New recommender system evaluation approaches based on user selections factor.

Area of Science:

• Computer Science
• Information Retrieval

Background:

• Recommender systems (RSs) are crucial in e-commerce and social networking.
• Existing RS evaluation metrics often lack clarity and are difficult to extend.
• There is a need for more reliable and human-behavior-aligned RS evaluation measures.

Purpose of the Study:

• To improve existing recommender system evaluation measures (diversity and novelty) into novel metrics (variety and newness).
• To develop evaluation rules that better reflect human behavior for increased reliability.
• To enhance the precision and centralization of recommender system evaluations.

Main Methods:

• Developed two novel recommender system evaluation measures: variety and newness.
• Based the new measures on human behavior, adapting previous diversity and novelty rules.
• Applied the new rules to improve suggestion weighting and user behavior consideration.

Main Results:

• The new variety and newness measures demonstrated improved precision and centralization.
• Variety measure showed a 22.54% improvement.
• Newness measure showed a 14.84% improvement.

Conclusions:

• The enhanced variety and newness measures offer greater reliability and compatibility for recommender systems.
• These improved metrics facilitate better comparative analyses and RS development, particularly in social networking.
• The study aims to foster advancements in recommender system research and competition.