Dries Buytaert

Mollom

Bright green graphic celebrating Mollom reaching 10,000 active users and blocking 100 million spam messages.
© Jamey Boje.
A graph shows spam attempts blocked (orange) and non-spam accepted (green) over time, with most activity in blocked spam.
A newspaper article about Mollom, a startup that protects the internet from spam and unwanted content.
A moderation interface for filtering user content based on spam, profanity, and reputation scores, with adjustable settings.
Two people sign documents at a desk while another person smiles nearby in an office setting.
© Thomas Erickson.
Screenshot of the redesigned Mollom website, showcasing its spam filtering and content quality evaluation features with call-to-action buttons.
The new Mollom.com website design.
Green and teal Mollom homepage with the slogan "Spot high-quality content, Stop spam, Save time moderating your site" and download steps.
The old Mollom.com design that we used from 2007 to early 2012.
Settings page for configuring comment protection using text analysis or CAPTCHA to filter spam and choose automatic or manual moderation.
A configuration page for comment form protection with options for spam detection, profanity filtering, and text analysis accuracy settings.
Mollom notice saying the service ended on 2 April 2018 after blocking more than 13.5 billion spam comments since 2008.
Mollom's end-of-life announcement that replaced the old https://mollom.com.
A world map showing spam activity, with darker colors representing higher spam levels in different regions.
A spam message with placeholder text and variable word choices, likely generated by an automated script.
Code snippet highlighting "Drupal speak" and a "Spam link", demonstrating content reuse in spam techniques.
Spam comment from an anonymous author with empty fields and a body that is only a link to facebook.com with rel="follow".
Highlighted HTML code snippet showing an embedded Flash file link, a technique commonly used in spam.
A spam message with random text, fake email, and suspicious links, likely generated for automated spam techniques.
Email screenshot requesting investment in a fiberglass factory, with capitalized text emphasizing urgency.
Highlighted spam technique using "google_ad_section_start" in a message promoting house projects.
Email spam message promoting a discount on jeans with an embedded image link, likely for phishing or scams.
Spam submission in which every field, including author and email, is null and the body only says "thanks for all."
A screenshot shows OpenID links, illustrating potential spam techniques related to user authentication and online identity.
A spam message with a fake author and a suspicious link, likely used for phishing or malicious purposes.
Mollom stats: 49,366 sites protected, 99.96% efficiency, 500,373,282 spam messages caught, 619,308 yesterday, 84% of messages spam.
Screenshot of the scorecard section on Molom.com.
Dries presents at a lectern beside a slide reading "Mollom: Website protection and moderation services, TechCrunch Belgium Meetup, Ghent".
© Bart Claeys.
Weekly graph of spam requests per second, steady near 7, jumping to about 21 on the 14th and falling back late on the 15th.
The Y-axis shows the number of requests per second. The X-axis shows the day of the month. The graph is for one Mollom server only.
Black T-shirt with a green "M" logo on the front and "Mollom" text on the upper back.
Pie chart showing Mollom usage: 44% personal websites, 27% non-profits, 22% company websites, and 7% customer websites.
Based on a sample of roughly 2,000 Drupal sites that use Mollom.
The same spaminess graph split into three zones, Spam, Unsure and Ham, with the overlapping curve tails falling in the Unsure zone.
The plot illustrates that having a classifier with three states avoids false positives and false negatives. The plot is generated from the actual data in Mollom's database.
Graph of spaminess scores with one cutoff line: an orange spam curve on the left and a green ham curve on the right overlap at the line.
The plot illustrates that having a binary classifier with only two states (ham and spam) is bound to make mistakes. The plot is generated from the actual data in Mollom's database.
A group of seven people stands together outside, smiling at the camera during a planning event.
Six people work on laptops around a long table covered with drinks and snacks, in a room with tall windows and plants.
A photo of the Mollom team at an offsite in 2011: it includes Daniel Kudwien, Benjamin Schrauwen, Cedric De Vleeschauwer, Thomas Meire, Johan Vos and Vicky Van Roeyen. Missing in the picture is Dries.
Dries smiles at a table stacked with paper and coffee flasks while Mary Jefts signs a document in the foreground.
Selling Mollom to Acquia involved signing a lot of documents. A photo of me signing the acquisition paperwork with Mary Jefts, Acquia's CFO at the time. It took three hours to sign all the paperwork.
Stacks of printed documents and a folder are neatly arranged on a table in an office setting.
Three people sit at a table sorting and signing documents, with stacks of paper around them.
Two people sit at a table reviewing and signing documents, possibly related to the closing of Mollom.
Two people sit at a table signing documents, with stacks of papers, coffee mugs, and snacks nearby.