How does weighted Jaccard stem deduplication catch reworded exam repeats?
A local Web Worker normalizes every question prompt by stripping HTML markup, leading question numbers, punctuation, and 18 common English stop words, then computes a weighted score combining 75% content-word overlap with 25% adjacent word-bigram overlap. Stems meeting or exceeding your similarity threshold merge automatically so cumulative finals never repeat near-identical questions.
Importing three years of midterm archives straight into Canvas or Moodle creates parallel folders full of forgotten clones. Exact text comparisons miss slight edits between semesters, such as changing array of n items to array of N items or rewriting indicates a permanent redirect as indicates that a page was permanently redirected.
Applying the Jaccard similarity coefficient formula across both content-word sets and two-word bigrams captures vocabulary overlap while respecting word order. Exact normalized matches score 100%, whereas reworded stems from earlier terms score in the 75% to 92% range and surface in the live audit log.
Combining QTI archives, Moodle XML, GIFT, and spreadsheet banks
Academic departments rarely keep multi-year assessments in one format. One instructor exports a Canvas QTI .zip, a colleague shares a Moodle .gift file, and a lab coordinator stores quizzes in Excel or Word.
Normalizing cross-LMS files up to 30 MB into one shared schema
Dropping files up to 30 MB (30,000,000 bytes) each into Add QTI .zip, Moodle XML, GIFT, Aiken, CSV, Excel, or Word banks converts every source into one typed question list. Distractors, multi-select weights, numerical tolerances, matching pairs, and item rationales stay intact across formats.
Brace-delimited pools following the Moodle GIFT question syntax reference merge cleanly beside Blackboard QTI archives and Aiken .txt files. To check media assets inside an archive first, preview QTI 1.2 and 2.1 archives in another tab.
Appending pasted Aiken or GIFT snippets with custom term labels
Expanding Or paste an additional text / Aiken / GIFT bank lets you enter a Source bank label like Spring 2026 Quiz 2, paste raw items into Pasted quiz text, and click Add pasted bank. Clicking × on any source chip removes that bank and recalculates Total imported, Unique kept, and Duplicates merged.
Which duplicate resolution mode fits cumulative finals and department pools?
Select Keep first & merge semester tags when building a longitudinal department pool where tracking term history and preserving rationales matters, Keep newest bank version when later semesters contain corrected distractors or updated wording, Keep earliest bank version to lock canonical baseline items, or Keep all (no deduplication) when concatenating distinct chapter files without filtering.
Curating longitudinal pools follows calibrated educational item bank overview practices that track item provenance across terms. Compare all four Duplicate resolution modes in the table.
| Duplicate resolution option | Surviving stem & choices | Category & feedback behavior | Best academic workflow |
|---|---|---|---|
| Keep first & merge semester tags | Keeps first bank’s prompt and choices | Joins term labels with / and backfills generalFeedback |
Longitudinal department pools and finals |
| Keep earliest bank version | Retains first uploaded item untouched | Keeps original category and feedback | Protecting a vetted baseline pool |
| Keep newest bank version | Adopts latest uploaded bank’s item | Uses updated wording, keys, and category | Rolling forward semester errata fixes |
| Keep all (no deduplication) | Retains 100% of imported items |
Preserves each source label as imported | Concatenating distinct chapter files |
Tuning the 55% to 100% similarity slider and match audit log
Different subjects need different deduplication sensitivity. A history stem reworded across semesters shares distinct nouns at 75%, whereas stoichiometry items may share 85% boilerplate phrasing while testing different molar ratios.
When to tighten the slider toward 90% or 100% for parallel math items
Dragging Near-duplicate threshold (#qbm-sim) adjusts the cutoff from 55% to 100% in 5% steps. The default 75% catches paraphrased conceptual questions across midterms without merging unrelated prompts.
When merging physics, accounting, or algebra banks that reuse sentence templates with changed numbers (12 V versus 24 V), raise the slider to 90% or 95%, or 100% for exact canonical matches only.
Inspecting kept versus merged stems in the live match log
Every merged pair appears inside Detected duplicate & near-duplicate stems showing the kept source, dropped source, and percentage score (100% match or 80% match). Authors drafting new questions from scratch can also use the plain-text to Canvas QTI packager before pooling archives here.
Tip: Click Load semester sample to watch
Fall 2025 MidtermandSpring 2026 Finalmerge two reworded binary-search and HTTP-301 stems while preserving unique TCP items.
Filtering question types and exporting unified LMS master packages
Once duplicates resolve, Master bank title names your assessment header and downloaded filename. You can filter by item format and export straight to your LMS or grading binder.
Isolating multiple-choice, true/false, or short-answer subsets
Switching Question type filter from All question types to Multiple choice only strips open-response, true/false, and multi-select items for Scantron or bubble-sheet finals, with dedicated filters for Multiple answers, True / False, and Short answer pools.
When printing paper tests from a marked Word or Markdown draft instead, generate unmarked student copies without leaked answers alongside a separate grading key.
Packaging QTI 1.2, QTI 2.1, Moodle XML, GIFT, and spreadsheets
Clicking Download merged QTI 1.2 (.zip) builds imsmanifest.xml and item XML per the IMS QTI 1.2 assessment XML binding for Canvas Classic and Blackboard, while QTI 2.1 (.zip), Moodle XML, and Moodle GIFT cover Brightspace and Moodle. Choose Excel (.xlsx), CSV table, or Master PDF for 12-column audits and printed keys.
Consolidating multi-instructor course banks without cloud uploads
Multi-section introductory courses in biology, psychology, nursing, and computer science often rotate across three or four lecturers each academic year. Pooling everyone’s quiz files locally solves three recurring coordination headaches.
- Ordering source banks intentionally: Upload your most thoroughly proofread bank first when using Keep first & merge semester tags, or upload your newest errata-corrected semester last when using Keep newest bank version.
- Backfilling missing explanations: Merging an older unannotated QTI export with a newer annotated GIFT file under Keep first & merge semester tags automatically attaches the newer
generalFeedbackrationale onto the kept item. - Protecting unreleased department finals: Because ZIP decompression, XML parsing, and Jaccard scoring run inside local browser memory, confidential accreditation and final exam pools never touch third-party servers.
After assembling your deduplicated cumulative pool, you can also build bell-ringer review slide decks from selected practice banks to run projector warm-up sessions during finals week without student logins.