Information filtering based on eliminating redundant diffusion and compensating balance
| Autor: | Xin Su, Xiang-Chun Liu, Jinming Ma, Yu-Xiao Zhu, Xuzhen Zhu, Hui Tian |
|---|---|
| Rok vydání: | 2019 |
| Předmět: | |
| Zdroj: | International Journal of Modern Physics B. 33:1950129 |
| ISSN: | 1793-6578 0217-9792 |
| DOI: | 10.1142/s0217979219501297 |
| Popis: | In statistical physics, researchers concentrate on mass diffusion and heat conduction-based information filtering models, which effectively facilitate recommendation accuracy and diversity. There are many improved methods combining mass diffusion with heat conduction theories. Research results show that the best results are achieved when the combination of mass diffusion and heat conduction reaches equilibrium. With elaborative analysis, we find that similarity redundancies derive from the attribute correlations of objects, and deduce the similarity estimation deviation. Considering the former deficiencies, we propose a novel model through eliminating redundant diffusion and compensating balance (shortly ERD-CB), which symmetrically combines mass diffusion with heat conduction process through balance compensation. Three benchmark datasets from Movielens, Amazon and Netflix are used in our extensive experiments. Experiment results show that the ERD-CB model outperforms the benchmarkbaselines for accuracy, diversity and novelty. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|