Diversity Focused Content-Based Filtering Recommendation System

Determining choices can be a challenging task for consumers. Recommendation systems serve as tools to provide suggestions, aiding users in decision-making processes. A major concern in recommendation systems is the issue of diversity, particularly in content-based approaches where recommendations are based on the attributes of items that users have previously liked. This can lead to overspecialization, where the system provides recommendations that are highly similar to each other, reducing the variety or diversity of suggestions. To address this issue, this research focuses on improving the recommendations diversity in a content-based filtering system. The proposed method in this study can be applied to recommendation systems that focus on improving recommendation diversity. The method used to improve recommendation diversity is the implementation of a similarity threshold. This threshold acts as a filtering mechanism, ensuring that the recommended items fall within a specific similarity range relative to the user's preferences, calculated using cosine similarity. Additionally, clustering is used to compare the performance of the model with and without cluster labels. The results demonstrate that applying a similarity threshold effectively enhances recommendation diversity, as reflected by improved individual diversity scores. Notably, the highest individual diversity values were achieved at the first threshold with top-n = 30, where scores reached 0.8240 without cluster labels and 0.8750 with cluster labels.