Marginalia Search - search rank algorithm

Marginalia is a very interesting index search engine for several reasons, one being the ability to change the search rank algorithm which i think is a great idea

1 Like

Yes, Marginalia is very interesting project and a great contribution to the community discussing, developing and innovating in search. We hope it survives: Uncertain Future for Marginalia Search | Hacker News

Giving users more control over search ranking is something we would like to do, and have all sorts of ideas that we have thought through, and an outline plan to tackle for when we have the bandwidth. Obviously there is a lot we could potentially do here given our full independence and not being a proxy for large search indexes.

Incidentally the Marginalia options are not so much a change of ranking algorithm, as far as I understand it, but more a case of searching across the web with a boost to the type of site. But perhaps someone else may know and can comment on that. It’s certainly not the same as using Mojeek Focus but another approach to givig users control.

right, that’s why i mentioned it here

the info it provides about the pages content in the search results is also interesting (JS, tracking, cookies, affiliates, media)

you can also use these operators manually to filter results, either as is, or preceded with ‘-’

special:media
special:scripts
special:affiliate
special:tracking
special:cookies

mechanical keyboard -special:affiliate -special:tracking

i talked to the developer and i think it will

2 Likes

Interesting! I didn’t know about those operators, do you know if there is a complete list of operators it supports ?

or for that matter is there one for mojeek too ?

apparently the documentation for Marginalia is rather sparse, so no, i don’t - i see it supports the site: operator though

1 Like

Mojeek operators here: Search Operators

You can also use some of the API request parameters as documented here: API Documentation - Search Request Parameters

For example this query gets 30 results (&t=30), starting at 21 (&s=21), displaying 2 results per site (&si=2): https://www.mojeek.com/search?q=mojeek&t=30&s=21&si=2

2 Likes

Specifically, Marginalia’s centrality algorithm is Custom Pagerank which allows the index to be biased towards sites that appear “close” to a selection of preferred sites. To bias the index towards non-commercial sites, it centered the index around explicitly anti-commercial sites.

I’m personally quite interested in centrality algos that try to behave differently from the OG PageRank algo. Alexandria, for instance, uses harmonic centrality on the Common Crawl corpus.