[Paper Review] Attention-based Group Recommendation

Tran Dang Quang Vinh, Tuan-Anh Nguyen Pham+2 more|arXiv (Cornell University)|Apr 12, 2018

Recommender Systems and Techniques38 references12 citations

TL;DR

This paper proposes an attention-based deep learning model for group recommendation that dynamically weights individual users' preferences within a group to improve recommendation accuracy. By modeling user-specific influence through an attention mechanism, the approach outperforms baseline methods across four datasets, demonstrating the effectiveness of attention in handling preference trade-offs in group settings.

ABSTRACT

Recommender systems are widely used in big information-based companies such as Google, Twitter, LinkedIn, and Netflix. A recommender system deals with the problem of information overload by filtering important information fragments according to users' preferences. However, most traditional recommendation techniques have limitations. In light of the increasing success of deep learning, recent studies have proved the benefits of using deep learning in various recommendation tasks. Recommendation architectures have been utilizing deep learning in order to overcome limitations of traditional recommendation techniques. We propose an extension of deep learning to solve the group recommendation problem. On the one hand, as different individual preferences in a group necessitate preference trade-offs in making group recommendations, it is essential that the recommendation model can discover substitutes among user behaviors. On the other hand, it has been observed that a user as an individual and as a group member behaves differently. To tackle such problems, we propose using an attention mechanism to capture the impact of each user in a group. Specifically, our model automatically learns the influence weight of each user in a group and recommends items to the group based on its members' weighted preferences. We conduct extensive experiments on four datasets. Our model significantly outperforms baseline methods and shows promising results in applying deep learning to the group recommendation problem.

Motivation & Objective

• Address the challenge of preference aggregation in group recommendation, where individual user preferences conflict and require trade-offs.
• Model the varying influence of individual users within a group, recognizing that group behavior differs from individual behavior.
• Overcome limitations of traditional recommendation systems by leveraging deep learning to capture complex user-item interactions.
• Develop a method that automatically learns user-specific attention weights to reflect each member's impact on group decisions.
• Demonstrate the effectiveness of attention mechanisms in improving recommendation performance for group settings using real-world datasets.

Proposed method

• Employ an attention mechanism to compute dynamic influence weights for each user in a group, reflecting their relative importance in shaping group preferences.
• Integrate user embeddings with attention weights to generate a group-level representation that aggregates individual preferences.
• Use deep neural networks to learn non-linear interactions between users and items, capturing complex preference patterns.
• Train the model end-to-end using a loss function that optimizes for group recommendation accuracy on item ranking or rating prediction.
• Apply multi-head attention to capture diverse aspects of user behavior and improve robustness to noisy or inconsistent preferences.
• Utilize user-item interaction data to train the model, with attention weights learned during backpropagation to reflect user influence in group decisions.

Experimental results

Research questions

• RQ1 How can attention mechanisms effectively model the varying influence of individual users within a group for recommendation?
• RQ2 To what extent does the proposed attention-based model improve recommendation accuracy compared to traditional group recommendation methods?
• RQ3 Can the model capture subtle preference trade-offs among group members without explicit user weighting?
• RQ4 How does the model generalize across diverse datasets with varying user and item interaction patterns?
• RQ5 What is the impact of attention weights on the final group recommendation quality in real-world scenarios?

Key findings

• The proposed attention-based model significantly outperforms baseline recommendation methods on all four evaluated datasets.
• The attention mechanism successfully learns meaningful influence weights for individual users, reflecting their actual contribution to group preferences.
• The model improves recommendation accuracy by effectively handling preference conflicts among group members through dynamic weighting.
• The use of deep learning with attention enables better representation of complex user behaviors in group settings compared to traditional collaborative filtering.
• The model demonstrates consistent performance gains across different types of group recommendation tasks, indicating strong generalization ability.
• The attention mechanism provides interpretable user influence scores, offering insights into group decision-making dynamics.