Taxonomy plays a key role in e-commerce, categorising items and facilitating both search and inventory management. Concept subsumption prediction is critical for taxonomy curation, and has been the subject of several studies, but they do not fully utilise the categorical information available in e-commerce settings. In this paper, we study the characteristics of e-commerce taxonomies, and propose a new subsumption prediction method based on the pre-trained language model BERT that is well adapted to the e-commerce setting. The proposed model utilises textual and structural semantics in a taxonomy, as well as the rich and noisy instance (item) information. We show through extensive evaluation on two large-scale e-commerce taxonomies from eBay and AliOpenKG, that our method offers substantial improvement over strong baselines.
Description
Subsumption Prediction for E-Commerce Taxonomies | SpringerLink
%0 Conference Paper
%1 10.1007/978-3-031-33455-9_15
%A Shi, Jingchuan
%A Chen, Jiaoyan
%A Dong, Hang
%A Khan, Ishita
%A Liang, Lizzie
%A Zhou, Qunzhi
%A Wu, Zhe
%A Horrocks, Ian
%B The Semantic Web
%C Cham
%D 2023
%E Pesquita, Catia
%E Jimenez-Ruiz, Ernesto
%E McCusker, Jamie
%E Faria, Daniel
%E Dragoni, Mauro
%E Dimou, Anastasia
%E Troncy, Raphael
%E Hertling, Sven
%I Springer Nature Switzerland
%K bert e-commerce knowledge-engineering lm myown ontology_learning subsumption subsumption_relations taxonomies
%P 244--261
%T Subsumption Prediction for E-Commerce Taxonomies
%X Taxonomy plays a key role in e-commerce, categorising items and facilitating both search and inventory management. Concept subsumption prediction is critical for taxonomy curation, and has been the subject of several studies, but they do not fully utilise the categorical information available in e-commerce settings. In this paper, we study the characteristics of e-commerce taxonomies, and propose a new subsumption prediction method based on the pre-trained language model BERT that is well adapted to the e-commerce setting. The proposed model utilises textual and structural semantics in a taxonomy, as well as the rich and noisy instance (item) information. We show through extensive evaluation on two large-scale e-commerce taxonomies from eBay and AliOpenKG, that our method offers substantial improvement over strong baselines.
%@ 978-3-031-33455-9
@inproceedings{10.1007/978-3-031-33455-9_15,
abstract = {Taxonomy plays a key role in e-commerce, categorising items and facilitating both search and inventory management. Concept subsumption prediction is critical for taxonomy curation, and has been the subject of several studies, but they do not fully utilise the categorical information available in e-commerce settings. In this paper, we study the characteristics of e-commerce taxonomies, and propose a new subsumption prediction method based on the pre-trained language model BERT that is well adapted to the e-commerce setting. The proposed model utilises textual and structural semantics in a taxonomy, as well as the rich and noisy instance (item) information. We show through extensive evaluation on two large-scale e-commerce taxonomies from eBay and AliOpenKG, that our method offers substantial improvement over strong baselines.},
added-at = {2023-05-22T10:07:04.000+0200},
address = {Cham},
author = {Shi, Jingchuan and Chen, Jiaoyan and Dong, Hang and Khan, Ishita and Liang, Lizzie and Zhou, Qunzhi and Wu, Zhe and Horrocks, Ian},
biburl = {https://www.bibsonomy.org/bibtex/28a652081e83067713c540cae7da995de/hangdong},
booktitle = {The Semantic Web},
description = {Subsumption Prediction for E-Commerce Taxonomies | SpringerLink},
editor = {Pesquita, Catia and Jimenez-Ruiz, Ernesto and McCusker, Jamie and Faria, Daniel and Dragoni, Mauro and Dimou, Anastasia and Troncy, Raphael and Hertling, Sven},
interhash = {dee604565f4398f0259ba8ee6c856a4b},
intrahash = {8a652081e83067713c540cae7da995de},
isbn = {978-3-031-33455-9},
keywords = {bert e-commerce knowledge-engineering lm myown ontology_learning subsumption subsumption_relations taxonomies},
pages = {244--261},
publisher = {Springer Nature Switzerland},
timestamp = {2023-05-22T10:07:04.000+0200},
title = {Subsumption Prediction for E-Commerce Taxonomies},
year = 2023
}