The original paper is in English. Non-English content has been machine-translated and may contain typographical errors or mistranslations. ex. Some numerals are expressed as "XNUMX".
Copyrights notice
The original paper is in English. Non-English content has been machine-translated and may contain typographical errors or mistranslations. Copyrights notice
La prédiction de lien, le problème informatique consistant à déterminer s'il existe un lien entre deux objets, est importante dans l'apprentissage automatique et l'exploration de données. La prédiction de liens basée sur les caractéristiques, dans laquelle les vecteurs de caractéristiques des deux objets sont donnés, présente un intérêt particulier car elle peut également être utilisée pour divers problèmes liés à l'identification. Bien que la machine de factorisation et la machine de factorisation d'ordre supérieur (HOFM) soient largement utilisées pour la prédiction de liens basée sur les caractéristiques, elles utilisent des combinaisons de caractéristiques non seulement entre les deux objets, mais également à partir du même objet. Les combinaisons de fonctionnalités du même objet ne sont pas pertinentes pour les problèmes majeurs de prédiction de liens, tels que la prédiction d'identité, car leur utilisation augmente le coût de calcul et dégrade la précision. Dans cet article, nous présentons de nouveaux modèles qui utilisent des combinaisons de caractéristiques d'ordre supérieur uniquement entre les deux objets. Puisqu'il n'existait aucun algorithme permettant de calculer efficacement des combinaisons de caractéristiques d'ordre supérieur uniquement sur deux objets, nous en obtenons un en tirant parti des résultats rapportés et nouvellement obtenus lors du calcul du noyau ANOVA. Nous présentons un algorithme de descente de coordonnées efficace pour les modèles proposés. Nous améliorons également l'efficacité de celui existant pour le HOFM. De plus, nous étendons les modèles proposés à un réseau neuronal profond. Les résultats expérimentaux ont démontré l'efficacité de nos modèles proposés.
Kyohei ATARASHI
Hokkaido University
Satoshi OYAMA
Hokkaido University,RIKEN AIP
Masahito KURIHARA
Hokkaido University
The copyright of the original papers published on this site belongs to IEICE. Unauthorized use of the original or translated papers is prohibited. See IEICE Provisions on Copyright for details.
Copier
Kyohei ATARASHI, Satoshi OYAMA, Masahito KURIHARA, "Link Prediction Using Higher-Order Feature Combinations across Objects" in IEICE TRANSACTIONS on Information,
vol. E103-D, no. 8, pp. 1833-1842, August 2020, doi: 10.1587/transinf.2019EDP7266.
Abstract: Link prediction, the computational problem of determining whether there is a link between two objects, is important in machine learning and data mining. Feature-based link prediction, in which the feature vectors of the two objects are given, is of particular interest because it can also be used for various identification-related problems. Although the factorization machine and the higher-order factorization machine (HOFM) are widely used for feature-based link prediction, they use feature combinations not only across the two objects but also from the same object. Feature combinations from the same object are irrelevant to major link prediction problems such as predicting identity because using them increases computational cost and degrades accuracy. In this paper, we present novel models that use higher-order feature combinations only across the two objects. Since there were no algorithms for efficiently computing higher-order feature combinations only across two objects, we derive one by leveraging reported and newly obtained results of calculating the ANOVA kernel. We present an efficient coordinate descent algorithm for proposed models. We also improve the effectiveness of the existing one for the HOFM. Furthermore, we extend proposed models to a deep neural network. Experimental results demonstrated the effectiveness of our proposed models.
URL: https://global.ieice.org/en_transactions/information/10.1587/transinf.2019EDP7266/_p
Copier
@ARTICLE{e103-d_8_1833,
author={Kyohei ATARASHI, Satoshi OYAMA, Masahito KURIHARA, },
journal={IEICE TRANSACTIONS on Information},
title={Link Prediction Using Higher-Order Feature Combinations across Objects},
year={2020},
volume={E103-D},
number={8},
pages={1833-1842},
abstract={Link prediction, the computational problem of determining whether there is a link between two objects, is important in machine learning and data mining. Feature-based link prediction, in which the feature vectors of the two objects are given, is of particular interest because it can also be used for various identification-related problems. Although the factorization machine and the higher-order factorization machine (HOFM) are widely used for feature-based link prediction, they use feature combinations not only across the two objects but also from the same object. Feature combinations from the same object are irrelevant to major link prediction problems such as predicting identity because using them increases computational cost and degrades accuracy. In this paper, we present novel models that use higher-order feature combinations only across the two objects. Since there were no algorithms for efficiently computing higher-order feature combinations only across two objects, we derive one by leveraging reported and newly obtained results of calculating the ANOVA kernel. We present an efficient coordinate descent algorithm for proposed models. We also improve the effectiveness of the existing one for the HOFM. Furthermore, we extend proposed models to a deep neural network. Experimental results demonstrated the effectiveness of our proposed models.},
keywords={},
doi={10.1587/transinf.2019EDP7266},
ISSN={1745-1361},
month={August},}
Copier
TY - JOUR
TI - Link Prediction Using Higher-Order Feature Combinations across Objects
T2 - IEICE TRANSACTIONS on Information
SP - 1833
EP - 1842
AU - Kyohei ATARASHI
AU - Satoshi OYAMA
AU - Masahito KURIHARA
PY - 2020
DO - 10.1587/transinf.2019EDP7266
JO - IEICE TRANSACTIONS on Information
SN - 1745-1361
VL - E103-D
IS - 8
JA - IEICE TRANSACTIONS on Information
Y1 - August 2020
AB - Link prediction, the computational problem of determining whether there is a link between two objects, is important in machine learning and data mining. Feature-based link prediction, in which the feature vectors of the two objects are given, is of particular interest because it can also be used for various identification-related problems. Although the factorization machine and the higher-order factorization machine (HOFM) are widely used for feature-based link prediction, they use feature combinations not only across the two objects but also from the same object. Feature combinations from the same object are irrelevant to major link prediction problems such as predicting identity because using them increases computational cost and degrades accuracy. In this paper, we present novel models that use higher-order feature combinations only across the two objects. Since there were no algorithms for efficiently computing higher-order feature combinations only across two objects, we derive one by leveraging reported and newly obtained results of calculating the ANOVA kernel. We present an efficient coordinate descent algorithm for proposed models. We also improve the effectiveness of the existing one for the HOFM. Furthermore, we extend proposed models to a deep neural network. Experimental results demonstrated the effectiveness of our proposed models.
ER -