Study Of A Bias Alleviated Personalized Recommendation Algorithm

With the rapid development of computer and internet technology, information hasgreatly increased, which brought about the so-called information overloaded problem. Itbecomes difficult for people to find out what they want from the massive amount ofinformation. The search system has been proposed in order to cope with this problemeffectively. The search system is a type of passive-service system that keywords shouldbe typed in when the system is used. Compared with the search system, therecommender system is oriented to a sort of active-service system, which can providedifferent information to different users without the limitation of the keywords. Therecommender system can dig out the potential interest to make recommendation, bycollecting and analyzing user’s historical behavior record. Generally, a recommendationsystem is composed of user modeling module, object modeling module andrecommendation algorithm module. Among these modules, the algorithm module isessential, which affects the performance of the whole system. In recent years, with thedevelopment of the complex network, study of network-based recommender system hasbecome a hot research topic.In this paper, we mainly study a network-based recommendation algorithm, andanalyze the effects of heterogeneity of recommendation objects on the recommendationperformance. Recommendation bias towards objects has been found to have an impacton personalized recommendation, since objects present heterogeneous characteristics insome network-based recommender systems. In the network-based recommendationalgorithm, user and object are abstracted as nodes in the network. Initial resources areassigned on the object nodes, and the resources can spread in the network. Due to thedirectional characteristic of the resource spreading, the object disseminating resources isnamed as the source object and the object receiving the resource is named as the targetobject. In this article, based on a biased heat conduction recommendation algorithm(BHC), we propose a heterogeneous Heat conduction algorithm (HHC), by furthertaking the heterogeneity of the source objects into account. Tested on three real datasets,the Netfix, RYM and MovieLens, the HHC algorithm is found to present betterrecommendation in both the accuracy and diversity than two benchmark algorithms, i.e.,the original BHC and a hybrid algorithm of heat conduction and mass diffusion (HHM),while not requiring any other accessorial information or parameter. Moreover, the HHCalgorithm also elevates the recommendation accuracy on cold objects, referring to the so-called cold-start problem. Eigenvalue analyses show that, the HHC algorithmeffectively alleviates the recommendation bias towards objects with different level ofpopularity, which is beneficial to solving the accuracy-diversity dilemma.