Wikimedia Discovery/Meetings/Portal ideas 2015-12-04

The portal team met to generate ideas for how to improve the very low click-through rate on the portal (

We got some interesting findings from initial Event Logging data. Numbers are (disturbingly) stable over 2 weeks. 

Questions from Moiz:

  1. How can we improve the 60% bounce rate? (60% of people who visit the portal leave without doing anything)
  2. What else can we instrument to find out more about these users?

Of the 26k events over the last 2 days, how many are from automata (crawlers/bots)? 8. Not 8k, but just 8. 

We are capturing actual users, because this is javascript, which automata tend to bypass.

AdamB's theory: Lots of people have wikipedia as their home page. 

Can we figure that out? Yes, we have referrer information that may help. 

We have Clickthrough traffic from google

(Scribe missed several minutes of discussion here)

Focus on one thing at a time. 

Our tests are focused on the 30% of users. Can we take on other low-hanging fruit?

Oliver's theory: Having all of those links leads to confusion/fatigue

Could we detect what languages the user is likely to want, and put those around the globe?

Reduces clutter, and provides most likely relevant links

Do we need to look at exactly what people do, such as scrolling?

Moiz's intuition: Most people don't scroll, and those who do probably don't click

Bottom section doesn't really do much

Dan: We don't know if people aren't clicking on the lower links because they didn't scroll to see it, or because they didn't like what they saw

If we did find that people weren't scrolling, would we want to pull all the links up?

Removing the links could be contentious, so more data would help us

Existing data already shows people aren't clicking on them, but more/other data might help

Even if you scroll to see the lower links, it's very difficult to find the one you want

Showing all the languages might be helping our brand by showing that we support all those languages. Could some other presentation achieve that?

Maybe we could bury them inside "more languages"?

What else could we do on that page, to engage the 60%? Trending articles? 

Trending would require language detection, so not easy. 

Showcase media? e.g. Commons picture of the day. 

What about community's reaction? How do we talk with them and engage them?

We should bring in a CL person for help

Moiz wants to share this data with people who maintain this page.

We could ask people what they think the page should do, but we would need to keep it grounded. "How would you improve search on the portal?"

Make it clear that this is idea generation, not a voting. CL person could help us with that. 

Would maintainers be disappointed? We think mxn realizes the current portal isn't that effective. 

Dan will reach out to mxn directly. 

Hackathon is April. Could we leverage January Dev Summit? Dev Summit is a very narrow slice (but vocal). 

Could we allow users to customize this page? 

We should get data about anonymous vs. logged-in users

Publishing early would help show why we did the gerrit migration. 

Moiz will reach out to CL by next week.