Limits...
Information Filtering via Heterogeneous Diffusion in Online Bipartite Networks.

Zhang FG, Zeng A - PLoS ONE (2015)

Bottom Line: Previous works considered the diffusion process from user to object, and from object to user to be equivalent.We show in this work that it is not the case and we improve the quality of the recommendation by taking into account the asymmetrical nature of this process.Finally, this modification is checked to be able to improve the recommendation in a realistic case.

View Article: PubMed Central - PubMed

Affiliation: School of Information Technology, Jiangxi University of Finance and Economics, Nanchang 330013, P.R. China; Jiangxi Key Laboratory of Data and Knowledge Engineering, Jiangxi University of Finance and Economics, Nanchang, 330013, P. R. China.

ABSTRACT
The rapid expansion of Internet brings us overwhelming online information, which is impossible for an individual to go through all of it. Therefore, recommender systems were created to help people dig through this abundance of information. In networks composed by users and objects, recommender algorithms based on diffusion have been proven to be one of the best performing methods. Previous works considered the diffusion process from user to object, and from object to user to be equivalent. We show in this work that it is not the case and we improve the quality of the recommendation by taking into account the asymmetrical nature of this process. We apply this idea to modify the state-of-the-art recommendation methods. The simulation results show that the new methods can outperform these existing methods in both recommendation accuracy and diversity. Finally, this modification is checked to be able to improve the recommendation in a realistic case.

No MeSH data available.


The (a) Ranking score, (b) Precision, (c) personlization and (d) novelty of the O-Hybrid method as a function of λ in Netflix network.The green lines mark the optimal λ* of the O-Hybrid method and the red lines mark the optimal results of the H-Hybrid method.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4488376&req=5

pone.0129459.g004: The (a) Ranking score, (b) Precision, (c) personlization and (d) novelty of the O-Hybrid method as a function of λ in Netflix network.The green lines mark the optimal λ* of the O-Hybrid method and the red lines mark the optimal results of the H-Hybrid method.

Mentions: In order to show in detail the advantage of H-Hybrid over O-Hybrid, we present in Fig 4 some curves from the heat maps in Fig 3. The blue curves are the results of the recommendation metrics versus the parameter λ in the O-Hybrid method, which are basically the diagonals in the heat map in Fig 3. Consistent with Ref. [15], we observe an optimal recommendation accuracy (in both RS and P) when λ is tuned. The green dashed lines mark the optimal λ when the optimal recommendation accuracy (RS) is achieved in O-Hybrid. Moreover, we mark the optimal RS and P of the H-Hybrid method by the red dashed lines ( and ). One can see that the H-Hybrid method can substantially outperform the O-Hybrid method in both RS and P. More specifically, the RS* in the O-Hybrid method is 0.0447 while the RS* in the H-Hybrid method can be as small as 0.0395. The improvement is 11.63%. For precision, P* is 0.1561 in O-Hybrid and P* is 0.1775 in H-Hybrid. The improvement of P is 13.71%. Fig 4(c)(d) show the results of recommendation diversity. Clearly, H-Hybrid recommendation is much more personalized and novel than O-Hybrid, with 9.0% improvement in D and 12.35% improvement in I.


Information Filtering via Heterogeneous Diffusion in Online Bipartite Networks.

Zhang FG, Zeng A - PLoS ONE (2015)

The (a) Ranking score, (b) Precision, (c) personlization and (d) novelty of the O-Hybrid method as a function of λ in Netflix network.The green lines mark the optimal λ* of the O-Hybrid method and the red lines mark the optimal results of the H-Hybrid method.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4488376&req=5

pone.0129459.g004: The (a) Ranking score, (b) Precision, (c) personlization and (d) novelty of the O-Hybrid method as a function of λ in Netflix network.The green lines mark the optimal λ* of the O-Hybrid method and the red lines mark the optimal results of the H-Hybrid method.
Mentions: In order to show in detail the advantage of H-Hybrid over O-Hybrid, we present in Fig 4 some curves from the heat maps in Fig 3. The blue curves are the results of the recommendation metrics versus the parameter λ in the O-Hybrid method, which are basically the diagonals in the heat map in Fig 3. Consistent with Ref. [15], we observe an optimal recommendation accuracy (in both RS and P) when λ is tuned. The green dashed lines mark the optimal λ when the optimal recommendation accuracy (RS) is achieved in O-Hybrid. Moreover, we mark the optimal RS and P of the H-Hybrid method by the red dashed lines ( and ). One can see that the H-Hybrid method can substantially outperform the O-Hybrid method in both RS and P. More specifically, the RS* in the O-Hybrid method is 0.0447 while the RS* in the H-Hybrid method can be as small as 0.0395. The improvement is 11.63%. For precision, P* is 0.1561 in O-Hybrid and P* is 0.1775 in H-Hybrid. The improvement of P is 13.71%. Fig 4(c)(d) show the results of recommendation diversity. Clearly, H-Hybrid recommendation is much more personalized and novel than O-Hybrid, with 9.0% improvement in D and 12.35% improvement in I.

Bottom Line: Previous works considered the diffusion process from user to object, and from object to user to be equivalent.We show in this work that it is not the case and we improve the quality of the recommendation by taking into account the asymmetrical nature of this process.Finally, this modification is checked to be able to improve the recommendation in a realistic case.

View Article: PubMed Central - PubMed

Affiliation: School of Information Technology, Jiangxi University of Finance and Economics, Nanchang 330013, P.R. China; Jiangxi Key Laboratory of Data and Knowledge Engineering, Jiangxi University of Finance and Economics, Nanchang, 330013, P. R. China.

ABSTRACT
The rapid expansion of Internet brings us overwhelming online information, which is impossible for an individual to go through all of it. Therefore, recommender systems were created to help people dig through this abundance of information. In networks composed by users and objects, recommender algorithms based on diffusion have been proven to be one of the best performing methods. Previous works considered the diffusion process from user to object, and from object to user to be equivalent. We show in this work that it is not the case and we improve the quality of the recommendation by taking into account the asymmetrical nature of this process. We apply this idea to modify the state-of-the-art recommendation methods. The simulation results show that the new methods can outperform these existing methods in both recommendation accuracy and diversity. Finally, this modification is checked to be able to improve the recommendation in a realistic case.

No MeSH data available.