Balancing diversity and accuracy of the recommendation system based on multi-objective optimization
Langlang Zhang, Anqi Pan, Hongrui Shi · 2021 China Automation Congress (CAC) · 2021
Recommendation systems are used in every aspect of people's lives in the context of today's Internet boom. However, most of the research on recommendation algorithms focus on improving the accuracy of algorithms, while the diversity and novelty of recommendation quality are more or less ignored. In addition, the accuracy and diversity of recommendation algorithms are usually contradictory. Therefore, the paper proposed a novel multi-objective framework to balance accuracy and diversity. Under the framework, similar products and diverse products are recommended by user clustering. Two objective functions are designed to represent the similarity and diversity of clustering. One objective function is to bring together users with similar preferences. Another objective function is to recommend diverse items to users. The novel multi-objective evolutionary algorithm can return a series of different trade-off solutions in one run. The proposed algorithm is applied on a real-world baseline dataset: MovieLens. Experimental results show that diversity and novelty are improved without sacrificing too much precision