United Kingdom·Public sector·2019 – ongoingCabinet Office
GOV.UK Related Links recommendation engine
GOV.UK expanded related links across nearly all its content and reported about 5% click-through
Problem
Editors manually maintained related links on only about 2,000 GOV.UK pages. Most content lacked suggestions for further reading.
Approach
The Government Digital Service built Related Links internally and launched it in May 2019. A node2vec model learned connections between pages from browsing patterns and hyperlinks. It refreshed recommendations every 3 weeks, while publishers could override links or exclude sensitive content.
Results
Takeaway
Automated recommendations expanded coverage while publishers retained control over unsuitable links. The reported click-through rate measured use, without establishing whether visitors completed their tasks.
Categories
- Size
- Large enterprise
- Industry
- Public sector
- Business function
- Customer service
- Innovation type
- Process innovation
- Adoption
- Built in-house
- Evidence
- Company's own claim
Sources
- GOV.UK Data Labs (Cabinet Office): Related LinksCabinet Office·29 Feb 2024Company publication
Source summaries
The government documented a production recommendation system that replaced much manual link selection. It reported an average click-through rate of about 5% and described publisher overrides.
Full details are in the original sources.