Enhancing Mobile App Recommendation by LLM Powered Explicit and Implicit Information Retrieval and Representation

The widespread adoption and rapid growth of mobile apps have opened up opportunities to maximize discovery and provide personalized recommendations to enhance user engagement and business performance. Advancements have been made in personalized product recommendations. In this study, we intend to leverage, the recently released large-scale mobile app dataset (mobileRec), along with the capabilities of Large Language Models (LLMs) to extract information from vast amounts of text and factorization machines that incorporate additional user and item features to improve the effectiveness of personalized app recommendations using the standard evaluation metrics such as HIT@K and NDCG@K. Our preliminary results indicate that hybrid algorithms with side information surpass existing methods of app recommendations on the said dataset. We further evaluated impact of different combinations of user and item features and their latent representations and achieved an average of 4X performance gains over the reported benchmarks on the data set across all K values. In addition, we propose, to making successful mobile app recommendations, limiting the scope of recommended items for a pre-defined app categories can enhance personalization where user preferences are often skewed. Taken together, our results indicate that, mobile app recommendations could be further enhanced by side information of users and items, latent space representation variations and user preferences over item categories play a critical role of achieving personalized experiences.