Three papers accepted in ACM WWW 2023 (full papers)


Title: Disentangling Degree-related Biases and Interest for Out-of-Distribution Generalized Directed Network Embedding
Author: Hyunsik Yoo, Yeon-Chang Lee, Kijung Shin and Sang-Wook Kim
Abstract
The goal of directed network embedding is to represent the nodes in a given directed network as embeddings (i.e., low-dimensional vectors) that preserve the asymmetric relationships between nodes. While a number of directed network embedding methods have been proposed, we empirically show that the existing methods lack out-of-distribution generalization abilities against degree-related distributional shifts. To mitigate this problem, we propose ODIN (Out-of-Distribution Generalized Directed Network Embedding), a new directed NE method where we model multiple factors in the formation of directed edges. Then, for each node, ODIN learns multiple embeddings, each of which preserves its corresponding factor, by disentangling interest factors and biases related to in- and out-degrees of nodes. Our experiments on four real-world directed networks demonstrate that disentangling multiple factors enables ODIN to yield out-of-distribution generalized embeddings that are consistently effective under various degrees of shifts in degree distributions. Specifically, ODIN universally outperforms 9 stateof-the-art competitors in 2 LP tasks on 4 real-world datasets under both identical distribution (ID) and non-ID settings.


Title: GELTOR: A Novel Graph Embedding Method based on Listwise Learning to Rank
Author: Masoud Rehyani Hamedani, JinSu Ryu and Sang-Wook Kim
Abstract
Similarity-based embedding methods have introduced a new perspective on graph embedding by conforming the similarity distribution of latent vectors in embedding space to that of nodes in the graph. These methods show significant effectiveness over conventional embedding methods in various machine learning tasks. In this paper, we first point out the three drawbacks of these methods: inaccurate similarity computation, optimization goal conflict, and impairing in/out-degree distributions. Then, motivated by these drawbacks, we propose a novel similarity measure for graphs called AdaSim, which is conducive to the similarity-based graph embedding. We finally propose GELTOR, an effective graph embedding method that employs AdaSim as a node similarity measure and the concept of learning-to-rank in the embedding process. Contrary to existing similarity-based methods, GELTOR does not aim to learn the similarity scores distribution. Instead, for any target node 𝑣, GELTOR learns a model that conforms the ranks of 𝑣’s top-𝑡 similar nodes in the embedding space to their original ranks based on AdaSim* scores. The philosophy behind GELTOR is that for any target node 𝑣, exploiting only its top-𝑡 similar nodes is more beneficial to graph embedding than exploiting all the nodes in the graph, since considering dissimilar nodes may adversely affect the learning quality. We conduct extensive experiments with six real-world datasets to evaluate the effectiveness of GELTOR in graph reconstruction, link prediction, and node classification tasks. Our experimental results show that (1) AdaSim* outperforms AdaSim, RWR, and MCT in computing nodes similarity in graphs, (2) our GETLOR outperforms existing state-of-the-arts and conventional embedding methods in most cases of the above machine learning tasks, thereby implying learning-to-rank is beneficial to graph embedding.


Title: KHAN: Knowledge-Aware Hierarchical Attention Networks for Accurate Political Stance Prediction
Author: Yunyong Ko, Seongeun Ryu, Soeun Han, Youngseung Jeon, Jaehoon Kim, Sohyun Park, Kyungsik Han, Hanghang Tong and Sang-Wook Kim
Abstract
The political stance prediction for news articles has been widely studied to mitigate the echo chamber effect – people fall into their thoughts and reinforce their pre-existing beliefs. The previous works for the political stance prediction focus on (1) identifying political factors that could reflect the political stance of a news article and (2) capturing those factors effectively. Despite their empirical successes, they are not sufficiently justified in terms of how effective their identified factors are in the political stance prediction. Motivated by this, in this work, we conduct a user study to investigate important factors in political stance prediction, and observe that the context and tone of a news article (implicit) and external knowledge for real-world entities appearing in the article (explicit) are important in determining its political stance. Based on this observation, we propose a novel knowledge-aware approach to accurate political stance prediction (KHAN), employing (1) hierarchical attention networks (HAN) to learn the relationships among words/sentences in different levels and (2) knowledge encoding (KE) to incorporate external knowledge (both common and political) for real-world entities into the process of political stance prediction. Also, to take into account the subtle difference between opposite political stances, we build two independent political knowledge graphs (KG) (i.e., KG-lib and KG-con) and learn to fuse the different political knowledge. Through extensive evaluations on three realworld datasets, we demonstrate the superiority of KHAN in terms of (1) accuracy, (2) efficiency, and (3) effectiveness. For the detailed information about KHAN, we have released the code of KHAN and the datasets at: https://anonymous.4open.science/r/khan.

업데이트: