Walking on a user similarity network towards personalized recommendations

Area of Science:

• Computer Science
• Information Retrieval

Background:

• Personalized recommender systems combat information overload on the web.
• Traditional collaborative filtering struggles with popular item bias, leading to suboptimal recommendations.
• Diffusion and random walk models show promise for improved recommendation accuracy and diversity.

Purpose of the Study:

• To propose a novel random walk with restart model for personalized recommendations.
• To enhance user similarity networks using power-law adjustment, nearest neighbor, and threshold filtration strategies.
• To evaluate the proposed method against state-of-the-art techniques on real-world datasets.

Main Methods:

• Construction of an adjusted user similarity network using historical user data.
• Application of a random walk with restart model on the constructed network.
• Cross-validation experiments on MovieLens and Netflix datasets.

Main Results:

• The proposed method significantly outperforms existing approaches.
• Demonstrated improvements in recommendation accuracy and diversity.
• Enhanced retrieval performance compared to baseline methods.

Conclusions:

• The random walk with restart model offers a superior alternative to traditional collaborative filtering.
• The network adjustment strategies effectively mitigate issues like popular item bias.
• The method provides more accurate, diverse, and efficient personalized recommendations.