Friend Story Ranking with Edge-Contextual Local Graph Convolutions

Abstract

Social platforms have paved the way in creating new, modern ways for users to communicate with each other. In recent years, multiple platforms have introduced "Stories" features, which enable broadcasting of ephemeral multimedia content. Specifically, "Friend Stories, " or Stories meant to be consumed by one's close friends, are a popular feature, promoting significant user-user interactions by allowing people to see (visually) what their friends and family are up to. A key challenge in surfacing Friend Stories for a given user, is in ranking over each viewing user's friends to efficiently prioritize and route limited user attention. In this work, we explore the novel problem of Friend Story Ranking from a graph representation learning perspective. More generally, our problem is a link ranking task, where inferences are made over existing links (relations), unlike common node or graph-based tasks, or link prediction tasks, where the goal is to make inferences about non-existing links. We propose ELR, an edge-contextual approach which carefully considers local graph structure, differences between local edge types and directionality, and rich edge attributes, building on the backbone of graph convolutions. ELR handles social sparsity challenges by considering and attending over neighboring nodes, and incorporating multiple edge types in local surrounding egonet structures. We validate ELR on two large country-level datasets with millions of users and tens of millions of links from Snapchat. ELR shows superior performance over alternatives by ≈ 8% and ≈ 5% error reduction measured by MSE and MAE correspondingly. Further generality, data efficiency and ablation experiments confirm the advantages of ELR.

MM-GNN: Mix-Moment Graph Neural Network towards Modeling Neighborhood Feature Distribution

“…Most existing GNNs follow the neighborhood aggregation scheme, where each node aggregates features of neighbors to update its own representation. Many GNNs [9,10,38,44] with different aggregators have been proposed to solve various downstream tasks, which can be divided into graph-level tasks [2,42], edge-level tasks, and node-level tasks, such as graph classification, link prediction [17,22,47], and node classification [8,15,19,27,36,48].…”

Deep learning for graph structured data

“…In fact, even a third hop neighbor set retrieval takes a significant amount of time for graphs with hundreds of millions of nodes, as the computational complexity of retrieving a node's l hop neighborhood is O(d l ), where d is the average node degree. Therefore, for efficient retrieval for large graphs, we establish only two hops retrieval as a key constraint; this limit is commonly adopted in large-scale graph learning applications [49,198,199].…”