Better ranking/sorting algorithm

[quantizor]

Describe the feature

First of all, great work creating a performant npm browser alternative!

One issue I've noticed is a lot of good packages tend to get buried by npm due to nuances in their relevance ranking algorithm. For example, when searching for "markdown", the markdown-to-jsx library doesn't show up for multiple pages despite having a ton of downloads and many dependent libraries.

Curious what improvement can be made here and what other relevancy and/or sorting algorithms could be experimented with to qualitatively improve the average browsing session / surface high quality but less-hyped packages.

Additional information

• Would you be willing to help implement this feature?

Final checks

• Read the contribution guide.
• Check existing issues.

[Wybxc]

The heuristic ranking algorithm used by lib.rs can serve as a reference. lib.rs is a catalog of programs and libraries of Rust. As an alternative to crates.io, I really like this website. I hope the npm ecosystem can also have a site with similar positioning, and npmx is a strong contender.

The following is the ranking algorithm used by lib.rs, from their website:

Crates are sorted by their overall quality score, which is a weighed combination of:

• The crate's popularity measured by number of downloads, but only as a small fraction of the score, and with some filtering of noise and obviously spammed crates,
• Number of direct and indirect reverse dependencies. The numbers are corrected for biases that affect application crates and dependencies of dependencies.
• The crate's usage trend — is it gaining or losing users.
• Availability of the crate's documentation, examples, and length and quality of the README.
• Stability estimated from release history, number of breaking versions, patch versions, and use of nightly features.
• Presence of tests, CI, code comments.
• Accuracy and completeness of the crate's metadata.
• Number of authors and contributors.
• Weight of the crate's unique dependencies (taking into account that some crates are very common and shared between projects).
• Whether the crate is actively maintained or at least stable and done, based on release frequency, age, maintenance status, use of deprecated/outdated dependencies, non-0.x releases, etc.
• Reputation of maintainers of the crate, calculated page-rank-like based on co-owning relationships.
• Plus special-case factors and blocklists for undesirable or junk crates.

The score is combined with relevance of crate's keywords to a given category or a search query.