The GReS workshop accepts papers of up to 14 pages following the standard single-column ACM RecSys template. This length does not include references and reviewers will be asked to comment on whether the length is appropriate for the contribution. Submissions are double-blind (and therefore should be anonymized) and to be provided in a single file in .pdf format.

Submission link: https://cmt3.research.microsoft.com/GReS2021/

Paper submission deadline: July 29th, 2021 (AoE)

Author notification: August 21st, 2021 (AoE)

Camera-ready version deadline: September 4th, 2021 (AoE)

Topics

The topics relevant to the GReS workshop include (but are not limited to):

• Top-k recommendation and matrix completion approaches based on GNNs;
• Session-based and next-item recommendation via GNNs and dynamic graphs;
• Knowledge graph and social network-enhanced recommendation models;
• Multimodal recommendation and search approaches based on GNNs;
• Explainability, fairness, accountability, transparency, and privacy issues in GNN-based recommendation;
• GNN-based result diversification in recommendation or search;
• Hypergraph neural networks for recommendation or search;
• GNNs for personalized recommendation via link prediction in multipartite or heterogeneous graphs;
• Temporal and Dynamic GNNs or applications of GNNs to next-item recommendation and dynamic environments;
• Graph topology inference for recommendation and search;
• Challenges, pitfalls, and negative results in applying GNNs to recommendation or search;
• Libraries, benchmarks, and datasets for GNN-based recommendation or search;
• Industrial applications and scalability of GNNs for recommendation or search.

3:00 – 3:05 PM: Workshop’s Opening
3:05 – 3:45 PM: Keynote I – Graph Neural Networks for Recommendation by Xiang Wang (National University of Singapore)
3:45 – 3:50 PM: Short Break
3:50 – 4:10 PM: Accepted Paper I – Knowledge Graph Attention for Sequential Recommendations by Amjadi et al
4:10 – 4:50 PM: Keynote II – Towards GNN explainability: Random Walk Graph Neural Networks by Michalis Vazirgiannis (Ecole Polytechnique)
4:50 – 5:20 PM: Long Break on wonder.me
5:20 – 5:40 PM: Accepted Paper II – CoRGi: Content-Rich Graph Neural Networks with Attention by Kim et al
5:40 – 6:00 PM: Accepted Paper III – Referral prediction in Healthcare using Graph Neural Networks by Duarte et al
6:00 – 6:10 PM: Short Break
6:10 – 6:50 PM: Keynote III – Powering Pinterest Recommendations with Graph Neural Networks by Andrew Zhai (Pinterest)
6:50 – 7:00 PM: Workshop’s Closing

Xiang Wang (National University of Singapore)
Title: Graph Neural Networks for Recommendation
Abstract: Graph Neural Networks (GNNs) have achieved remarkable success in many domains and shown great potentials in personalized recommendation. In this talk, I will give a brief introduction on why GNNs are suitable for recommendation, and how to incorporate GNNs into some recommendation scenarios (e.g., collaborative filtering, knowledge-aware recommendation). I will also share some points on what advances we are researching into (e.g., self-supervised learning, contrastive learning). In the last part, I will discuss in the “Achilles’ Heel” of the GNN-based recommender models (e.g., amplify popularity bias), as well as some possible solutions (e.g., causal inference).

Michalis Vazirgiannis (Ecole Polytechnique)
Title: Towards GNN explainability: Random Walk Graph Neural Networks
Abstract: In recent years, graph neural networks (GNNs) have become the de facto tool for performing machine learning tasks on graphs. Most GNNs belong to the family of message passing neural networks (MPNNs). These models employ an iterative neighborhood aggregation scheme to update vertex representations. Then, to compute vector representations of graphs, they aggregate the representations of the vertices using some permutation invariant function. One would expect the hidden layers of a GNN to be composed of parameters that take the form of graphs. However, this is not the case for MPNNs since their update procedure is parameterized by fully-connected layers. In this paper, we propose a more intuitive and transparent architecture for graph-structured data, so-called Random Walk Graph Neural Network (RWNN). The first layer of the model consists of a number of trainable “hidden graphs” which are compared against the input graphs using a random walk kernel to produce graph representations. These representations are then passed on to a fully-connected neural network which produces the output. The employed random walk kernel is differentiable, and therefore, the proposed model is end-to-end trainable. We demonstrate the model’s transparency on synthetic datasets. Furthermore, we empirically evaluate the model on graph classification datasets and show that it achieves competitive performance.

Andrew Zhai (Pinterest)
Title: Powering Pinterest Recommendations with Graph Neural Networks
Abstract: Pinterest is the home of inspiration to over 450M monthly active users. Inspiration comes from our personalized recommender systems generating content from our catalog of billions of ideas. Graph Neural Networks (GNNs) have been a key method in improving the predictive performance of these systems through helping us understand content, search queries, and users more comprehensively. In our talk we will discuss the evolution of these representations starting with (1) using GNNs to model content (pins) combining multimodal node features with our web scale pin-board graph of billions of nodes and edges (2) showing how our content embeddings enable us to learn good representations of search queries and users, combining our GNN embeddings with techniques such as sequence models to capture temporal behavior of users across all of Pinterest. Beyond sharing technical details, we will quantify our learnings with online experimentation showing the impact of our methods.