TESTING: Clustering

We have implemented a new clustering option, which groups together top ranked results from the same domain. The clustered results are shown indented, as you can see in the example below:

  • on the left with clustering off, using: /search?q=bbc&clufmt=0
  • on the left with clustering on, using: /search?q=bbc&clufmt=4

To use clustering use &clufmt=4 in the search URL. Currently you cannot use &clufmt=4 with &si (maximum number of results to return per site, per complete query result set).

For this feature to work, you might need to refresh to see the indentation; on:

  • Windows/Linux:
    • Chrome/Edge/Firefox: Press Ctrl + F5 or Shift + F5
  • Mac:
    • Chrome/Firefox: Press Cmd + Shift + R
    • Safari: Press Option + Cmd + E to empty cache, then Cmd + R to reload

We expect to move this setting to be the default quite soon, so feedback would be appreciated. We haven’t found any cases where it is undesirable, so if you come across one please let us know here, or using the “Submit Feedback”, or by email.

Thank you.

6 Likes

This can also be set as a cookie:

In Firefox, you can manually add a cookie by

  • Using your browser’s Developer Tools while visiting mojeek.com
  • Visiting the Storage tab in Developer Tools.
  • Clicking on the “+” button in the top right (not shown above).

Then manually click each field to modify the newly created cookie’s properties:

  • Name
  • Value
  • Expires

Similar instructions for Chrome can be found elsewhere under the Discourse Testing category.

2 Likes

nice - as the Inuit said in ‘Never Cry Wolf’, “Good idea”

1 Like

+1 from me! I’ve enabled it and this is soooo much better.

Question: what are the implications for the Results per domain setting? Is it obsoleted, or will the default change from 1 to ~4? The value 4 makes me think it’s now 4 by default.

Looks a great addition, awesome

Going by substack - Mojeek Search it works at the exact domain level?

Yes, subdomains are clustered separately. For example: github - Mojeek Search

1 Like

Thank you @all for your feedback. This new clustering is now live.

It is turned it on by default, with what is effectively ‘clufmt=5’. This gives you one result plus up to 4 indented.

For those that like to customise, a value of between 0 and 5 is valid:

  • 0 = disable new clustering and revert to the original using ‘si=n’; where n is the number of results from a site’.
  • 1 = show a maximum of one result per host per 10 results.
  • 2 to 5 = show a maximum of 2 to 5 results per host per 10 results.

This clustering also works for Focus where the default setting is ‘clufmt=2’. You can also customise this, as above, if you wish

We hope you like it. As always, feedback is encouraged.

3 Likes

This update has made a dramatic improvement to my SERPs on Mojeek. I now feel like I’m getting so much more out of the pages in Mojeek’s index because they’re presented in a much better way.

(Still running into pages I want that aren’t in the index yet, but I’ll keep submitting feedback!)

2 Likes

Agree. I’m finding it useful for example on imdb searches where there’s a closely matching film name that ranks, the clustering gives a closer page that I intended to visit. Means the target page although not ideally ranking, is a click away.

1 Like

There is some user feedback here asking for a clufmt preference at mojeek.com/preferences

Also, I noticed the si preference was removed when clustering was launched.

I have some additional feedback after reading the recent announcement.

Colin wrote:

But, I thought I found a slight bug which is some site: searches will show search results which are all indented.

For example, The World Factbook (cia.gov) will indent all results:

But, other hosts won’t do the same. For example, Wikipedia:

Edit:

I think I see the issue.

I have The World Factbook saved as a Mojeek Focus.

{"Factbook":"i=www.cia.gov"}

That seems to be required to trigger the bug.

For example, here is the same search using a site: operator. And, no indentation is shown:

@mike This is a quirk of having one site only in a Focus definition.

1 Like