Addressing popularity bias in personality-aware recommender systems

Recommender systems are extensively employed in diverse domains and assist users to identify the most suitable product of their choice. Nevertheless, the presence of a notorious problem of popularity bias in recommendation algorithms highlights only popular items (head items) and ignores less popular items (tail items) in the recommendation lists. In this way, the tail items remain unseen, and the user only sees the head items as recommendation. The reason behind this imbalance approach is the imperfect distribution of ratings. There exist several techniques via popularity bias problem can be resolved such as pre-processing, in-processing, and post-processing techniques. This research presents a pre-processing technique to diminish the bias effect in the recommendation process. The proposed technique introduces the concept of injection of synthetic ratings to tail items in the domain of personality-aware recommendations. Rating imbalance reduces by using the synthetic injection method resulting in efficient and fair recommendations. For this purpose, at first the proposed method calculates the average rating count of head items to find the number of ratings to inject. After the synthetic rating count estimation, the system calculates the value of synthetic ratings to be injected. The performance of the proposed system is estimated using accuracy and beyond-accuracy metrics. This study proposes a new evaluation metric called general performance indicator that balances the effectiveness of both accuracy and beyond-accuracy metrics.