Conference report: CIKM 2019 Learning and Reasoning on Graph for Recommendation

class=”markdown_views prism-atom-one-dark”>

Foreword

This is a conference report on graphs and recommendations at CIKM 2019
Link: https ://dl.acm.org/doi/10.1145/3357384.3360317

1. Summary

Recommendation methods build predictive models to estimate the likelihood of user-item interactions. Previous models have largely followed a common supervised learning paradigm—treating each interaction as a separate data instance and making predictions based on “silos of information.” This approach ignores the relationships between data instances and may lead to poor performance, especially in sparse cases. In addition, models built on individual data instances have difficulty showing the reasons behind recommendations, making the recommendation process difficult to understand. We will revisit the recommendation problem from the perspective of graph learning. Common recommendation data sources can be organized into graphs, such as user-item interactions (bipartite graphs), social networks, knowledge graphs (heterogeneous graphs), etc. This graph-based organization connects isolated data instances, bringing benefits for developing higher-order connections that encode meaningful patterns for collaborative filtering, content-based filtering, social impact modeling, and knowledge-aware reasoning. Coupled with the recent success of graph neural networks (GNNs), graph-based models have the potential to become the next generation of recommendation system technology. And this article reviews graph-based recommendation learning methods, paying special attention to the latest developments in GNNs and knowledge graph-based recommendations. By introducing this emerging and promising topic in this article, we hope that readers will gain a deep understanding and accurate insights into the field, stimulate more ideas and discussions, and promote the development of technology.

2. Introduction

The main purpose of recommendations is to estimate the likelihood that a user will accept a target product, or more formally, the likelihood that a user will interact with the product. Existing methods [3, 4, 6, 7] largely follow the general supervised learning paradigm, which has two key components

• (1) Convert each interaction and its associated side information into a separate data instance
• (2) Construct a prediction model and make predictions based on examples. These methods have achieved great success and are widely used in industry.

However, there is an information silo problem in this paradigm – modeling each user-item interaction as an independent instance – ignoring the relationship between instances, which may lead to poor performance [11, 12, 16 ]. Furthermore, a model built on a single instance of data is largely a black box—only able to provide predictions but little about the reasoning behind the recommendations. This black-box nature makes decision-making processes opaque, difficult to understand, and hinders their further application. Therefore, it is of great significance to explore and mine interactive relationships.

Graphs are a powerful representation that represent data instances as nodes and the relationships between them as edges, rather than considering each instance in isolation. In recent years, people have generated great interest in graph neural networks (GNNs) [2, 5, 10]. The core idea is an information propagation mechanism—aggregating information from a node’s neighbors to enrich the node’s representation and improve downstream supervised learning. Thanks to this propagation effect, GNN-based methods have shown promising results and improved the current best methods in many challenging tasks. Inspired by the recent success of GNNs, we believe that graph learning techniques can serve as the infrastructure for next-generation recommendation. Therefore, it is very timely to re-examine the recommendation problem from the perspective of graph learning and introduce the latest research results of GNN-based recommenders. Here, we mainly focus on the following recommended solutions:

• Collaborative filtering: User-item interactions are organized into a bipartite graph between user and item nodes. Some recent efforts, such as GC-MC [9] and NGCF [12] recursively propagate embedding information on the graph to encode collaborative signals along high-order connectivity into representations of users and items and empirically obtain better representation [12].
• Social recommendations: Social networks represent social relationships between users, and connected users influence each other. Recent methods such as DANSER [14], GraphRec [1] and DiffNet [13] use GNN to simulate this social influence model – propagating similar interests through high-order social relationships – for better social recommendations.
• Sequence recommendation: The historical conversation sequence of user behavior is reorganized into a conversation graph, representing the conversion of items. Recently proposed works such as DGRec [8] and SR-GNN [15] perform information propagation on such graphs to model the dynamic user preferences of the session.
• Recommendation based on knowledge graph: External project knowledge, such as common sense knowledge and project properties, can be well represented as a knowledge graph (also known as a heterogeneous information network), where real-world entities and relationships are represented by subject-propertyobject ternary Factual representation. Among them, multi-hop relationship paths serve as supporting evidence of user preference for invisible interactions. Some recent efforts, such as KGAT [11] utilize GNN to synthesize information from this connectivity, enhance expressive capabilities, and enrich the relationships between users and items.

