Congratulations are extended to Dr. Poppy Nicolette Riddle on successfully defending her doctoral thesis on Thursday, April 9th, 2026!
Dr. Riddle’s work “An analysis of Crossref and OpenAlex as external knowledge sources for Retrieval-Augmented Generation” looked at mistakes and “noise” (messy or incorrect text) in research article titles and abstracts in major scholarly databases, and whether those issues affect AI systems used to retrieve and generate information. It found that errors are common—about 42% of titles and 63% of abstracts in a sample from Crossref contained some kind of problem. Data from OpenAlex had slightly less missing information, but errors still remained.
The study focused on two specific issues: formatting/encoding problems (like garbled characters) and errors involving multilingual text. It found that multilingual content could actually improve AI retrieval when the system used models designed for multiple languages. But when English-only AI models processed multilingual abstracts, the quality of generated answers dropped. By contrast, formatting and encoding problems did not significantly affect performance.
Overall, the findings suggest that while some metadata errors may not matter much, language-related issues can strongly influence how well AI systems work, especially depending on the models used.
Read Dr. Riddle’s dissertation here.
Dr. Riddle was supervised by Dr. Philippe Mongeon and is the inaugural graduate of the Department of Information Science’s PhD in Information program at Dalhousie! Dr. Riddle will continue to collaborate with the QSSLab and is moving into a position with Crossref!