earticle

영어

To investigate how training data bias impacts recommendation quality, this study experimented with simulated data to observe changes in recommendation performance and result bias across varying levels of data bias. The experiment revealed that while imbalances in popularity and preference distributions within the training data did not significantly affect recommendation accuracy, they did strongly influence recommendation result bias, leading to amplification effects. Specifically, increased preference distribution imbalance intensified recommendation result bias, whereas a certain level of popularity imbalance had no effect on recommendation result bias nor caused amplification. These findings challenge previous research suggesting that data bias degrades recommendation accuracy but support claims that data bias amplifies recommendation result bias. This study is notable for providing insights into how data bias mechanisms impact recommendation quality, offering valuable guidance for future research focused on bias mitigation.