Query Resolution of Literature Knowledge Graphs using Hybrid Document Embeddings.

Coenen, Frans ORCID: 0000-0003-1026-6649, Muhammad, Iqra, Gamble, Carrol ORCID: 0000-0002-3021-1955, Kearney, Anna ORCID: 0000-0003-1404-3370 and Williams, Paula (2022) Query Resolution of Literature Knowledge Graphs using Hybrid Document Embeddings. .

PDF sgaiConference_2022a.pdf - Other Download (362kB) | Preview

Abstract

Literature Knowledge Graphs play a critical role in helping domain experts carry out query resolution for finding relevant articles in published literature. Such knowledge graphs are usually in the form of Curated Document Databases (CDDs). Domain Experts and researchers typically query such literature knowledge graphs using some form of query-resolution mechanism. Machine learning techniques can be used to automate query-resolution. This paper presents a document query-resolution mechanism, given a query and set of documents in a knowledge graph, based on a hybrid word embedding that combines knowledge graph embeddings with “traditional” embeddings. A query-document data set extracted from a clinical trials CDD (the ORRCA CDD) was used. Three “traditional” word embeddings were considered: CBOW, BERT and SciBERT. The evaluation demonstrated that hybrid embeddings produced better results than when the embedding models were used in isolation. A best Mean Average Precision of 0.486 was obtained when using a CBOW and random walk knowledge graph hybrid embedding.

Item Type:	Conference Item (Unspecified)
Uncontrolled Keywords:	Query resolution, Word embedding, Document ranking
Divisions:	Faculty of Health & Life Sciences Faculty of Health & Life Sciences > Inst. Population Health Faculty of Science & Engineering > School of Electrical Engineering, Electronics and Computer Science
Depositing User:	Symplectic Admin
Date Deposited:	14 Mar 2023 10:39
Last Modified:	01 Mar 2026 06:53
DOI:	10.1007/978-3-031-21441-7_7
Related Websites:	Author Publisher
URI:	https://livrepository.liverpool.ac.uk/id/eprint/3169038
Disclaimer:	The University of Liverpool is not responsible for content contained on other websites from links within repository metadata. Please contact us if you notice anything that appears incorrect or inappropriate.