Kullanici-komşuluk-tabanli işbirlikçi filtreleme algoritmalarinin popülerlik ayrimciliğini analiz etme

This thesis aims to study the popularity bias present in user-neighbourhood based collaborative filtering algorithms. Popularity bias refers to the tendency of recommendation systems to recommend popular items more frequently than less popular items, resulting in a need for more diversity in recommendations. To examine this phenomenon, we test three algorithms, KNNwithMeans, KNNwithZscore, and KNNBaseline, using three different similarity metrics: Cosine, MSD and Pearson's corelation coefficient. We analyse three distinct datasets: Movielens-1M, Yahoo Music, and Douban Books. The results of our study indicate that the use of different similarity metrics and algorithms can significantly impact the level of popularity bias present in the recommendations. Furthermore, our findings suggest that the best approach to mitigate popularity bias is to use a combination of different similarity metrics and algorithms. This approach allows for a more diverse set of recommendations, which can lead to a more personalised user experience. Our study provides valuable insights into the impact of popularity bias on recommendation systems and the effectiveness of various methods for addressing it. This thesis highlights the importance of addressing popularity bias in recommendation systems. The results of this study can be used to guide the development of more diverse recommendation systems, and personalised, in turn, can lead to a better user experience