UniRecSys: A Unified Framework for Personalized, Group, Package, and Package-to-Group Recommendations

TL;DR

UniRecSys presents a unified matrix factorization framework that simultaneously learns latent representations for users, items, groups, and packages to support personalized, group, package, and package-to-group recommendations. By forming compatible groups/packages via spectral clustering and coupling their latent factors through a gamma-weighted alignment term, the approach extends RMF and MMMF to a single adaptable model. Empirical results on MovieLens 100K/1M show consistent gains across MAE, RMSE, and Precision@k compared with traditional baselines and task-specific models, highlighting the value of incorporating group and package information. The framework offers a flexible, scalable path for delivering diverse, stakeholder-aware recommendations in domains such as e-commerce and travel, while acknowledging overhead and dataset limitations as avenues for future improvement.

Abstract

Recommender systems aim to enhance the overall user experience by providing tailored recommendations for a variety of products and services. These systems help users make more informed decisions, leading to greater user engagement with the platform. However, the implementation of these systems largely depends on the context, which can vary from recommending an item or package to a user or a group. This requires careful exploration of several models during the deployment, as there is no comprehensive and unified approach that deals with recommendations at different levels. Furthermore, these individual models must be closely attuned to their generated recommendations depending on the context to prevent significant variation in their generated recommendations. In this paper, we propose a novel unified recommendation framework that addresses all four recommendation tasks, namely, personalized, group, package, and package-to-group recommendation, filling the gap in the current research landscape. The proposed framework can be integrated with most of the traditional matrix factorization-based collaborative filtering (CF) models. This research underscores the significance of including group and package information while learning latent representations of users and items for personalized recommendations. These components help in exploiting a rich latent representation of the user/item by enforcing them to align closely with their corresponding group/package representation. We consider two prominent CF techniques, namely Regularized Matrix Factorization and Maximum Margin Matrix factorization, as the baseline models and demonstrate their customization to various recommendation tasks. Experimental results on two publicly available datasets are reported, comparing them to other baseline approaches for various recommendation tasks.

Figures (9)

• Figure 1: An example of Matrix Factorization.
• Figure 2: An example of the user-item rating matrix.
• Figure 3: Outline of the proposed approach.
• Figure 4: An example of train-test split for group recommendation scenario.
• Figure 5: Effect of $K_1$ and $K_2$ in URMF for personalized recommendation tasks over MovieLens 100K dataset.