Mapping Item-wise Rating into Attribute-wise Ratings for Enhanced Personalized Recommendations
Pushya Chaparala, Panduranga Naidu Nagabhushan, Lakshmi Tulasi Ponduri, Pravallika Kummari · 2024
Personalization is the key to any recommendation service, as it relies heavily on analyzing user interactions with items to uncover patterns and suggest relevant items. While users provide multi-level ratings for every item in their interactions, the conventional recommendation system generally employs an average rating value. This may not be efficient in realizing personalized recommendations. To make recommendations more personalized, it is proposed to map items-wise multi-level ratings into attribute wise ratings employing the framework of Symbolic Data Analysis (SDA). This will create more informative user profiles, enabling better similarity-dissimilarity analysis between users, which ultimately generates better recommendations. The efficiency of the recommendation list is assessed using the NDCG and MRR evaluation metrics. Baseline algorithms are employed to compare the proposed model with a focus on the top-k (Top-5) recommendations to demonstrate its potential. In the evaluation, the proposed method significantly outperformed baseline algorithms. This highlights the effectiveness of utilizing attribute-wise ratings and SDA in generating personalized recommendations.