By introducing this emerging and promising topic, we hope that this article can help researchers and practitioners gain in-depth understanding and accurate insights on this topic, exchange productive ideas, and promote the development of technology.

REFERENCES

[1] Wenqi Fan, Yao Ma, Qing Li, Yuan He, Yihong Eric Zhao, Jiliang Tang, and
Dawei Yin. 2019. Graph Neural Networks for Social Recommendation. In WWW.
417–426 .
[2] William L. Hamilton, Zhitao Ying, and Jure Leskovec. 2017. Inductive
Representation Learning on Large Graphs. In NeurIPS. 1025–1035.
[3] Xiangnan He and Tat-Seng Chua. 2017. Neural Factorization Machines for Sparse
Predictive Analytics. In SIGIR. 355–364.
[4] Xiangnan He, Lizi Liao, Hanwang Zhang, Liqiang Nie, Xia Hu, and Tat-Seng
Chua. 2017. Neural Collaborative Filtering. In WWW. 173–182.
[5] Thomas N. Kipf and Max Welling. 2017. Semi-Supervised Classification with
Graph Convolutional Networks. In ICLR.
[ 6] Steffen Rendle, Christoph Freudenthaler, Zeno Gantner, and Lars Schmidt-Thieme.
2009. BPR: Bayesian Personalized Ranking from Implicit Feedback. In UAI. 452–
461.
[7] Steffen Rendle, Zeno Gantner, Christoph Freudenthaler, and Lars Schmidt-Thieme.
2011. Fast context-aware recommendations with factorization machines. In SIGIR.
635–644.
[8] Weiping Song, Zhiping Xiao, Yifan Wang , Laurent Charlin, Ming Zhang, and Jian
Tang. 2019. Session-Based Social Recommendation via Dynamic Graph Attention
Networks. In WSDM. 555–563.
[9] Rianne van den Berg, Thomas N . Kipf, and Max Welling. 2017. Graph
Convolutional Matrix Completion. In KDD.
[10] Petar Velickovic, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro
Liò, and Yoshua Bengio. 2018. Graph Attention Networks. In ICLR.
[11] Xiang Wang, Xiangnan He, Yixin Cao, Meng Liu, and Tat-Seng Chua. 2019. KGAT:
Knowledge Graph Attention Network for Recommendation. In KDD. 950– 958.
[12] Xiang Wang, Xiangnan He, Meng Wang, Fuli Feng, and Tat-Seng Chua. 2019.
Neural Graph Collaborative Filtering. In SIGIR. 165–174.
[13] Le Wu, Peijie Sun, Yanjie Fu, Richang Hong, Xiting Wang, and Meng Wang. 2019.
A Neural Influence Diffusion Model for Social Recommendation. In SIGIR.
[14] Qitian Wu, Hengrui Zhang, Xiaofeng Gao, Peng He, Paul Weng, Han Gao,
and Guihai Chen. 2019. Dual Graph Attention Networks for Deep Latent
Representation of Multifaceted Social Effects in Recommender Systems. In WWW.
2091–2102.
[15] Shu Wu, Yuyuan Tang, Yanqiao Zhu, Liang Wang, Xing Xie, and Tieniu Tan. 2019.
Session-based Recommendation with Graph Neural Networks. In AAAI.
[16] Fuzheng Zhang, Nicholas Jing Yuan, Defu Lian, Xing Xie, and Wei-Ying Ma. 2016.
Collaborative Knowledge Base Embedding for Recommender Systems. In KDD.
353–362.