Collaborative Filtering Recommendation on Users' Interest Sequences.

Area of Science:

• Computer Science
• Information Retrieval
• Artificial Intelligence

Background:

• Recommender systems often use time information to model dynamic user preferences.
• However, the sequential nature of user behavior and its impact on similarity is underexplored.

Purpose of the Study:

• To propose a novel collaborative filtering recommendation method that incorporates users' interest sequences (IS).
• To enhance user similarity calculations by considering sequential behavior patterns and interest transitions.

Main Methods:

• Developed a method to rank user behaviors based on timestamps to form interest sequences (IS).
• Extracted semantic features from IS using longest common sub-IS (LCSIS) and total common sub-IS (ACSIS).
• Refined traditional collaborative filtering similarities with IS-based similarities to capture dynamic preferences.

Main Results:

• The proposed method effectively captures transition characteristics and dynamic evolution of user preferences.
• Experimental results on MovieLens, Flixster, and Ciao datasets demonstrate improved accuracy in rating prediction.
• Outperformed several state-of-the-art time-aware collaborative filtering algorithms.

Conclusions:

• Incorporating user interest sequences significantly improves recommendation accuracy.
• The method provides a more nuanced understanding of user preferences by considering behavioral evolution.
• This approach offers a valuable advancement for time-aware recommender systems.