Algorithm Research And Implementation Of Online Education Platform Recommendation System Based On Graph Computing Technology

Thanks to the rapid development of big data and data mining,online education platform has obvious advantage over traditional education in the Internet area.Especially in the field of personalized education,the recommendation system makes it possible for the realization of individuation education.This thesis applies the graph computing technology to the construction of personalized recommendation system,and turns to develop one recommendation algorithm which is fit for graph computing.One rating score based recommendation system is designed basing on the data of online education platform.Firstly,the cold start and data sparseness problem are discussed in detail,including conception,common solutions and a detailed description of data sparsity solution based on data transformation and student group.The optimized solution will enrich the user evaluation matrix in two steps,taking account of the personalized characteristics while alleviate the problem of data sparsity.Secondly,this thesis gives the module division and functional design,including the UI module,the log module,the data processing module and the recommendation module.The log module takes on the data collecting work,the data to be collected is something like user behavior data relating to the rating behavior and some user attribution data.The major work of data processing module is giving a definition of the data module which is the basis of recommendation module.Then it need to use different processing technology to transform the data according to the type of data source.In the recommendation module,algorithm group is innovatively proposed to deal with the cold start problem in the recommendation system and realize the individualization in the algorithm level.Then several traditional recommendation algorithms are introduced with detailed description of their advantages,disadvantages and fitness for the graph computing technology.Then bipartite recommendation algorithm is chosen to be the core algorithm of this recommendation system because of its controllable performance in time complexity and perfect fitness for the graph calculation model.Then we detailed describe the optimization and improvement of bipartite algorithm,including a optimized random walk strategy,putting the node degree in the influence factors of target selecting,optimizing rating strategy and introducing the user similarity conception of user-based collaborative filtering algorithm to improving the accuracy of the recommendation system.In the final experimental section,the performance of traditional algorithm and bipartite algorithm is compared in some index.The experiment results show that although bipartite algorithm has a slight lack of stability,it has a great advantage in the accuracy of recommendation.Specially,the accuracy and recall are improved partly compared with collaborative filtering algorithm.