Research On The Diversity And Novelty Of Personalized Recommendation Systems

In recent years,with the rapid development of mobile Internet applications and the popularization of Web 2.0 applications,personalized recommendation systems have played an increasingly important role in helping individual users mine interesting content and helping companies do product promotion.Personalized recommendation systems are widely used in almost all aspects of people's lives,such as finance,news,social network-ing,tourism,education,etc.At present,more and more research scholars are devoted to study how to optimize the accuracy,diversity,novelty and other indicators of the recom-mendation algorithm of the recommendation system to meet the increasing personalized needs of users.In the practical application-oriented recommendation system,an important research branch is related to the recommendation algorithm based on the bipartite graph network structure for how the recommendation system meets the research goals of accuracy,di-versity and novelty at the same time.This kind of algorithm has aroused strong research interest of research scholars,but there are still some problems:(1)As far as optimization goals are concerned,most of the research on recommendation algorithms focuses on im-proving the accuracy of the algorithm,and other quality of the recommendations,such as diversity or novelty,are more or less ignored?(2)The accuracy and diversity(and nov-elty)of the personalized recommendation system have the problem of ‘trade-off'?(3)The performance on accuraty and diversity of the algorithms based on the network structure are limited,and the advantages of diversity and novelty of other algorithms need to be in-tegrated.To deal with these problems,we further analyze the important role of user activ-ity and item popularity in the recommendation algorithm,and conduct an in-depth study on the accuracy,diversity and novelty of the diffusion-like recommendation algorithm based on the bipartite graph network structure.In addition,this thesis also innovatively applies the recommendation system to the field of online education,and studies how to recommend diverse online learning resources to users.In summary,the main research innovations of this thesis are as follows:(1)We proposed to enhance the Resource-Allocation(RA)similarity in resource transfer equations of diffusion-like models,by giving a tunable exponent to the RA sim-ilarity,and traversing the value of this exponent to achieve the optimal recommendation results.In this way,we can increase the recommendation scores(allocated resource)of many unpopular objects.Experiments on three benchmark datasets,MovieLens,Netflix and RYM show that the modified models can yield remarkable performance improvement compared with the original ones.(2)We introduced a family of approaches to extract fabricated experts from users in RSes,named as the Expert Tracking Approaches(ExTrA for short),and explored the capability of these fabricated experts in improving the recommendation diversity,by high-lighting them in a well-known bipartite network-based method,called the Mass Diffusion(MD for short)model.These ExTrA-based models are compared with two state-of-the-art MD-improved models HHP and BHC,with respect to recommendation accuracy and di-versity.Comprehensive empirical results on three real-world datasets MovieLens,Netflix and RYM show that,our proposed ExTrA-based models can achieve significant diversity gain while maintain comparable level of recommendation accuracy.(3)We proposed a hybrid algorithm by using the diversity of item-based collabora-tive filtering(ItemCF)method and the accuracy of the mass diffusion(MD)algorithm(referred to as the HI-series algorithm).When studying the diffusion-like algorithms,we found that the most basic mass diffusion algorithm has better accuracy,but lower diversity performance? and item-based collaborative filtering algorithm has excellent diversity,but the accuracy performance is poor.Based on this analysis,we get accurate and diverse recommendation results by mixing the characteristics of these two algorithms.We call this algorithm Hybrid of ItemCF and MD(HI-MD).The ItemCF is further applied to the diffusion-like algorithms HHP,BHC and BD.The experimental results on the real data sets MovieLens,Netflix and RYM data show that the HI-series algorithms we proposed can not only improve the diversity,but also guarantee the accuracy,and also are applicable to sparse data sets or dense data sets,keeping a certain degree of robustness.(4)Besides,in addition to the research on the above commonly used data sets,we innovatively applied the personalized recommendation system to the field of online ed-ucation,and designed a learning resource recommendation system for Massive Online Open Courses(MOOC)platforms(referred to as Wikification system).In the background of the practicality of the recommendation system in real life and the era of on-demand online learning,this thesis attempts to automatically extract the user's key phrases about learning resources in the discussion forum of the MOOC platform and analyze the user's demand for learning resources,thus recommend diverse learning resources to users ac-curately.The data of this work comes from the real MOOC platform—Coursera.After collection,processing,manual annotation and application to the recommendation system,the recommendation results show the accuracy and feasibility of the Wikification system.