Using Negative User Actions to Improve the Quality of Recommender Systems

Recommendation systems are finding increasingly wide application, encompassing a variety of domains and diverse data types. However, in scenarios with a limited number of items, traditional approaches often prove to be insufficiently effective. In such cases, methods based on boosting algorithms offer a more efficient solution. This paper proposes a way to improve recommendation quality within this approach by incorporating users' negative interactions with items. Integrating these data enables more accurate modeling of both preferences and avoidances. The presented method enhances recommendation personalization even under conditions of high interdependence and limited item availability.