Project Details
Home/Project Details
PlagCheck NG
Project description
Plagiarism detection built for Nigerian academic writing
The Challenge
General-purpose plagiarism checkers are trained and tuned on English academic writing from elsewhere, and they behave badly on Nigerian theses: names, institutions, local terminology and the conventions of the form all read as anomalies, and legitimate quotation gets flagged while paraphrased copying gets through.
A plagiarism report is an accusation, so a false positive is not a nuisance — it is a person's problem. The system has to be conservative and explainable, and every match has to be something a human can look at and agree with.
The Solution
Build the pipeline around quote detection and citation structure first, so properly attributed material is excluded before similarity is measured, then run the comparison over what remains. Every flagged span carries its source and the matched text, because a report nobody can audit is a report nobody should act on.
Skills and deliverables
Features delivered
The hard part
Telling legitimate academic reuse from copying. Referencing previous work is the discipline, not a failure of it, so the model has to distinguish a cited paraphrase from a lifted one — which is a structure problem more than a similarity problem.
The Impact
A working detection service with quote and citation handling in front of the similarity engine. Deployed nowhere deliberately: it needs a document corpus, and a public instance would mean uploading students' work to it.
