[Paper Review] An Intelligent Group Event Recommendation System in Social networks

Guoqiong Liao, Huang Xiao-mei+2 more|arXiv (Cornell University)|Jun 16, 2020

Recommender Systems and Techniques42 references4 citations

TL;DR

This paper proposes ACGER, an attention-based context-aware group event recommendation model for Event-Based Social Networks (EBSNs) that dynamically learns context influences on users, groups, and events using neural networks. By integrating indirect (attention-based aggregation) and direct (neural network-learned) group preferences and co-training with individual recommendation, ACGER significantly outperforms state-of-the-art models on two real-world datasets.

ABSTRACT

The importance of contexts has been widely recognized in recommender systems for individuals. However, most existing group recommendation models in Event-Based Social Networks (EBSNs) focus on how to aggregate group members' preferences to form group preferences. In these models, the influence of contexts on groups is considered but simply defined in a manual way, which cannot model the complex and deep interactions between contexts and groups. In this paper, we propose an Attention-based Context-aware Group Event Recommendation model (ACGER) in EBSNs. ACGER models the deep, non-linear influence of contexts on users, groups, and events through multi-layer neural networks. Especially, a novel attention mechanism is designed to enable the influence weights of contexts on users/groups change dynamically with the events concerned. Considering that groups may have completely different behavior patterns from group members, we propose that the preference of a group need to be obtained from indirect and direct perspectives (called indirect preference and direct preference respectively). In order to obtain the indirect preference, we propose a method of aggregating preferences based on attention mechanism. Compared with existing predefined strategies, this method can flexibly adapt the strategy according to the events concerned by the group. In order to obtain the direct preference, we employ neural networks to directly learn it from group-event interactions. Furthermore, to make full use of rich user-event interactions in EBSNs, we integrate the context-aware individual recommendation task into ACGER, which enhances the accuracy of learning of user embeddings and event embeddings. Extensive experiments on two real datasets from Meetup show that our model ACGER significantly outperforms the state-of-the-art models.

Motivation & Objective

• Address the limitation of predefined, inflexible group aggregation strategies in existing group recommendation models.
• Model the complex, non-linear, and dynamic influence of contextual factors (e.g., time, location, social ties) on group decision-making in EBSNs.
• Capture group preferences through both indirect (member preference aggregation) and direct (group-specific interaction patterns) perspectives.
• Improve embedding quality and recommendation accuracy by jointly training group and individual recommendation tasks.
• Enable the model to adapt its decision strategy based on the event type by learning dynamic attention weights for contextual factors.

Proposed method

• Propose a novel attention mechanism that jointly models user/group-context and event-context interactions to compute dynamic influence weights of contexts on group decisions.
• Use a neural network-based attention mechanism to learn adaptive group aggregation strategies from data, replacing fixed rules like average or approval voting.
• Model indirect group preference by aggregating individual member preferences through learned attention weights that vary with the event type.
• Learn direct group preference directly from group-event interaction data using deep neural networks to capture unique group behavior patterns.
• Co-train the group recommendation task with an individual recommendation task to enhance user and event embedding quality via shared representation learning.
• Employ multi-layer neural networks to model deep, non-linear interactions among users, groups, events, and contextual factors.

Experimental results

Research questions

• RQ1 How can we model the dynamic, non-linear influence of contextual factors (e.g., time, location, social ties) on group preferences in EBSNs?
• RQ2 Can a learnable attention mechanism improve group preference aggregation compared to predefined strategies like average or least misery?
• RQ3 To what extent do indirect (member-aggregated) and direct (group-specific) group preferences contribute to accurate event recommendation?
• RQ4 Does integrating individual recommendation into the group recommendation framework enhance the performance of group recommendation?
• RQ5 How do contextual factors affect the model’s ability to learn accurate group preferences and improve recommendation accuracy?

Key findings

• ACGER significantly outperforms state-of-the-art models on two real-world Meetup datasets across all evaluation metrics (P@5, R@5, NDCG@5).
• The performance of ACGER_U (with indirect preference only) exceeds ACGER_G (with direct preference only), indicating that indirect preference plays a dominant role in group preference learning.
• ACGER_Grp, which omits the individual recommendation task, performs significantly worse than ACGER, with a 10.03% drop in P@5 on the New York dataset, showing the benefit of co-training.
• ACGER2, which excludes context information, performs markedly worse than ACGER, confirming that context modeling is critical for high recommendation accuracy.
• The attention mechanism effectively captures dynamic context influence, as the model adapts its attention weights based on the event type, improving context-aware decision-making.
• The integration of individual and group recommendation tasks enhances the quality of user and event embeddings, leading to more accurate group preference modeling.