Using ontology-based semantic similarity to facilitate the article screening process for systematic reviews

Area of Science:

• Biomedical Informatics
• Information Science

Background:

• Systematic Reviews (SRs) are crucial for evidence-based practice (EBP) but face challenges in efficient article screening.
• Previous methods relied on lexical relationships, which have limitations in capturing full article relevance.
• Enhancing article relationships with semantic knowledge from Unified Medical Language System (UMLS) offers a novel solution.

Purpose of the Study:

• To develop and evaluate an ontology-based semantic approach for improving article identification in Systematic Reviews (SRs).
• To leverage background semantic knowledge from UMLS concepts and ontologies to enrich article relationships.

Main Methods:

• A pipelined semantic concepts representation process was developed using UMLS, SNOMED-CT, and MeSH ontologies.
• Article relationships were established and visualized as a semantic article network.
• Active learning was incorporated to simulate interactive article recommendation, evaluated on 15 SRs using Work Saved over Sampling at 95% recall (WSS95).

Main Results:

• The ontology-based semantic approach demonstrated superior WSS95 performance compared to lexical and corpus-based methods in most SRs.
• Achieved an average WSS95 of 43.81% and a total WSS95 of 657.18%, indicating significant efficiency gains.
• The semantic network approach effectively assisted in the article screening process.

Conclusions:

• Ontology-based semantics effectively facilitate the identification of relevant articles for SRs.
• UMLS-derived concepts and relationships enhance article semantic connections.
• The proposed approach shows promising generalizability across biomedical SR topics.