CF框架下物质扩散算法改进策略研究.docxVIP

协同过滤框架下物质扩散算法的改进 2014年11月30日 Recommender Systems 1 2 References:  Collaborative Filtering Content-based methods Network-based methods Hybrid methods Similarity-based methods Diffusion-based methods Dimensionality reduction techniques Social filtering Meta approaches 刘建国,周涛,汪秉宏. 个性化推荐系统研究进展[J]. 自然科学进展, 2009, 19(1):1-15. Linyuan Lü, Matú? Medo,et al, Recommender systems, Physics Reports, Volume 519, Issue 1, October 2012, Pages 1-49, ISSN 0370-1573. Diffusion vs. Collaborative Filtering Mass Diffusion Collaborative Filtering MD is a kind of ICF m 1 m w?? ? 1 ? al? al? s??Salton ? ?al? al? 1 kl k ? l ?1 ? k? k? ?2 l ?1 n n a m a a f i? ? ? ai ? w?? ? ? i ? ? l? l? kl ? ?1 ? ?1 k ? l ?1 n n ai? m fi? ? ? ai ? s??salton ? ? ?al? al? ? k? k? ? 1 ? ?1 ? ?1 l ?1 2 Key factors of item similarity Similarity of item α and β depends on kα and kβ Number of common users of α and β Interest diversity of common user CF usually captures the former two factors. MD captures all the three factors. Diffusion-like methods 1 m al? al? ? MD-Mass Diffusion: w?? ? ? k ? kl l ?1 1 m ? HC-Heat Conduction: w?? ? ? al? al? k? m l ?1 kl 1 al? al? ? HHP: w?? ? ? ? k ? 1?? k? l ?1 kl 1 m ? Biased HC: w?? ? ? al? al? ? k? l ?1 kl 1 m a a k ? ? Preferential Diffusion: w?? ? ? l? l? ? n k? l ?1 ?r ?1 alr kr ? Part One: Information filtering via balanced diffusion on bipartite networks Balanced Diffusion [1] 1 m al? al? w?? ? ? k? a k ? b kl l ?1 1 m al? al? w?? ? ? k? ? k ? ? kl l ?1 References: D.-C. Nie, Y.-H. An, Q. Dong, Y. Fu, T. Zhou, Information filtering via balanced diffusion on bipartite networks, Physica A (2014), /10.1016/j.physa.2014.11.018 Optimal parameter λ Comparison of methods Degree of items in recommending lists Part Two: Information Filtering via strengthening the correlation of small degree objects m ? m a a ?? cs?? ? ? al? al? cs?? ? ? ? l? l? ? l l ?1kl ? l 1 k ? ? Distribution of Correlation between Objects MovieLens Netflix p(c)  100 10?2 10?4 10?6 10?8 10?