AI-Enabled Intelligent Recommendation Engine for Personalized Digital Services

Abstract

While the use of recommendation engines to connect users to the service they need is becoming central to the success of digital platforms, the traditional approaches to recommendation engines are faced with a number of challenges in this context: lack of interactions, contextual variation, and rapidly changing preferences. This study aims to suggest an intelligent recommendation engine with AI, which is based on the Attention-based Hybrid Deep Learning Recommendation method based on PyTorch. It brings together embeddings of user information, embeddings of service identities, features characterizing service categories and contextual signals from the surrounding content, and extracts the relative weights of the heterogeneous representations by employing an attention mechanism, to utilize these representations for ranking candidate services. A synthetic benchmark was used to create a reproducible proof-of-concept experiment and contains 360 users, 180 digital services, 12 service categories, 8 contexts, 5,760 training interactions and 1,080 test interactions. The proposed model achieved better ranking quality in all of the overall ranking metric with HR at 10=0.3315, NDCG at 10=0.1694, MRR at 10=0.1208 and AUC=0.6492 compared to popularity, matrix factorization, neural collaborative filtering and non-attentive hybrid model. Results show the usefulness of adaptive feature fusion to personalize tasks in a context-specific way.