A hybrid recommendation model based on the label propagation and VSM clustering
Kai Lei, Kun Zhang, Yanchao Xiang, Wenming Wang · 2012
Recommendation systems try to dig out the most relevant data items according to users' interests by means of data mining and machine learning. Currently, content-based recommendation, collaborative filtering, knowledge-based recommendation are most widely used. However, it is difficult to just use one of them to solve all the problems like cold start, data sparseness, over-fitting etc. together. A hybrid recommendation model based on label propagation and VSM clustering is presented in this paper, which can avoid bias caused by a single algorithm and improve the recommendation system's validity, usability, and portability. After implementing and deploying our model in Maze system [1], we were pleased to discover some rules on how the thermal diffusion model and probability diffusion model could affect the quality of recommendation results and proved that our hybrid model can improve result precision by 47%.