Unsupervised Large-Scale Social Network Alignment via Cross Network Embedding

Liang, Zhehan; Rong, Yu; Li, Chenxin; Zhang, Yunlong; Huang, Yue; Xu, Tingyang; Ding, Xinghao; Huang, Junzhou

doi:10.1145/3459637.3482310

Cited by 23 publications

(9 citation statements)

References 36 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Similarly, for the high-bias algorithm L3 [6], which counts paths of length 3 and applies a soft normalization, we create a lower-biased version called L3n by applying a stronger normalization based on node degrees. Besides these, we consider preferential attachment (a purely-biased baseline that specifically brings forward interactions between high degree nodes), LINE [36] a neural-network based embedding algorithm with high bias (since its learning process captures node degree information in its embeddings; S. Figure 2), two network propagation algorithms von Neumann [37] and random walks with restarts (RWR) [38], both having low-bias due to strong normalization based on node degrees in their formulation.…”

Section: Resultsmentioning

confidence: 99%

See 1 more Smart Citation

Are under-studied proteins under-represented? How to fairly evaluate link prediction algorithms in network biology

Yılmaz

Yorgancioglu

Koyutürk

2022

Preprint

View full text Add to dashboard Cite

In the context of biomedical applications, new link prediction algorithms are continuously being developed and these algorithms are typically evaluated computationally, using test sets generated by sampling the edges uniformly at random. However, as we demonstrate, this creates a bias in the evaluation towards "the rich nodes", i.e., those with higher degrees in the network. More concerningly, we demonstrate that this bias is prevalent even when different snapshots of the network are used for evaluation as recommended in the machine learning community. This leads to a cycle in research where newly developed algorithms generate more knowledge on well-studied biological entities while the under-studied entities are commonly ignored. To overcome this issue, we propose a weighted validation setting focusing on under-studied entities and present strategies to facilitate bias-aware evaluation of link prediction algorithms. These strategies can help researchers gain better insights from computational evaluations and promote the development of new algorithms focusing on novel findings and under-studied proteins. We provide a web tool to assess the bias in evaluation data at: <a href = https://yilmazs.shinyapps.io/colipe/>https://yilmazs.shinyapps.io/colipe/</a>

show abstract

Section: Resultsmentioning

confidence: 99%

“…Thus, we consider both to be low-bias algorithms. Embedding/Learning Methods: We consider two types of embedding methods: Deepwalk [13] (Random walk based) and Line [36] (Neural network based). For each of these methods, we train a logistic regression model using the embeddings as features.…”

Section: Link Prediction Algorithms -Verbal Descriptionsmentioning

confidence: 99%

Are under-studied proteins under-represented? How to fairly evaluate link prediction algorithms in network biology

Yılmaz

Yorgancioglu

Koyutürk

2022

Preprint

View full text Add to dashboard Cite

show abstract

“…Karakasis et al [19] propose to learn the node embeddings and matches the nodes of two graphs simultaneously. Considering the big success in various graph mining tasks, graph neural networks are integrated in graph alignment algorithms to acquire better node representations [14], [31], and adversarial learning strategies are utilized to further improve model performance [4], [9]. Besides, many unsupervised methods have been proposed for KG alignment [34], [37].…”

Section: Related Workmentioning

confidence: 99%

“…However, these correspondences are usually unavailable and further suffer from the labor expensiveness issue in real-world applications. Thus, unsupervised graph alignment methods have attracted increasing attention [4], [9], [14], [17], [19], [31], [72]. Also, graph nodes are often associated with wealthy side information, such as the user information of social network accounts or the embedding of knowledge graph entities.…”

Section: Introductionmentioning

confidence: 99%

“…In previous works, a popular paradigm for unsupervised graph alignment is the "embed-then-cross-compare" procedure [4], [9], [14], [17], [23], [31], as shown in Figure 2(a). As the name suggests, it first embeds nodes in each graph into a common feature space (e.g., using a graph neural network), and then compares embeddings across two graphs to obtain node correspondences.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Robust Attributed Graph Alignment via Joint Structure Learning and Optimal Transport

Tang¹,

Zhang²,

Li³

et al. 2023

Preprint

View full text Add to dashboard Cite

Graph alignment, which aims at identifying corresponding entities across multiple networks, has been widely applied in various domains. As the graphs to be aligned are usually constructed from different sources, the inconsistency issues of structures and features between two graphs are ubiquitous in real-world applications. Most existing methods follow the "embed-then-cross-compare" paradigm, which computes node embeddings in each graph and then processes node correspondences based on cross-graph embedding comparison. However, we find these methods are unstable and sub-optimal when structure or feature inconsistency appears. To this end, we propose SLOTAlign, an unsupervised graph alignment framework that jointly performs Structure Learning and Optimal Transport Alignment. We convert graph alignment to an optimal transport problem between two intra-graph matrices without the requirement of cross-graph comparison. We further incorporate multi-view structure learning to enhance graph representation power and reduce the effect of structure and feature inconsistency inherited across graphs. Moreover, an alternating scheme based algorithm has been developed to address the joint optimization problem in SLOTAlign, and the provable convergence result is also established. Finally, we conduct extensive experiments on six unsupervised graph alignment datasets and the DBP15K knowledge graph (KG) alignment benchmark dataset. The proposed SLOTAlign shows superior performance and strongest robustness over seven unsupervised graph alignment methods and five specialized KG alignment methods. 1

show abstract