Serendipitous Recommendation in E-Commerce Using Innovator-Based Collaborative Filtering.

Area of Science:

• Computer Science
• Information Retrieval

Background:

• Collaborative filtering (CF) is prevalent in recommender systems but suffers from the Matthew effect, recommending only popular items.
• This leads to poor performance in discovering cold items (new or niche items) and user dissatisfaction.

Purpose of the Study:

• To address the Matthew effect in recommender systems.
• To propose a novel recommendation algorithm capable of recommending cold items.
• To balance serendipity and accuracy in recommendations.

Main Methods:

• A user survey on online shopping habits in China was conducted.
• A novel innovator-based CF algorithm was developed, introducing the concept of 'innovators'—users who discover cold items.
• Extensive experiments were performed on a large-scale e-commerce dataset from Alibaba.

Main Results:

• The proposed innovator-based CF algorithm effectively recommends cold items.
• The algorithm balances serendipity and accuracy in recommendation lists.
• Experimental results validated the algorithm's effectiveness in a real-world e-commerce environment.

Conclusions:

• Innovator-based CF offers a viable solution to the Matthew effect in recommender systems.
• The approach enhances the discovery of less popular items, improving recommendation diversity.
• This method provides a more balanced and effective recommendation experience for users.