure recommendation algorithm to calculate the weights of the edges of the two points in this paper.Compared with the original algorithm,the weighted recommendation algorithm integrates the user's explicit score into the resource allocation process of the two graph,which make itemsthe user score higher priority recommended.Secondly,the network structure for users and items with the number of users and items on the network is growing more and more complex,which leads to the problem of poor timeliness of operation algorithm,so this paper has introduced the concept of user's nearest neighbor.The algorithm firstly use cosine similarity and modified cosine similarity of two methods to compute the first N nearest neighbor users of the target users,then according to these nearest neighbor userspredict the interested goods of the target users.However,recommendation system needs according to the user's current demand generating different recommendation list display to the user,so the context aware information is starting to get the attention of many scholars,and it has been successfully used in many fields.Since the proposed recommendation algorithm can only according to user's past behavior recommend items to the user,can not represent what users want the moment.So the context information of user's mood is added to therecommendation algorithm,this will recommend items for the user according to the mood of the current user.Finally,through the experiment,verified to the two graph edge weighted,cosine similarity join,modified cosine similarity and integrate into the context information,all the improved algorithms can make accuracy and recall rate and F values of the recommendation system have significantly improved,this also illustrates the research significance and practical value of this topic.With the explosive growth of network data,the traditional recommendation algorithm in stand-alone mode has been unable to adapt to the processing of large data sets.this paper puts forwardparallel computing model based on Hadoop platform to solve this problem,and develops a WEB application on the Hadoop platform,which showing the visual page for recommendation algorithm using experimental results form and improving the user experience effect,and also verifying the feasibility of the recommendation algorithm.