diff --git a/.gitignore b/.gitignore
index 78d7ad44d..77efe00cf 100644
--- a/.gitignore
+++ b/.gitignore
@@ -52,3 +52,5 @@ config/environments/*.local.yml
 !/app/assets/builds/.keep
 
 /test-data
+
+/dataset
diff --git a/lib/tasks/dataset.rake b/lib/tasks/dataset.rake
new file mode 100644
index 000000000..85f8ddc28
--- /dev/null
+++ b/lib/tasks/dataset.rake
@@ -0,0 +1,103 @@
+# frozen_string_literal: true
+
+desc 'Create a dataset of articles'
+task dataset: :environment do
+  # Text article works that have a single file and that file is a PDF.
+  # Omits the following collections:
+  # * Non-governmental Organizations (NGOs) Collection (2)
+  # * Acquisitions Serials (3)
+  # * Hopkins Marine Station Collection (5)
+  # * Stanford Rare Books and Early Manuscripts (6)
+  # * Dept. of Special Collections, Manuscripts Collections - Supplemental Materials (7)
+  # * Archive of Recorded Sound/Music Library - Supplemental Materials (8)
+  # * Water in the West Reports and Working Papers (40)
+  # * Stanford Center for Ocean Solutions Reports and Working Papers (55)
+  # * Undergraduate Theses, Department of Physics (70)
+  # * Oakland Police Department Race Relations (81)
+  # * John A. Blume Earthquake Engineering Center Technical Report Series (82)
+  # * Undergraduate Honors Theses, Graduate School of Education (102)
+  # * Boothe Prize Winners, Stanford University (116)
+  # * Stanford Iran 2040 Project (125)
+  # * Electronic Acquisitions (143)
+  # * The Korea Program Prize for Writing in Korean Studies (145)
+  # * Free EEMs (144)
+  # * Rigler and Deutsch Record Index project at Stanford, revisited (155)
+  # *	Stanford Research Data (168)
+  # * Publications and flyers by the Red Guard and mass organizations in Guangdong and other cities (218)
+  # * Xu Liangying papers (293)
+
+  sql = <<~SQL.squish
+    SELECT pdf_blobs.id
+    FROM
+    (
+    SELECT works.id, count(*) AS blobs
+    FROM works
+    INNER JOIN work_versions ON works.head_id = work_versions.id
+    INNER JOIN attached_files ON attached_files.work_version_id=work_versions.id
+    INNER JOIN active_storage_attachments ON active_storage_attachments.record_id=attached_files.id
+    INNER JOIN active_storage_blobs ON active_storage_attachments.blob_id=active_storage_blobs.id
+    WHERE active_storage_blobs.content_type = 'application/pdf'
+    AND works.collection_id NOT IN (2,3,5,6,7,8,40,55,70,81,82,102,116,125,143,144,145,155,168,218,293)
+    AND work_versions.work_type='text'
+    AND work_versions.subtype='{Article}'
+    GROUP BY works.id
+    HAVING count(*)=1
+    ) AS pdf_blobs
+    INNER JOIN
+    (
+    SELECT works.id, count(*) AS blobs
+    FROM works
+    INNER JOIN work_versions ON works.head_id = work_versions.id
+    INNER JOIN attached_files ON attached_files.work_version_id=work_versions.id
+    INNER JOIN active_storage_attachments ON active_storage_attachments.record_id=attached_files.id
+    INNER JOIN active_storage_blobs ON active_storage_attachments.blob_id=active_storage_blobs.id
+    GROUP BY works.id
+    HAVING count(*)=1
+    ) AS all_blobs
+    ON pdf_blobs.id=all_blobs.id;
+  SQL
+
+  work_ids = ActiveRecord::Base.connection.execute(sql).values.flatten
+
+  FileUtils.rm_rf('dataset')
+  FileUtils.mkdir_p('dataset')
+
+  rows = nil
+  File.open('dataset/metadata.jsonl', 'w') do |metadata_file|
+    rows = work_ids.map do |work_id|
+      puts work_id
+      work = Work.find(work_id)
+      work_version = work.head
+      blob = work_version.attached_files.first.file.blob
+      pdf_filename = "#{work_id}.pdf"
+      blob.open { |file| FileUtils.cp(file.path, "dataset/#{pdf_filename}") }
+
+      metadata = {
+        work_id:,
+        filename: pdf_filename,
+        title: work_version.title,
+        abstract: work_version.abstract,
+        published_edtf: work_version.published_edtf,
+        authors: work_version.authors.map { |contributor| map_contributor(contributor) },
+        contributors: work_version.contributors.map { |contributor| map_contributor(contributor) },
+        keywords: work_version.keywords.map(&:label)
+      }
+      metadata_file.write("#{metadata.to_json}\n")
+      [work_id, work.druid, pdf_filename, work_version.title, work.collection.head.name]
+    end
+  end
+
+  CSV.open('dataset/dataset.csv', 'w') do |csv|
+    csv << ['Work ID', 'Druid', 'PDF Filename', 'Title', 'Collection']
+    rows.shuffle.each { |row| csv << row }
+  end
+end
+
+def map_contributor(contributor)
+  {
+    first_name: contributor.first_name,
+    last_name: contributor.last_name,
+    full_name: contributor.full_name,
+    orcid: contributor.orcid
+  }
+end