Active learning algorithm for alleviating the user cold start problem of recommender systems.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Recommender systems face challenges in profiling new users.
• Active learning strategies are commonly used to address this by requesting user ratings for selected items.

Purpose of the Study:

• To propose and evaluate a novel decision tree-based algorithm for item selection in active learning for recommender systems.
• To investigate the effectiveness of this algorithm in improving recommender system performance.

Main Methods:

• Developed a decision tree-based algorithm to select items for user interviews.
• Treated the recommender system as a black box, feeding collected ratings back to improve performance.
• Conducted extensive offline evaluations using two datasets and various recommender algorithms.
• Performed online evaluations with 50 real users.

Main Results:

• Offline evaluations demonstrated that the proposed algorithm improves recommender performance when users can rate most presented items.
• Online evaluations with real users failed to show a significant positive impact on recommender performance.
• A discrepancy was observed between offline and online evaluation results.

Conclusions:

• The proposed active learning algorithm shows promise in offline settings but its effectiveness with real users is uncertain.
• Real users' inability to rate all selected items poses a challenge for active learning in recommender systems.
• Further research is needed to bridge the gap between offline and online evaluation results for active learning strategies.