PME: projected metric embedding on heterogeneous networks for link prediction

Hongxu Chen, Hao Wang, Hongzhi Yin, Quoc Viet Hung Nguyen, Weiqing Wang, Xue Li

Research output: Chapter in Book/Report/Conference proceedingConference PaperResearchpeer-review

Abstract

Heterogenous information network embedding aims to embed heterogenous information networks (HINs) into low dimensional spaces, in which each vertex is represented as a low-dimensional vector, and both global and local network structures in the original space are preserved. However, most of existing heterogenous information network embedding models adopt the dot product to measure the proximity in the low dimensional space, and thus they can only preserve the first-order proximity and are insufficient to capture the global structure. Compared with homogenous information networks, there are multiple types of links (i.e., multiple relations) in HINs, and the link distribution w.r.t relations is highly skewed. To address the above challenging issues, we propose a novel heterogenous information network embedding model PME based on the metric learning to capture both first-order and second-order proximities in a unified way. To alleviate the potential geometrical inflexibility of existing metric learning approaches, we propose to build object and relation embeddings in separate object space and relation spaces rather than in a common space. Afterwards, we learn embeddings by firstly projecting vertices from object space to corresponding relation space and then calculate the proximity between projected vertices. To overcome the heavy skewness of the link distribution w.r.t relations and avoid “over-sampling” or “under-sampling” for each relation, we propose a novel loss-aware adaptive sampling approach for the model optimization. Extensive experiments have been conducted on a large-scale HIN dataset, and the experimental results show superiority of our proposed PME model in terms of prediction accuracy and scalability.

Original languageEnglish
Title of host publicationKDD' 18 - Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining
EditorsChih-Jen Lin, Hui Xiong
Place of PublicationNew York NY USA
PublisherAssociation for Computing Machinery (ACM)
Pages1177-1186
Number of pages10
ISBN (Print)9781450355520
DOIs
Publication statusPublished - 19 Jul 2018
EventACM International Conference on Knowledge Discovery and Data Mining 2018 - London, United Kingdom
Duration: 19 Aug 201823 Aug 2018
Conference number: 24th
http://www.kdd.org/kdd2018/ (Conference website)

Conference

ConferenceACM International Conference on Knowledge Discovery and Data Mining 2018
Abbreviated titleSIGKDD 2018
CountryUnited Kingdom
CityLondon
Period19/08/1823/08/18
Internet address

Keywords

  • Heterogenous Network Embedding
  • Link Prediction

Cite this

Chen, H., Wang, H., Yin, H., Nguyen, Q. V. H., Wang, W., & Li, X. (2018). PME: projected metric embedding on heterogeneous networks for link prediction. In C-J. Lin, & H. Xiong (Eds.), KDD' 18 - Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (pp. 1177-1186). New York NY USA: Association for Computing Machinery (ACM). https://doi.org/10.1145/3219819.3219986
Chen, Hongxu ; Wang, Hao ; Yin, Hongzhi ; Nguyen, Quoc Viet Hung ; Wang, Weiqing ; Li, Xue. / PME : projected metric embedding on heterogeneous networks for link prediction. KDD' 18 - Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. editor / Chih-Jen Lin ; Hui Xiong. New York NY USA : Association for Computing Machinery (ACM), 2018. pp. 1177-1186
@inproceedings{daf324d0f7c34ff8b5f8374081c2eb26,
title = "PME: projected metric embedding on heterogeneous networks for link prediction",
abstract = "Heterogenous information network embedding aims to embed heterogenous information networks (HINs) into low dimensional spaces, in which each vertex is represented as a low-dimensional vector, and both global and local network structures in the original space are preserved. However, most of existing heterogenous information network embedding models adopt the dot product to measure the proximity in the low dimensional space, and thus they can only preserve the first-order proximity and are insufficient to capture the global structure. Compared with homogenous information networks, there are multiple types of links (i.e., multiple relations) in HINs, and the link distribution w.r.t relations is highly skewed. To address the above challenging issues, we propose a novel heterogenous information network embedding model PME based on the metric learning to capture both first-order and second-order proximities in a unified way. To alleviate the potential geometrical inflexibility of existing metric learning approaches, we propose to build object and relation embeddings in separate object space and relation spaces rather than in a common space. Afterwards, we learn embeddings by firstly projecting vertices from object space to corresponding relation space and then calculate the proximity between projected vertices. To overcome the heavy skewness of the link distribution w.r.t relations and avoid “over-sampling” or “under-sampling” for each relation, we propose a novel loss-aware adaptive sampling approach for the model optimization. Extensive experiments have been conducted on a large-scale HIN dataset, and the experimental results show superiority of our proposed PME model in terms of prediction accuracy and scalability.",
keywords = "Heterogenous Network Embedding, Link Prediction",
author = "Hongxu Chen and Hao Wang and Hongzhi Yin and Nguyen, {Quoc Viet Hung} and Weiqing Wang and Xue Li",
year = "2018",
month = "7",
day = "19",
doi = "10.1145/3219819.3219986",
language = "English",
isbn = "9781450355520",
pages = "1177--1186",
editor = "Chih-Jen Lin and Hui Xiong",
booktitle = "KDD' 18 - Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining",
publisher = "Association for Computing Machinery (ACM)",
address = "United States of America",

