Research On Novelty Of Network Recommender System Based Time Information

With the rapid development of information technology and internet computing,Web 2.0 technologies make people act as both consumers and producers of information.Humans enter the age of information overload from information poverty for explosive development of internet information.Recommender systems no longer require users to provide definite requirements.Instead,user preference model is established to provide information satisfying users' interests and needs by analyzing their behavior history.Variety recommender systems are more widely used and applied by almost all large-scale e-commerce systems.Accuracy indicator has been discussed in above 90%of theses concerning recommender systems since they came into existence.Actually,recommendations with high accuracy are useless.Users will be satisfied if results generated by recommender systems are not only beyond users' expectations but also liked by them.Novel recommendation becomes the hot topic in the field in recent years.Novel item is defined from the perspective of user perception,and offline experimental methods and evaluation metrics is designed for user temporal correlation according to the novelty of item.process integration strategy and result integration strategy are adopted for the fusion of unknown n and dissimilarity metrics with traditional algorithms.The experiment result indicated that the result integration strategy is more effective.The novelty recommendation algorithm based on the product life cycle theory is used for the analysis of the temporal variations of the popularity of the item,and a model of item popularity prediction is established,then users' acceptance of the item is determined with the predicted value of the item popularity.How users prefer and accept the item are used as two standards for the selection of alternative sets and two algorithms are designed,which were ER and PP.The experiment result showed that the novelty,accuracy and coverage of ER algorithm are relatively good.Users'initiative to adopt innovation is modeled with the diffusion of innovation theory,making the recommendation system capable of recognizing the possibility of whether users may accept certain kinds of items at a given moment,and the accuracy and novelty of the recommendation system are improved.The novel item also includes differentiation,namely the distance between the recommended items and user preferences.Global clustering and the clustering of the items accepted by users are adopted for the modelling of user preferences,and weighted distance is used to calculate the differentiation of recommended items towards users,reducing the influences on the recommendation system and improving coverage.For users,the novelty and time of items are closely related.First,time-aware traditional algorithms may improve the accuracy of recommended results;secondly,time-aware innovation diffusion algorithm is designed for more accurate calculation of users' initiative to accept the innovations of various items;lastly,the quantity of each type of items accepted by users and the time took to accept were used for the design of the time perceptive differentiation algorithm.The experiment result shows that the time-aware traditional algorithm in this paper may not effectively improve the novelty of recommended results,while time-aware innovation diffusion algorithm may effectively improve the accuracy and novelty of recommended results.While ensuring that the novelty remains unchanged,the time-aware differentiation algorithm simultaneously the coverage and popularity of recommended results,further improving the comprehensive performance of the recommendation system.