}

Chen, H, Wang, H, Yin, H, Nguyen, QVH, Wang, W & Li, X 2018, PME: projected metric embedding on heterogeneous networks for link prediction. in C-J Lin & H Xiong (eds), KDD' 18 - Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. Association for Computing Machinery (ACM), New York NY USA, pp. 1177-1186, ACM International Conference on Knowledge Discovery and Data Mining 2018, London, United Kingdom, 19/08/18. https://doi.org/10.1145/3219819.3219986

PME : projected metric embedding on heterogeneous networks for link prediction. / Chen, Hongxu; Wang, Hao; Yin, Hongzhi; Nguyen, Quoc Viet Hung; Wang, Weiqing; Li, Xue.

KDD' 18 - Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. ed. / Chih-Jen Lin; Hui Xiong. New York NY USA : Association for Computing Machinery (ACM), 2018. p. 1177-1186.

Research output: Chapter in Book/Report/Conference proceedingConference PaperResearchpeer-review

TY - GEN

T1 - PME

T2 - projected metric embedding on heterogeneous networks for link prediction

AU - Chen, Hongxu

AU - Wang, Hao

AU - Yin, Hongzhi

AU - Nguyen, Quoc Viet Hung

AU - Wang, Weiqing

AU - Li, Xue

PY - 2018/7/19

Y1 - 2018/7/19

N2 - Heterogenous information network embedding aims to embed heterogenous information networks (HINs) into low dimensional spaces, in which each vertex is represented as a low-dimensional vector, and both global and local network structures in the original space are preserved. However, most of existing heterogenous information network embedding models adopt the dot product to measure the proximity in the low dimensional space, and thus they can only preserve the first-order proximity and are insufficient to capture the global structure. Compared with homogenous information networks, there are multiple types of links (i.e., multiple relations) in HINs, and the link distribution w.r.t relations is highly skewed. To address the above challenging issues, we propose a novel heterogenous information network embedding model PME based on the metric learning to capture both first-order and second-order proximities in a unified way. To alleviate the potential geometrical inflexibility of existing metric learning approaches, we propose to build object and relation embeddings in separate object space and relation spaces rather than in a common space. Afterwards, we learn embeddings by firstly projecting vertices from object space to corresponding relation space and then calculate the proximity between projected vertices. To overcome the heavy skewness of the link distribution w.r.t relations and avoid “over-sampling” or “under-sampling” for each relation, we propose a novel loss-aware adaptive sampling approach for the model optimization. Extensive experiments have been conducted on a large-scale HIN dataset, and the experimental results show superiority of our proposed PME model in terms of prediction accuracy and scalability.

AB - Heterogenous information network embedding aims to embed heterogenous information networks (HINs) into low dimensional spaces, in which each vertex is represented as a low-dimensional vector, and both global and local network structures in the original space are preserved. However, most of existing heterogenous information network embedding models adopt the dot product to measure the proximity in the low dimensional space, and thus they can only preserve the first-order proximity and are insufficient to capture the global structure. Compared with homogenous information networks, there are multiple types of links (i.e., multiple relations) in HINs, and the link distribution w.r.t relations is highly skewed. To address the above challenging issues, we propose a novel heterogenous information network embedding model PME based on the metric learning to capture both first-order and second-order proximities in a unified way. To alleviate the potential geometrical inflexibility of existing metric learning approaches, we propose to build object and relation embeddings in separate object space and relation spaces rather than in a common space. Afterwards, we learn embeddings by firstly projecting vertices from object space to corresponding relation space and then calculate the proximity between projected vertices. To overcome the heavy skewness of the link distribution w.r.t relations and avoid “over-sampling” or “under-sampling” for each relation, we propose a novel loss-aware adaptive sampling approach for the model optimization. Extensive experiments have been conducted on a large-scale HIN dataset, and the experimental results show superiority of our proposed PME model in terms of prediction accuracy and scalability.

KW - Heterogenous Network Embedding

KW - Link Prediction

UR - http://www.scopus.com/inward/record.url?scp=85051459634&partnerID=8YFLogxK

U2 - 10.1145/3219819.3219986

DO - 10.1145/3219819.3219986

M3 - Conference Paper

SN - 9781450355520

SP - 1177

EP - 1186

BT - KDD' 18 - Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining

A2 - Lin, Chih-Jen

A2 - Xiong, Hui

PB - Association for Computing Machinery (ACM)

CY - New York NY USA

ER -

Chen H, Wang H, Yin H, Nguyen QVH, Wang W, Li X. PME: projected metric embedding on heterogeneous networks for link prediction. In Lin C-J, Xiong H, editors, KDD' 18 - Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. New York NY USA: Association for Computing Machinery (ACM). 2018. p. 1177-1186 https://doi.org/10.1145/3219819.3219986